Merge upstream 0.2.2 and resolve reroll conflicts (#408)

2026-10-10 00:46:46 +08:00
commit be095d8638
@@ -34,8 +34,12 @@ effects), [DATA.md](DATA.md) (generated data), [ASSETS.md](ASSETS.md) (art and a
  against the spec (`server/match/fields.js` `validateClientResult`), runs the battles no connected human owns (bots, a
  dropped player, an authority that missed its deadline) and can re-simulate results (`SP_VERIFY`). `SP_COMBAT=server`
  is the older mode in which the server runs every battle and streams snapshots. DESIGN §14.
- **Determinism** is what makes this work: fixed 1/30 s ticks, a seeded PRNG (`server/sim/rng.js`), no wall clock and
-  no `Math.random` in the sim, so the same spec and data give the same battle on every machine, bit for bit. The golden
+- **Determinism** is what makes this work: fixed 1/30 s ticks, a seeded PRNG (`server/sim/rng.js`), no wall clock,
+  no `Math.random` and only correctly rounded arithmetic in the sim, so the same spec and data give the same battle on
+  every machine, bit for bit. ECMA-262 leaves `Math.hypot`, `sin`, `cos`, `atan2`, `pow` (and `**`) implementation-
+  approximated and the engines differ in their last bits (V8, SpiderMonkey and JavaScriptCore each compute `hypot`
+  differently; Chrome's and Node's `sin` differ), so the sim takes them from `server/sim/detmath.js`, built only from
+  + − × ÷ and `Math.sqrt`; ESLint and `test/sim/detmath.test.js` refuse the Math ones in `server/sim/`. The golden
  results (§5) rely on it.

 ## 2. The WebSocket protocol
@@ -89,7 +93,7 @@ so old imports keep working: `public/js/ui/gameLogic.js` (`public/js/ui/gameLogi
 | `server/sim/content/enemies/` | the enemy kits by special type (`invisible.js`, `times.js`, `element.js`, `dot.js`, `reflection.js`, `fly.js`, `special.js`) and `leaders.js`; `server/sim/content/enemies.js` dispatches them, `server/sim/content/bosses.js` scripts the leaders |
 | `server/sim/content/garrisons/`, `items/`, `bands/` | 特质, equipment and strategies: `battle.js` is the battle side, `meta.js` the prep side (`registerMeta`, META §2) |
 | `server/sim/content/bonds/` | the 23 bonds: `core.js` the 8 core bonds (both sides), `server/sim/content/bonds/addon/` the 15 add-on bonds (`battle.js`, `meta.js`) |
-| `server/sim/content/` (the rest) | `tokens.js` (summons), `devices.js` (terrain and stage devices), `generic.js` (the kit built from a skill's data when a chess has none), `choices.js` (机变 cards in battle) |
+| `server/sim/content/` (the rest) | `tokens.js` (summons), `devices.js` (terrain and stage devices), `generic.js` (the kit built from a skill's data when a chess has none), `choices.js` (机变 cards in battle), `traitMods.js` (the module trait line the engine applies to every operator: 「攻击范围内存在N名及以上敌人时攻击速度+X」) |
 | `shared/` | imported by the server and the browser: `protocol.js`, `constants.js`, `i18n.js`, `i18nData.js` and `i18nPacks.js` (languages), `packs.js` (the content-pack format), `standIn.js` (补位), `diy.js` (自选编队), `highGround.js`, `loadoutRecord.js` |

 ### Client (`public/`)
@@ -99,7 +103,7 @@ so old imports keep working: `public/js/ui/gameLogic.js` (`public/js/ui/gameLogi
 | `public/index.html`, `public/js/main.js` | the page and its entry: boot, the router (title → lobby → room → game) |
 | `public/js/net.js`, `public/js/store.js`, `public/js/data.js` | the socket client, the observable store, the data loader (`/data/*.json`, with the English overlay) |
 | `public/js/battle/` | `runner.js` (the local battle: loads `/sim/`, steps it, reports), `observe.js` (who may watch which field) |
-| `public/js/screens/` | `title.js`, `lobby.js`, `room.js`, `loadout.js` (干员调配), `ownership.js` (干员持有), `diy.js` (自选编队), `briefing.js`, `bandDraft.js`, `game.js` with `public/js/screens/game/`, `result.js` |
+| `public/js/screens/` | `title.js`, `lobby.js`, `room.js`, `loadout.js` (干员调配), `cultivation.js` (its 潜能 / 练度 controls), `ownership.js` (干员持有), `diy.js` (自选编队), `briefing.js`, `bandDraft.js`, `game.js` with `public/js/screens/game/`, `result.js` |
 | `public/js/ui/` | the HUD components (`hud.js`, `shopBar.js`, `detailPanel.js`, `bondStrip.js`, `teamPanel.js` …); `public/js/ui/gameLogic/` the pure in-match logic, unit-tested in Node |
 | `public/js/render/` | the field view: `app.js` with `public/js/render/app/`, `units.js` and `spine.js` (models), `tiles.js`, `projection.js`, `interp.js`, `pick.js`, `drag.js`, `public/js/render/fx/` (effects; `kinds.js` maps the fx kinds), `public/js/render/board3d/` (the official 3D board) |
 | `public/css/`, `public/i18n/<code>.json` | the styles; the UI strings of each language pack (English ships) |
@@ -20,7 +20,7 @@ npm run assets       # = node tools/vendor.mjs && node tools/fetch-assets.mjs
 | `--offline` | No network. Re-runs post-processing (atlas fixes, skeleton parsing, WOFF2) on what is already on disk, then rebuilds `data/assets.json`. |
 | `--dry-run` | Print the plan (file and model counts, alias notes) and exit. |
 | `--refresh-index` | Re-download the upstream indexes: `audio_data.json`, `charword_table.json` (the 干员战斗语音 slots) and `models_data.json`. |
-| `--voice-lang=cn` | 干员战斗语音 language: `cn` (default) | `jp` | `en` | `kr` — the same file names under `voice_cn/`, `voice/`, `voice_en/`, `voice_kr/`. |
+| `--voice-lang=cn` | The dub of `audio.voice` (the 中文 voice setting): `cn` (default) | `jp` | `en` | `kr` — the same file names under `voice_cn/`, `voice/`, `voice_en/`, `voice_kr/`. `audio.voiceJp` (the 日本語 setting) is always the JP dub: every run plans both trees. |
 | `--voice-all` | Plan every official voice slot, including the prep-only lines no battle plays (干员报到 / 编入队伍 / 任命队长 — 360 files, one per operator and slot). Off by default: nothing requests them, so planning them only makes every run download more. |
 | `--prune` | Delete files under `public/assets/` that the manifest no longer references, for example after a mapping change. Without this flag they are only listed in the report. `public/assets/local/` (written by `tools/local-extract`) is never pruned. Implies `--allow-shrink`. |
 | `--allow-shrink` | Write `data/assets.json` even when it loses entries the current one has (see "The manifest never shrinks by accident" below). |
@@ -53,7 +53,7 @@ How downloads are fetched:
 - A manifest entry with fallbacks (for example an enemy icon that falls back to its base enemy's icon) only moves on to the next alternative after a **definitive 404**. When the primary fails transiently (network error, 5xx or an invalid payload after all retries), no fallback is fetched. The path is listed under `downloadErrors` in the report, and the next run retries the primary.
 - A skeleton that fails to parse is deleted and removed from the ledger, so the next online run downloads it again.

-The first run downloads about **309 MiB in about 5,690 files** (it took 134 s on a ~3 MB/s link before the 55 emote and 玩法说明 files, 21.3 MiB, and the 1,680 干员战斗语音 files, 40.1 MiB, were added). A re-run takes about 1 s. The voice count is the twelve slots a battle plays; the three prep-only slots the official client uses elsewhere (干员报到 / 编入队伍 / 任命队长, 360 more files, 18.4 MiB) are left out unless `--voice-all` is passed.
+The first run downloads about **548 MiB in 10,643 files** (`stats` of `data/assets.json` in 0.2.2: 574,449,900 bytes). The two 干员战斗语音 dubs are 2,674 files each, the twelve slots a battle plays for 191 operators: the Chinese one 65.6 MiB, the Japanese one 85.2 MiB (89,388,041 bytes; adding it took 107 s at about 1 MB/s from raw.githubusercontent.com). A re-run takes about 1 s. The three prep-only slots the official client uses elsewhere (干员报到 / 编入队伍 / 任命队长, 360 more files per dub, 18.4 MiB of them Chinese) are left out unless `--voice-all` is passed.

 Outputs:
 - `data/assets.json`: the manifest (committed).
@@ -93,7 +93,7 @@ The research JSONs in `docs/research/` (03, 05, 07) define **which** ids are nee
 | Token Spine that no dump carries (39 summons: most 自选 summons, 凯瑟琳's 爬行号·防护单元, 凛御银灰's 风雪之眼) | the local client only (`tools/local-extract/extract.py TOKEN_SPINES`, optional); never downloaded and never required: an overlay (`tokens[id].spineLocal`; without it the avatar diamond) | `local/spine/token/{tokenId}/{stem}.*` (listed in `data/local-assets.json`) |
 | BGM | AA2 `voice` branch `audio/sound_beta_2/music/**` (大厅/休整期 `act1autochess`, 开战 `act13side/m_bat_kazimierz2_{1,2}` — 骑士之日 / 无畏者; the 开战 track follows the round: `_2` 无畏者 rounds 1–7, `_1` 骑士之日 from round 8) | `audio/bgm/{file}.mp3` |
 | SFX (UI, battle, per unit) | AA2 `voice` `audio/sound_beta_2/**`, mapped from `audio_data.json` banks | `audio/sfx/{same sub-path}.mp3` |
-| 干员战斗语音 | AA2 `voice` `audio/sound_beta_2/voice_cn/{charId}/cn_nn.mp3` — the lines `charword_table.json` lists (`placeType` = when the game plays one, `voiceAsset` = the path); `--voice-lang=jp|en|kr` takes the same file names from `voice/`, `voice_en/`, `voice_kr/` | `audio/voice/{lang}/{charId}/{cn_nn}.mp3` |
+| 干员战斗语音 | AA2 `voice` `audio/sound_beta_2/voice_cn/{charId}/cn_nn.mp3` (Chinese, `audio.voice`) and `voice/{charId}/cn_nn.mp3` (Japanese, `audio.voiceJp`: the same file names) — the lines `charword_table.json` lists (`placeType` = when the game plays one, `voiceAsset` = the path); `--voice-lang=jp|en|kr` fills `audio.voice` from `voice/`, `voice_en/`, `voice_kr/` instead | `audio/voice/cn/{charId}/{cn_nn}.mp3`, `audio/voice/jp/{charId}/{cn_nn}.mp3` (`audio/voice/{lang}/…` for another `--voice-lang`) |
 | Fonts: Bender Regular and Light, Novecento Wide | TimWangZi/The-font-of-Arknights | `public/fonts/*.{otf,ttf,woff2}`, `public/fonts/fonts.css` |

 The `stem` of a Spine model is the upstream file name. Two examples: `char_107_liskam` has the stem `char_107_liskarm`, and `enemy_9032_aclionk` uses `enemy_1559_vtlionk`. The skel and atlas of a model always share one stem. pixi-spine locates the atlas by swapping the extension, so this matters.
@@ -148,7 +148,7 @@ All paths are URL paths relative to the site root, for example `/assets/char/ava
  hash: 'a1b2c3d4e5f6',             // content hash (cache busting)
  generator: 'tools/fetch-assets.mjs',
  stats: { files, bytes, chars, charsWithBack, enemies, enemiesWithSpine, tokens, tokensWithSpine,
-           spineModels, bonds, items, bands, skills, modules, ui, sfxUnits, voiceChars },
+           spineModels, bonds, items, bands, skills, modules, ui, sfxUnits, voiceChars, voiceJpChars },
  chars:   { [charId]: { avatar, avatarE2?, portrait, portraitE2?, spine: { front: Spine, back?: Spine } } },
  enemies: { [enemyId]: { icon, spine?: Spine, spineAliasOf?: enemyId,
                          spineLocal?: { group, skel, atlas, textures, …Spine } } },
@@ -186,6 +186,12 @@ All paths are URL paths relative to the site root, for example `/assets/char/ava
                           // prep-only slots 干员报到 / 编入队伍 / 任命队长 are left out of the plan by default —
                           // nothing requests them and they cost 360 files (19.3 MB) per run — and `--voice-all` adds
                           // them (audio.mjs VOICE_PREP_SLOTS) for the complete official set
+    voiceJp: { [charId]: { …the slots of `voice` } },
+                           // the Japanese dub (0.2.2, the owner's request 「全套的日配语音」): the same slots, lines and file
+                           // names as `voice`, under /assets/audio/voice/jp/ (AA2 `voice/`). The client plays it when the
+                           // player picks 日本語 (settings 语音语言, not tied to the interface language) and falls back to
+                           // `voice` for a slot it lacks or a file the host does not have (public/js/audio.js voiceLine);
+                           // the full zip ships it unless tools/package.mjs FULL_ZIP_JP_VOICE is off
    sfx: {
      ui:     { click, back, confirm, tab, pick, drop, error, buy, sell, income, refresh, freeze, levelup,
                merge, equip, itemMerge, bondUp, artPlace, ready, timer, draft, yourTurn, yourTurnCircle,
@@ -273,8 +279,12 @@ catch-up frames and hidden tabs (`keepsState`), the game screen's pre-entry buff
 event queue (`render/interp.js isCosmeticEvent`) too — or, for a view built mid-battle, UnitInfo `form`, which `render/app/info.js renderInfo` passes to the view; a
 `change` clip plays once first, an `end` clip is timed from the fx's `dur` to finish as that state ends, keeping the
 current form's death clip until the next form's fx). A blocked or revealed 隐匿 enemy is drawn solid: the sim sends the
-stealth bit only while its 隐匿 is on:
+stealth bit only while its 隐匿 is on — and 假想敌：骨刺 changes model with that bit, not with an fx (`STEALTH_FORMS`, GitHub
+#296, PR #365): the manifest's `*_A` (the three-headed snake) while the bit is on, its `*_B` set (form `revealed`) the
+moment it is blocked or revealed, `*_A` again when its 隐匿 is back; no change clip, a running attack, stun or freeze
+pose carried over (`render/spine.js syncFormPose`). The forms:
 - 掠海漂移体's crawl (`Change`, then `*_02`);
+- 假想敌：骨刺 (`*_A` / `*_B`, above);
 - 转译基底·α's three forms (`A_Die_B` / `_C` / `_D`, 2 s each, then `B_*` 寻仇者, `C_*` 幽灵, `D_*` 特战术师);
 - the 深池逐火 embers (`Die`, then `Idle_2` / `Move_2` / `Die_2`; `Revive` ends as it stands up) and 假想敌：再生's puppet
  (`A_Die`, then `B_*`; `B_Revive`);
@@ -393,12 +393,13 @@ heuristics CPU 36 / 88 → 35 / 86 ms and 16 / 40 → 16 / 40 ms.

 * Bounties with battles left (multi-round cards — "之后的每场作战" — last two battles since playtest #6, META §1.2)
  also spawn in the Final Assault / Hidden Core, on the owner's half (route to its goal), are shown in the boss-round
-  preview, and the boss battle uses up one of their battles (kill coins → pending funds for the Hidden Core prep).
+  preview, and the boss battle uses up one of their battles (kill coins, and a perfect-payout 战术特训 card's coins when the player's own field is perfect and the team wins → pending funds for the Hidden Core prep).
 * A 驰援 card is never offered when its bond has no chess left in the match's pool (every member banned).
 * docs/META.md §2.6 example guards `ctx.source.kind === 'choice'` (its EffectRef reuses the handler's key).
 * 坚若磐石 = least LP lost (`titles.comment_3 { stat: 'lpLost', rule: 'min' }` in tuning.json; results.js supports
  `rule: 'min'` among the players still alive).
-* 教鞭 / “神秘顾客” stay a random bounty, documented in docs/META.md §2.5.
+* 教鞭 offers a personal choice of three 战术特训 cards (a bot picks by the bounty scorer, the prep's deadline at random);
+  “神秘顾客” stays a random bounty — docs/META.md §2.5.

 ---

@@ -12,7 +12,7 @@ downloads failed on this machine unless `--allow-shrink` (or `--prune`) is passe
 ```
 node tools/build-data.mjs              # build (downloads missing official files into .cache/gamedata/)
 node tools/build-data.mjs --offline    # never download; fail if a cached file is missing
-node tools/build-data.mjs --refresh    # re-download every official file
+node tools/build-data.mjs --refresh    # re-download every official file; fail if any download fails
 node tools/build-data.mjs --no-research --out /tmp/x   # research-free build (fallback defaults), other dir
 node tools/build-data.mjs --cache <dir> --report <file> # other cache dir / report file (tests, CI)
 node tools/build-data.mjs --force      # write data/ even when integrity checks fail (debugging only)
@@ -23,8 +23,11 @@ Unknown options or a missing option value are errors (exit code 2); `--refresh`
 - **Cache.** Official files live under `.cache/gamedata/<repo path>` (`excel/activity_table.json`,
  `levels/enemydata/enemy_database.json`, `levels/activities/act1autochess/level_*.json`, …) and are fetched from
  `https://raw.githubusercontent.com/Kengxxiao/ArknightsGameData/master/zh_CN/gamedata/<path>` when missing.
-  Downloads are validated (must parse as JSON) and written atomically.
- **Report.** Every run writes `.cache/build-data-report.json` (counts, byte sizes, warnings (each once), errors, `written`).
+  Downloads are validated (must parse as JSON) and written atomically. With `--refresh`, a download that still fails
+  after four attempts aborts the build (exit code 1), even if an older cached file exists. Generated files and the
+  previous report remain untouched, including with `--force`; successfully downloaded cache entries are not rolled
+  back. Retry `--refresh` to complete the refresh; cached files are not verified to share an upstream revision.
+- **Report.** A run that reaches integrity validation writes `.cache/build-data-report.json` (counts, byte sizes, warnings (each once), errors, `written`).
  Integrity errors (see §17) make the exit code 1 **and leave the previous output untouched** (unless `--force`);
  warnings never do. Each output file is written atomically (temp file + rename).
 - **Determinism.** Same inputs ⇒ byte-identical outputs (stable key order, no timestamps, no randomness).
@@ -161,7 +164,7 @@ Top level: `{ season, seasonName, modes, economy, lpCapPerRound, bossOvertimeAft
 | `price`, `sellPrice` | `2`, `1` | |
 | `upgradeNum`, `upgradeChessId` | `3`, `"chess_char_1_01_b"` | copies needed to merge (风丸 2; golden 0) |
 | `status` | `{"phase":2,"level":50,"skillLevel":7,"equipLevel":1}` | training status used for all numbers |
-| `stats` | `{"maxHp":1421,"atk":531,"def":167,"res":0,"cost":12,"blockCnt":1,"bat":1,"aspd":100,"respawnTime":70,"spRecovery":1,"hpRecoveryPerSec":0,"moveSpeed":1,"tauntLevel":0,"massLevel":0,"deployLimit":1,"deckStack":0}` | keyframes linearly interpolated at `status` **plus every potential attribute modifier** (full potential, 潜能 6: character_table `potentialRanks` 1–5 — 「攻击力+N」, 「部署费用-1」, 「再部署时间-N秒」, 生命 / 防御 / 法抗 / 攻速 — added to the raw values; the owner's decision of 2026-10-07, GitHub #252 / PR #255: the official records carry no potential, a tournament video shows 刺玫 at ATK 435 / cost 15 = her E1 Lv55 413 / 17 + 攻击力+22 and two 部署费用-1; `tools/build-data.mjs OPERATOR_POTENTIAL`), hp/atk/def/cost/block/respawn rounded; **golden includes the module attribute bonus**. `aspd` 100 = base (module attack_speed added). Talent stat boosts are NOT folded in. |
+| `stats` | `{"maxHp":1421,"atk":531,"def":167,"res":0,"cost":12,"blockCnt":1,"bat":1,"aspd":100,"respawnTime":70,"spRecovery":1,"hpRecoveryPerSec":0,"moveSpeed":1,"tauntLevel":0,"massLevel":0,"deployLimit":1,"deckStack":0}` | keyframes linearly interpolated at `status` **plus every potential attribute modifier** (full potential, 潜能 6 — the default; character_table `potentialRanks` 1–5 — 「攻击力+N」, 「部署费用-1」, 「再部署时间-N秒」, 生命 / 防御 / 法抗 / 攻速 — added to the raw values: the official records carry no potential, a tournament video shows 刺玫 at ATK 435 / cost 15 = her E1 Lv55 413 / 17 + 攻击力+22 and two 部署费用-1; a lower potential — the player's 干员调配 setting since 0.2.2 — reads `potDown`, §2.3), hp/atk/def/cost/block/respawn rounded; **golden includes the module attribute bonus**. `aspd` 100 = base (module attack_speed added). Talent stat boosts are NOT folded in. |
 | `immunities` | `{"stun":false,"silence":false,"sleep":false,"frozen":false,"levitate":false}` | |
 | `rangeId`, `rangeGrid` | `"3-3"`, `[[1,0],[1,1],…]` | attack range at `status` |
 | `dmgType` | `"phys"` | `phys` / `arts` / `heal` (derived, see §2.1) |
@@ -169,15 +172,16 @@ Top level: `{ season, seasonName, modes, economy, lpCapPerRound, bossOvertimeAft
 | `projectile` | `"arrow"` | `arrow` (phys ranged) / `bolt` (arts ranged) / `orb` (heal) / `none` |
 | `canHitFly` | `true` | |
 | `targetPriority` | `"fly"` | from trait text: `fly` (优先攻击空中单位), `lowestDef` (防御力最低), else `null` |
-| `trait` | `{"desc":"优先攻击空中单位","descRaw":"…","bb":{"atk_scale":1.1},"bbStr":{},"rangeGrid":null,"moduleDesc":"攻击空中单位时攻击力提升至110%","moduleDescRaw":"…"}` | profession trait (+ golden module trait upgrade merged into `bb`; `rangeGrid` = trait-effect area, e.g. 散射手 front row — **not** the attack range). `desc` = the class trait, or the module's rewrite of it (official `overrideDescripton`); `moduleDesc` = the module's added line (official `additionalDescription`, PRTS 「特性追加」), which the client shows after `desc`, never alone (0.2.0, community report item 16.2) |
+| `trait` | `{"desc":"优先攻击空中单位","descRaw":"…","bb":{"atk_scale":1.1},"bbStr":{},"rangeGrid":null,"moduleDesc":"攻击空中单位时攻击力提升至110%","moduleDescRaw":"…"}` | profession trait (+ golden module trait upgrade merged into `bb`; `rangeGrid` = trait-effect area, e.g. 散射手 front row — **not** the attack range). `desc` = the class trait, or the module's rewrite of it (official `overrideDescripton`); `moduleDesc` = the module's added line (official `additionalDescription`, PRTS 「特性追加」), which the client shows after `desc`, never alone (0.2.0, community report item 16.2). A module part that only adds a display line (official target `DISPLAY`, no rewrite) fills that line with its blackboard and leaves `desc` at the class trait's numbers — 圣约送葬人 REA-Y: `desc` heals 50, `moduleDesc` 攻击速度+12 (0.2.2, GitHub #400); `bb` still merges every part (what the kits read) |
 | `skill` | see below | default skill at `status.skillLevel` (normal 4, golden 7) |
 | `skills[]` | `[{…skill record…, "index":0, "isDefault":false}, {…, "index":1, "isDefault":true}]` | **loadout choices** (DESIGN §16): every skill unlocked at `status` (E1 ⇒ S1–S2, E2 ⇒ S1–S3; the default is always listed), same shape as `skill` + `isDefault`, at the chess skill level, `trigger` resolved **for that skill index** (§2.2). The `isDefault` entry equals `skill` |
 | `modules[]`, `statsBase`, `traitBase`, `talentsBase` | see §2.2 | golden chess with `equipLevel > 0` only: selectable modules + the no-module base they apply to |
-| `talents[]` | `{"index":0,"name":"火力支援","desc":"…","descRaw":"…","bb":{"self_ammo":3,"duration":20,"ally_ammo":1},"bbStr":{},"rangeGrid":null,"tokenKey":null,"hidden":false,"fromModule":false}` | best unlocked candidate at `status` at full potential (candidates up to `requiredPotentialRank` 5: the 「天赋效果增强」 steps; module talent candidates the same); golden: module talent upgrades applied (`fromModule`); module data-only talents have `name:null, hidden:true`. A module upgrade of an existing talent **merges** blackboards (module keys win, base keys it does not restate are kept — e.g. 宴 keeps `min_attack_speed`; 仇白's upgrade adds `atk_scale_t` next to the old `atk_scale`: prefer the key the text uses). Module parts flagged `isToken` are **not** applied to the operator; they upgrade its summons (tokens.json variants). `containerTokenKey`: the official talent token id when it is a container missing from character_table (凛御银灰), `tokenKey` then holds the default skill's token |
+| `talents[]` | `{"index":0,"name":"火力支援","desc":"…","descRaw":"…","bb":{"self_ammo":3,"duration":20,"ally_ammo":1},"bbStr":{},"rangeGrid":null,"tokenKey":null,"hidden":false,"fromModule":false}` | best unlocked candidate at `status` at full potential (candidates up to `requiredPotentialRank` 5: the 「天赋效果增强」 steps; module talent candidates the same — an entry a lower potential changes chains its lower candidates, `potMin` / `potBelow`, §2.3); golden: module talent upgrades applied (`fromModule`); module data-only talents have `name:null, hidden:true`. A module upgrade of an existing talent **merges** blackboards (module keys win, base keys it does not restate are kept — e.g. 宴 keeps `min_attack_speed`; 仇白's upgrade adds `atk_scale_t` next to the old `atk_scale`: prefer the key the text uses). Module parts flagged `isToken` are **not** applied to the operator; they upgrade its summons (tokens.json variants). `containerTokenKey`: the official talent token id when it is a container missing from character_table (凛御银灰), `tokenKey` then holds the default skill's token |
 | `tokens[]` | `["token_10028_vigil_wolf"]` | summons (→ `tokens.json`): displayTokenDict + default-skill `overrideTokenKey` + talent `tokenKey`; `tokens.json → variants[chessId].sources` tells which (a `display`-only token is not produced by this chess's default skill or talents) |
 | `module` | `{"id":"uniequip_002_inside","name":"“最初的惊喜”","type":"MAR-X","level":1,"active":true}` | active only on golden chess |
 | `assets` | `{"avatar":"char_498_inside_2","portrait":"char_498_inside_2","spine":"char_498_inside","skillIcon":"skchr_inside_2","subProfIcon":"sub_fastshot_icon"}` | asset **ids** (URLs in `data/assets.json`); golden uses the E2 art when it exists |
 | `diyRequirement` | `"TIER_6"` | DIY only |
+| `potDown` | `{"0":{"stats.cost":17},"1":{"stats.respawnTime":70},"2":{"stats.atk":413},"4":{"stats.cost":16}}` (`chess_char_1_06_a`, 刺玫) | what a lower potential changes (§2.3; absent when nothing does: the DIY slots, 盟约·辅助干员) |

 `skill`:

@@ -242,6 +246,30 @@ everything needed to resolve a unit for `(chessId, skillIndex, moduleId)` (`simd
  variants[owner].byModule` (§14). Module choices never change the combat classification (`dmgType`…; the build warns
  if one ever would).

+### 2.3 Potential below full potential: `potDown` and chained talents (0.2.2)
+
+Potential (潜能 1–6) is the player's per-operator setting (干员调配, the owner's decision of 2026-10-08 — default 潜能 6;
+official: 调度手册 「卫戍协议中干员潜能由自身已持有干员潜能决定」). Every record is built at **full potential** and carries
+what each lower one changes; `shared/potential.js atPotential(rec, potential)` composes the record at a potential
+(data rank r = potential − 1, character_table's `potentialRanks[r]` being the step from rank r to r + 1). The same
+annotation is on chess records, `backups.json` unit forms (§18) and token variants (`tokens.json` and `backups.json`
+`tokens` variants, §14 — the owner's potential; a token record's top-level defaults stay full):
+
+- `potDown`: `{ "<rank>": { "<dotted.path>": value } }` — for each rank 0–4, the exact value at that rank of every leaf
+  that differs from the rank above (`stats.*` / `statsBase.*` of a record, `stats.*` of a form; a variant's
+  `stats.deployLimit`, `stats.deckStack`, `count`, `bySkill.<i>.count`, `byModule.<id>.stats.*`). Applied from rank 4
+  down to the wanted rank, each value overwriting, so every leaf ends at its value for that rank.
+- a talent entry (`talents`, `talentsBase`, `modules[].talentChanges`; a variant's `talents` and `byModule[*].talents`)
+  that a lower rank changes: `potMin` = the lowest rank it holds for, `potBelow` = the fields the entry below it
+  changes (`desc`, `descRaw`, `bb` …), itself chained the same way — the candidates by `requiredPotentialRank`
+  (刺玫's 土壤基肥改良 `potMin` 4: 11% → 8%; Mon3tr's 不毁重构 below its owner's rank 4: 3 s / 1200).
+
+`tools/build-data.mjs` runs the chess / backups / tokens builders at every rank (`ctx.potRank`) and annotates the
+full-rank build from the others; the build fails unless `atRank` rebuilds every rank exactly, and the unit-form parity
+(§18) holds at every rank. Nothing else depends on potential (no skill, trait or list length). The data translations
+(`data/i18n/*.json`) translate the chained texts too. 练度 (the 自持有 bonus) is no record field: it is the
+`char_attribute_mul` of the effects.json CHAR_MAP records `aceffect_char_1…4` (§7), applied to the unit.
+
 ---

 ## 3. `bonds.json` — `{ [bondId]: Bond }` (23 bonds, 8 core)
@@ -317,7 +345,7 @@ everything needed to resolve a unit for `(chessId, skillIndex, moduleId)` (`simd

 ## 7. `effects.json` — `{ [effectId]: Effect }` (361: every effect of the season)

-Types: `BAND_INITIAL` 41, `BOND` 23, `EQUIP` 115, `ENEMY_GAIN` 129 (悬赏), `BUFF_GAIN` 43 (战术决策), `ENEMY` 6 (`aceffect_enemy_*` global modifiers), `CHAR_MAP` 4.
+Types: `BAND_INITIAL` 41, `BOND` 23, `EQUIP` 115, `ENEMY_GAIN` 129 (悬赏), `BUFF_GAIN` 43 (战术决策), `ENEMY` 6 (`aceffect_enemy_*` global modifiers), `CHAR_MAP` 4 (角色练度映射 — the 自持有 练度 tiers `aceffect_char_1…4`, 未精英化 / 精英阶段1 / 精英阶段2 / 精英阶段2-60级, in `autoChessData.cultivateEffectList` order: their `char_attribute_mul` buff — `atk` / `def` / `max_hp` 1, 1.05, 1.1 — is the unit's 练度 multiplier, `shared/potential.js cultivateMul`, 0.2.2).

 | Field | Example | Meaning |
 |---|---|---|
@@ -442,7 +470,7 @@ as official level data, and sim tests use them as a plain two-halves road.
 | `special` | `{"mire":{"intervalSec":1,"aspdPerStack":-0.05,"moveMulPerStack":-0.05,"maxStacks":10,"heavyWeight":3,…}}`, `{"infection":{"bb":{"damage":70,"atk":0.2,"attack_speed":20,"duration":300}}}`, `{"deepsea":{"skillId":"sktok_tidectrl_3","bb":{…}}}`, `{"smog":{"rule":"operatorsNotTargetableByEnemyRanged"}}`, `{"blower":{…}}` | special-terrain parameters |
 | `runes[]`, `globalBuffs[]` | `{"key":"map_tile_blackb_assign","bb":{…},"bbStr":{…}}` | level runes (act2 m03 reed rune has no reed tiles — leftover) |
 | `deployTiles` | `{"normal":{"melee":[[9,3],…],"rangedOnly":[[10,4],…],"changedByDevices":[…]},"bossLeft":{…},"bossRight":{…}}` | deployable tiles at match start: `melee` = LOW (effective) buildable ALL/MELEE (ranged may use them too) — and a 深水区 under an active 特制水上平台 (`waterPlatform`, "在水上建立可以部署任意单位的平台"; act1 m05 only — the prep's deploy map only: the sim does not model the canoes, so its Grid keeps those tiles NONE, automatic placements avoid them and the path tie-break counts them as non-blockable [ASSUMED, weight 0 this season]) —, `rangedOnly` = high ground / active platforms; tiles under active crates/mounds and the 深水区 excluded. `bossLeft` / `bossRight` (field coordinates, boss rows 1–5) are what a player may deploy on in a boss round's prep: board (r, c) = boss (r − 7, c) / the right half mirrored (r − 7, 20 − c) (server/match/board.js `field`, user playtest #5 item 7) |
-| `groundPaths` / `groundPathsWithDevices` | `{"12,10->9,2":[[12,10],[12,9],…,[12,3],[11,3],[10,3],[9,3],[9,2]]}` | the ground route of the sim's own `server/sim/grid.js` flow field (research 08 §3.4: 4-direction SPFA from the goal, crates cost 1000, Bresenham line-of-sight smoothing — official route lengths — kept unless the remake's (0.1.0's) road-over-floor preference route crosses strictly fewer non-blockable floor / gate / 深水区 tiles — the fewest among equal-length chains, smoothing that never cuts across floor its grid route does not walk; a corner touch does not count as a crossing — e.g. 战场#01's lower gate takes the col-8 road, not the col-9 floor lane, while 战场#04's keeps the official diagonal from row 9 into row 10; DESIGN §14 corrections, test/sim/pathing-official.test.js) as the tiles an enemy crosses, over the whole map, without / with the match-start devices (crates cost 1000, platforms / mounds blocking [ASSUMED]); the battle re-reads its flow fields live |
+| `groundPaths` / `groundPathsWithDevices` | `{"12,10->9,2":[[12,10],[12,9],…,[12,3],[11,3],[10,3],[9,3],[9,2]]}` | the ground route of the sim's own `server/sim/grid.js` flow field (research 08 §3.4: 4-direction SPFA from the goal, crates cost 1000, Bresenham line-of-sight smoothing — official route lengths — kept unless the remake's (0.1.0's) road-over-floor preference route crosses strictly fewer non-blockable floor / gate / 深水区 tiles and no more 深水区 (GitHub #375) — the fewest among equal-length chains, smoothing that never cuts across floor its grid route does not walk; a corner touch does not count as a crossing — e.g. 战场#01's lower gate takes the col-8 road, not the col-9 floor lane, while 战场#04's keeps the official diagonal from row 9 into row 10; DESIGN §14 corrections, test/sim/pathing-official.test.js) as the tiles an enemy crosses, over the whole map, without / with the match-start devices (crates cost 1000, platforms / mounds blocking [ASSUMED]); the battle re-reads its flow fields live |
 | `options`, `config` | `{"characterLimit":8,"moveMultiplier":0.5}`, `{…configBlackBoard strings…}` | |

 Glyph legend (`rows`):
@@ -494,7 +522,7 @@ Glyph legend (`rows`):
 | `skill` | `{"skillId":"sktok_vigil_wolf_3","bb":{…}}` | default token skill (same slot as the owner's skill) |
 | `deployLimit`, `count` | `1`, `1` | `deployLimit` = the first owner's `stats.deployLimit`. `count` = copies sent to the hand / spawned (talent/skill `cnt`); `null` ⇒ use `deployLimit` |
 | `abnormal[]` | `["healFree"]` | abnormal effects the summon holds from the start, no official table carries them — `tools/build-data.mjs TOKEN_ABNORMAL` from the PRTS summon pages (user playtest #6 item 18): `healFree` = 禁疗 (“小自在”, “耀阳”, 斯卡蒂的海嗣, 沙之碑, 流形, 狼群, 迷迭香的战术装备, 黄金盟誓, 保护目标（冻结状态）), `isolated` = 孤立 "无法被同阵营选中" (“炎佑”, 从不混淆的方向); `[]` otherwise. The sim sets `noHeal` / `isolated` (docs/SIM.md §3) |
-| `variants[ownerChessId]` | `{"phase":2,"level":1,"stats":{…},"immunities":{…},"rangeGrid":…,"trait":{…},"dmgType":…,"skill":{full skill record},"talents":[…],"count":1,"sources":["talent","display"]}` | stats at the owner's phase/level (clamped to the token's max level) — **none of the owner's potential attribute modifiers** (a token has no potential ranks) — + golden module `tokenAttributeBlackboard` + the token's own talent additions to `deployLimit` / `deckStack` (blackboard `max_deploy_count` / `max_deck_stack_cnt` — the hidden "TOKEN数" talent of 麦哲伦's / 令's summons and 白铁's devices, 夜莺's 幻影; a module token part of the same talent replaces it: `tools/build-data.mjs tokenTalentDeckBonus`, 0.2.0 — no tokens.json summon has one; PRTS 幻影 备注 "最大可部署数量为3", the owners' "最多同时部署3个"); its `talents` / `trait` candidates are picked at the **owner's potential** (full: their `requiredPotentialRank` mirrors the owner's 「天赋效果增强」 — 夕's “小自在” 18 层, 凯尔希's Mon3tr, 望's 棋子 +1 持有 / 部署; `tools/build-data.mjs OPERATOR_POTENTIAL`); the owner's module parts flagged `isToken` upgrade the variant's `trait` (+`moduleDesc`) and `talents` (伺夜's wolves, 缪尔赛思's 流形 `scale` 1, 浊心斯卡蒂's 海嗣 30 s, “耀阳” `atk_scale` 1.15). `sources` ⊆ `talent`/`skill`/`display`: how the owner produces it (`display` only = listed on the character but unused by its default skill/talents, e.g. 迷迭香 S2, 凛御银灰 eagle1/3). `count` = copies from a talent `cnt` or the default skill's `cnt` when that skill overrides this token; `null` ⇒ use `deployLimit` |
+| `variants[ownerChessId]` | `{"phase":2,"level":1,"stats":{…},"immunities":{…},"rangeGrid":…,"trait":{…},"dmgType":…,"skill":{full skill record},"talents":[…],"count":1,"sources":["talent","display"]}` | stats at the owner's phase/level (clamped to the token's max level) — **none of the owner's potential attribute modifiers** (a token has no potential ranks) — + golden module `tokenAttributeBlackboard` + the token's own talent additions to `deployLimit` / `deckStack` (blackboard `max_deploy_count` / `max_deck_stack_cnt` — the hidden "TOKEN数" talent of 麦哲伦's / 令's summons and 白铁's devices, 夜莺's 幻影; a module token part of the same talent replaces it: `tools/build-data.mjs tokenTalentDeckBonus`, 0.2.0 — no tokens.json summon has one; PRTS 幻影 备注 "最大可部署数量为3", the owners' "最多同时部署3个"); its `talents` / `trait` candidates are picked at the **owner's potential** (full; their `requiredPotentialRank` mirrors the owner's 「天赋效果增强」 — 夕's “小自在” 18 层, 凯尔希's Mon3tr, 望's 棋子 +1 持有 / 部署 — and the variant chains the lower ones, §2.3: `atPotential(variant, ownerPotential)`); the owner's module parts flagged `isToken` upgrade the variant's `trait` (+`moduleDesc`) and `talents` (伺夜's wolves, 缪尔赛思's 流形 `scale` 1, 浊心斯卡蒂's 海嗣 30 s, “耀阳” `atk_scale` 1.15). `sources` ⊆ `talent`/`skill`/`display`: how the owner produces it (`display` only = listed on the character but unused by its default skill/talents, e.g. 迷迭香 S2, 凛御银灰 eagle1/3). `count` = copies from a talent `cnt` or the default skill's `cnt` when that skill overrides this token; `null` ⇒ use `deployLimit` |
 | `variants[o].bySkill[i]` | `{"skill":{…},"count":1,"sources":["talent","display"]}` | owner loadout with the non-default skill index `i` (one entry per other selectable owner skill): the token skill of that slot (伺夜's wolves, 缪尔赛思's 流形, 凛御银灰's eagles…), the count and how the chess then produces it (`sources` may be `[]`: 风丸 S1 makes no 纸偶; 赫默 / 巫恋 S1 only `display` ⇒ no hand piece). The sim resolves them for an owner loadout: `simdata getToken(id, ownerChessId, loadout)` → `def.sources` / `def.count` |
 | `variants[o].byModule[m]` | `{"stats":{…},"immunities":{…},"trait":{…},"talents":[…]}` | golden owner with another module `m` or `'none'`: the token as that module makes it (module `tokenAttributeBlackboard`, `isToken` trait/talent parts) |
 | `assets` | `{"avatar":"token_10028_vigil_wolf","spine":"token_10028_vigil_wolf"}` | |
@@ -574,7 +602,7 @@ Glyph legend (`rows`):

 `chess 266 (112 visible; 283 selectable skills over the visible chess, 184 module choices over 129 goldens; 74 PRESET / 55 NORMAL / 4 DIY base chess)`, `bonds 23`, `garrisons 249 (43 effect keys)`, `items 115`, `bands 40`, `effects 361`,
 `enemies 249`, `factions 67 entries`, `waves 38`, `stages 11 (8 active)`, `bosses 10`, `tokens 22`, `choice events 109`,
-`bounty cards 129`, `tactic cards 43`, `backups: 88 units (17 stand-ins, 71 owned-6★ picks; 256 forms), 38 自选 summons, 4 DIY slots, 15 / 9 prototype picks (tier 5 / 6), 71 owned-6★ picks (7 collab operators excluded)`.
+`bounty cards 129`, `tactic cards 43`, `backups: 88 units (17 stand-ins, 71 owned-6★ picks; 256 forms), 38 自选 summons, 4 DIY slots, 15 / 9 prototype picks (tier 5 / 6), 71 owned-6★ picks (7 collab operators excluded)`, potential (§2.3): `256 chess records, 219 forms / variants with potDown; 442 + 401 + 19 chained talent entries (chess, backups, tokens)`.

 ## 17. Integrity guarantees (checked by the builder and `test/data.test.js`)

@@ -623,8 +651,8 @@ server/match/player/diy.js, docs/META.md §3). The rules in the data (activity_t
 - A stand-in keeps the chess's bonds, 特质, tier, price, merge and status (the elite form uses the same backup at the
  elite row) and fights as `backup.charId` at that status with skill `backup.skillIndex` (fixed by the chess: the same
  character takes S2 on one chess and S3 on another), module `backup.uniEquipId` (null at tiers 3–4 and for every 4★, so
-  even their elite form has none) and the row's potential (`potRank` 0 on all 55) — moot: every unit form is built at
-  full potential like a chess (the owner's decision of 2026-10-07), and the 17 原型干员 have no potential ranks.
+  even their elite form has none) and the row's potential (`potRank` 0 on all 55) — moot: the 17 原型干员 have no
+  potential ranks, and a stand-in takes neither the player's potential nor the 自持有 练度 (0.2.2).

 `units[charId]` — first the 17 stand-ins, then the 71 owned-6★ 自选 picks (`diy.ownedPool`); no unit for a PRESET or DIY
 chess:
@@ -638,7 +666,7 @@ chess:
 | `forms[statusKey]` | keys `"2/1/4/0"`, `"2/60/7/1"`, `"2/60/7/3"` | the character at every status it fights at — of the chess it stands in for and of the DIY slots it may fill (`statusKey(status)` = `phase/level/skillLevel/equipLevel`): 3 forms per 6★ (an owned pick: exactly the three DIY slot statuses — E2 Lv1 skill rank 4 without a module, E2 Lv60 rank 7 with every module at stage 1 and at stage 3), 2 per 4★ |

 A **form** holds the operator fields of a chess record with **nothing selected**: `status`; `stats`, `trait`, `talents`
-**without** a module, at full potential as on a chess; `immunities`, `rangeId`, `rangeGrid`, `dmgType`, `attackKind`, `projectile`, `canHitFly`,
+**without** a module, at full potential as on a chess (`potDown` / chained talents for the lower ones, §2.3); `immunities`, `rangeId`, `rangeGrid`, `dmgType`, `attackKind`, `projectile`, `canHitFly`,
 `targetPriority`; `skills[]` — every skill unlocked at the status, at its skill level, `trigger` resolved per skill, no
 `isDefault`; `displayTokens` / `tokens` (summons — none of the 17 stand-ins has one; 27 of the owned picks do, their
 records in `tokens` below); at `equipLevel > 0` `modules[]` (§2.2 shape without `isDefault`). `buildUnitForm` uses buildChess's helpers and rules, and the build fails when `buildUnitForm` +
@@ -705,5 +733,6 @@ loadout. No id is also a tokens.json record.

 Not in this data: a player's ownership roster and 自选 picks (browser settings sent with `room.ownership` / `room.diy`),
 the DIY stock (per player in the match, `PlayerState.diyStock`: `config.economy.poolCopies` of the slot's tier — no excel
-field), 助战 borrows (`borrowCount` 20), the owned-operator training bonus (`prepareStateDict`), a per-player potential (every
-form is built at full potential, as every chess: the owner's decision of 2026-10-07).
+field), 助战 borrows (`borrowCount` 20), and a player's potential / 练度 per operator (the 干员调配 settings sent with
+`room.loadout`; the forms are built at full potential and carry the lower ranks, §2.3; 练度 = effects.json
+`aceffect_char_1…4`).
@@ -10,7 +10,7 @@
 | 服务器 CPU | 战斗在各玩家浏览器里模拟（DESIGN §14），服务器只负责回合、经济和校验：**每个房间每个作战回合约 1 ms CPU**。AI 队友 / 掉线玩家的战场由服务器模拟：作战开始时 3 个 AI 战场在开发机上约 0.2–0.5 s CPU，小主机上可能要几秒（分成 8 ms 小片执行，不会卡住其他房间）。`SP_VERIFY=all` 会复算每个真人战场，CPU 明显增加，小主机建议保持 `off` 或 `sample`。 |
 | 服务器内存 | 空闲约 100 MB，每个进行中的对局再增加几 MB。 |
 | 网络 | 4 人对局中服务器每回合下行约 0.25 MB（DESIGN §14 实测）。首次进入游戏时浏览器要从主机下载所需的图片 / Spine 模型 / 音频（按需加载，之后走浏览器缓存），公网隧道带宽小时第一次会慢一些。 |
-| 磁盘 | 素材约 460 MB（`public/assets`）+ 依赖约 125 MB（`node_modules`；整合包只带运行依赖，约 65 MB）；可选的本地提取约 40 MB（`.venv-extract`）+ 70 MB 贴图（见第 6 节）。完整包解压后约 625 MB。 |
+| 磁盘 | 素材约 550 MB（`public/assets`，含中文约 65 MB、日文约 85 MB 两套干员语音）+ 依赖约 125 MB（`node_modules`；整合包只带运行依赖，约 65 MB）；可选的本地提取约 40 MB（`.venv-extract`）+ 70 MB 贴图（见第 6 节）。完整包解压后约 710 MB。 |
 | 玩家设备 | 支持 WebGL 的现代浏览器（Chrome / Edge / Firefox / Safari 最新版），电脑或手机平板（横屏）。老旧设备可在设置里调低画质或访问 `/?board=2d`。 |

 服务器**无状态**：房间和对局只存在内存里，没有数据库和存档，**不需要备份**。重启服务器会结束正在进行的对局（包括断线后本可在 24 小时内回来继续的独立模拟）。
@@ -26,13 +26,13 @@
   ```
   装完**关闭并重新打开**终端，`node -v` 应显示 v22 或更高（winget 的 LTS 目前是 v24.x，同样可用）。没有 winget 时从 <https://nodejs.org/zh-cn/download> 和 <https://git-scm.com/download/win> 下载安装。
 2. 下载，三选一。建议放在一个固定、短、**不在 OneDrive 同步范围内**的目录，例如 `C:\Stronghold-Protocol`：
-   - **完整包（推荐）**：在仓库的 [Releases](https://github.com/sganggs/Stronghold-Protocol/releases) 页面下载最新版本的 `Stronghold-Protocol-v<版本>.zip`（约 430 MB，解压后约 625 MB；已含运行依赖、前端库和全部素材，包括官方 3D 棋盘等本地客户端素材），解压后把里面的 `Stronghold-Protocol` 文件夹放到上述位置。不需要 Git，首次启动也不用再下载素材。素材版权归上海鹰角网络 / Yostar，仅限非商业使用，见 [NOTICE.md](../NOTICE.md)。
-   - **精简包**：同一页面的 `Stronghold-Protocol-v<版本>-lite.zip`（约 22 MB）。代码、运行依赖和前端库与完整包相同，但不带素材：美术、Spine 模型、音频、字体、表情和「玩法说明」教程图在首次启动时由 setup 从公开镜像下载（约 460 MB，显示进度，可中断续传；镜像设置见下面的「国内镜像下载」）。官方 3D 棋盘等本地客户端素材需要用本机客户端提取，或从同一版本的完整包复制（第 6 节）。适合下载大文件不方便、或想先下一个小包的情况；放置方式同完整包。
+   - **完整包（推荐）**：在仓库的 [Releases](https://github.com/sganggs/Stronghold-Protocol/releases) 页面下载最新版本的 `Stronghold-Protocol-v<版本>.zip`（约 505 MB，解压后约 710 MB；已含运行依赖、前端库和全部素材，包括中文、日文两套干员语音和官方 3D 棋盘等本地客户端素材），解压后把里面的 `Stronghold-Protocol` 文件夹放到上述位置。不需要 Git，首次启动也不用再下载素材。素材版权归上海鹰角网络 / Yostar，仅限非商业使用，见 [NOTICE.md](../NOTICE.md)。
+   - **精简包**：同一页面的 `Stronghold-Protocol-v<版本>-lite.zip`（约 22 MB）。代码、运行依赖和前端库与完整包相同，但不带素材：美术、Spine 模型、音频（含两套干员语音）、字体、表情和「玩法说明」教程图在首次启动时由 setup 从公开镜像下载（约 550 MB，显示进度，可中断续传；镜像设置见下面的「国内镜像下载」）。官方 3D 棋盘等本地客户端素材需要用本机客户端提取，或从同一版本的完整包复制（第 6 节）。适合下载大文件不方便、或想先下一个小包的情况；放置方式同完整包。
   - **源码**：
     ```powershell
     git clone https://github.com/sganggs/Stronghold-Protocol.git C:\Stronghold-Protocol
     ```
-3. 双击 `C:\Stronghold-Protocol\scripts\start-windows.bat`。首次会：安装依赖（`npm ci`；整合包已含，跳过）→ 复制前端库（整合包已含，跳过）→ 下载约 460 MB 素材（完整包已含，跳过；精简包和源码在这一步下载，显示进度，中断后再次启动会续传）→ 若检测到本机的明日方舟客户端，询问是否提取官方贴图（可跳过）→ 启动服务器并打开浏览器。
+3. 双击 `C:\Stronghold-Protocol\scripts\start-windows.bat`。首次会：安装依赖（`npm ci`；整合包已含，跳过）→ 复制前端库（整合包已含，跳过）→ 下载约 550 MB 素材（完整包已含，跳过；精简包和源码在这一步下载，显示进度，中断后再次启动会续传）→ 若检测到本机的明日方舟客户端，询问是否提取官方贴图（可跳过）→ 启动服务器并打开浏览器。
 4. 窗口里会打印朋友可用的地址，例如 `http://192.168.1.23:3000`。用另一台设备打开它确认能进入。关闭窗口即停止服务器。

 等价的手动命令：`npm ci`、`node tools/setup.mjs`、`npm start`。
@@ -106,7 +106,7 @@ powershell -ExecutionPolicy Bypass -File scripts\install-service-windows.ps1
 winget install NSSM.NSSM            # 或从 https://nssm.cc 下载
 nssm install StrongholdProtocol "C:\Program Files\nodejs\node.exe" server\index.js
 nssm set StrongholdProtocol AppDirectory C:\Stronghold-Protocol
-nssm set StrongholdProtocol AppEnvironmentExtra PORT=3000 HOST=0.0.0.0
+nssm set StrongholdProtocol AppEnvironmentExtra PORT=3000 HOST=::
 nssm set StrongholdProtocol AppStdout C:\Stronghold-Protocol\logs\server.log
 nssm set StrongholdProtocol AppStderr C:\Stronghold-Protocol\logs\server.log
 nssm start StrongholdProtocol
@@ -226,7 +226,7 @@ https / wss 说明：页面通过 https 打开时客户端自动连接 `wss://
 ## 3. Docker

 ```bash
-# A) 构建时下载素材（需要联网，约 460 MB）
+# A) 构建时下载素材（需要联网，约 550 MB）
 docker build -t stronghold-protocol --build-arg FETCH_ASSETS=1 .
 docker run -d --name stronghold -p 3000:3000 --restart unless-stopped stronghold-protocol

@@ -266,7 +266,7 @@ services:
  [Service]
  WorkingDirectory=/opt/Stronghold-Protocol
  ExecStart=/usr/bin/node server/index.js
-  Environment=PORT=3000 HOST=0.0.0.0
+  Environment=PORT=3000 HOST=::
  Restart=always
  RestartSec=5
  User=stronghold
@@ -323,7 +323,7 @@ npm run package:lite -- --out <目录>   # 精简包 Stronghold-Protocol-v<版
 npm run package -- --update --from <旧版本的完整包>[,<…>] --out <目录>   # 更新包 Stronghold-Protocol-v<版本>-update.zip
 ```

- **打进去的**：`git ls-files` 里的 `server/`、`shared/`、`data/`、`public/`（不含 `public/dev/`）、`packs/`（随仓库提交的内容包；只在本机安装、没提交的不打进去）、启动脚本、玩家会运行的工具（setup、vendor、fetch-assets 与 `tools/assets/`、doctor，以及 setup 调用的 `tools/local-extract/` 和 `crop-board-atlas.mjs`）、服务器和 fetch-assets 读取的 4 张研究数据表（`docs/research/` 的 `03-operators`、`05-enemies`、`05-maps`、`07-assets` 四个 JSON）、`package.json` / `package-lock.json`、许可证与说明（`LICENSE`、`NOTICE.md`、`THIRD-PARTY-NOTICES.md`、`README.md`、`CHANGELOG.md`）、`docs/PLAYING.md` 和本文；然后在临时目录里生成 `packs/index.json`（打进去的语言包和内容包的列表，供纯静态托管使用；服务器自己会实时列出，见 [PACKS.md](PACKS.md)），再 `npm ci --omit=dev` 装上运行依赖和 `public/vendor`。完整包再加上 `data/assets.json` 列出的素材、`public/fonts`，以及本地提取的 `public/assets/local/` 和 `data/local-assets.json`。磁盘上有、清单却没列出的文件不打进去（例如 0.2.0 移出自选的焰狐龙梓兰的旧素材）。
+- **打进去的**：`git ls-files` 里的 `server/`、`shared/`、`data/`、`public/`（不含 `public/dev/`）、`packs/`（随仓库提交的内容包；只在本机安装、没提交的不打进去）、启动脚本、玩家会运行的工具（setup、vendor、fetch-assets 与 `tools/assets/`、doctor，以及 setup 调用的 `tools/local-extract/` 和 `crop-board-atlas.mjs`）、服务器和 fetch-assets 读取的 4 张研究数据表（`docs/research/` 的 `03-operators`、`05-enemies`、`05-maps`、`07-assets` 四个 JSON）、`package.json` / `package-lock.json`、许可证与说明（`LICENSE`、`NOTICE.md`、`THIRD-PARTY-NOTICES.md`、`README.md`、`CHANGELOG.md`）、`docs/PLAYING.md` 和本文；然后在临时目录里生成 `packs/index.json`（打进去的语言包和内容包的列表，供纯静态托管使用；服务器自己会实时列出，见 [PACKS.md](PACKS.md)），再 `npm ci --omit=dev` 装上运行依赖和 `public/vendor`。完整包再加上 `data/assets.json` 列出的素材、`public/fonts`，以及本地提取的 `public/assets/local/` 和 `data/local-assets.json`。磁盘上有、清单却没列出的文件不打进去（例如 0.2.0 移出自选的焰狐龙梓兰的旧素材）。日文干员语音（`audio.voiceJp`，`public/assets/audio/voice/jp/` 的 2674 个文件，约 85 MB，zip 后约 76 MB）默认也打进完整包；`tools/package.mjs` 里的开关 `FULL_ZIP_JP_VOICE` 改成 `false` 时，完整包（以及由它比较出的更新包）不带这些文件：玩家首次启动时 setup 会像精简包那样下载它们，下载完成前选「日本語」会播中文语音，更新包也不会删除玩家已有的日文语音。
 - **不打进去的**：`test/`、维护用的工具（数据构建、golden、botbench、i18n、导入检查、本工具等）、`scripts/make-windows-bundle.mjs`（Windows 便携包，见 [WINDOWS.md](WINDOWS.md)）、其他文档、研究笔记和 `docs/img/`、`handoff/`、`.github/`、`types/`、lint / 编辑器 / Docker 配置。和 0.1.x 的整树打包（全部跟踪文件加上 `public/assets` 的全部内容）相比，0.2.0 的完整包少了约 640 个文件、解压后小约 26 MB，zip 小约 8 MB。
 - **打包前的检查**（`--dry-run` 也全部做一遍）：拒绝名单（`pv`、`review`、`.cache`、`.claude`、`.git`、`logs`、`.env`、`scripts/service.env.cmd`、`handoff`、`test` 等）；每个打进去的模块的相对导入、玩家用的 npm 脚本（start / setup / doctor / launch / postinstall / vendor / assets）都指向包里的文件；完整包里 `data/assets.json` 和 `data/local-assets.json` 列出的文件都在（缺了先运行 `node tools/fetch-assets.mjs`）；没有只差大小写的两个路径；包里的文件（二进制素材也查）不含个人目录路径（`/Users/…`、`C:\Users\…`、`/home/…`）或本机的账户名（运行时从系统读取；`SP_PACKAGE_SCAN_NAMES=a,b` 可以再加名字）；打进去的已跟踪文件没有未提交的改动（重新生成的 `data/assets.json` 要先提交）。有任何问题都会列出原因、以非零状态结束，不写 zip；正式打包时还会核对临时目录里的文件和计划完全一致。
 - **MANIFEST.json**：三种包的根目录都有（`npm ci` 之后写入）：除素材（`public/assets/`、`public/fonts/`、`data/assets.json`、`data/local-assets.json`，归 setup 管）以外每个文件的大小和 sha256，同一版本的三种包内容相同。`npm run doctor` 和更新包的启动检查（`server/update.js`）用它核对安装。
@@ -4,7 +4,7 @@ This is the **single source of truth** for every implementer. Research lives in

 Language: player-facing text is **Simplified Chinese** by default, with an English switch since 0.2.0 (§25.2, docs/I18N.md: UI strings through `t('中文')`, game texts from the official Chinese data or the official EN client's). Code, comments and identifiers are English.

-Versions: the first public release was **0.1.0** (2026-10-02, the state of §0–§20.15); the releases after it add the player feedback and GitHub reports — 0.1.1 §21, 0.1.2 §22, 0.1.3 §23, 0.1.4 §24 — and **0.2.0** (2026-10-07; `CHANGELOG.md`) the maintainability refactor, two languages, 补位, 自选编队 and their fidelity work — §25 — and **0.2.1** (2026-10-07; `package.json`, `shared/constants.js APP_VERSION`) the 联防 battlefield restored, full potential and the GitHub fixes after it — §26 — and is the state described by this document. The labels v1 / v2 / v2.1–v2.5.2 in §0, §14–§20 and in the BALANCE / SIM comparisons name the design generations and the private playtest builds that came before it; they are kept as history.
+Versions: the first public release was **0.1.0** (2026-10-02, the state of §0–§20.15); the releases after it add the player feedback and GitHub reports — 0.1.1 §21, 0.1.2 §22, 0.1.3 §23, 0.1.4 §24 — and **0.2.0** (2026-10-07; `CHANGELOG.md`) the maintainability refactor, two languages, 补位, 自选编队 and their fidelity work — §25 — **0.2.1** (2026-10-07) the 联防 battlefield restored, full potential and the GitHub fixes after it — §26 — and **0.2.2** (2026-10-09; `package.json`, `shared/constants.js APP_VERSION`) per-operator 潜能 / 练度, Japanese voices, the statistics page, 失衡, whole-frame attack timing and the GitHub fixes after 0.2.1 — §27, the state described by this document. The labels v1 / v2 / v2.1–v2.5.2 in §0, §14–§20 and in the BALANCE / SIM comparisons name the design generations and the private playtest builds that came before it; they are kept as history.

 ---

@@ -33,3 +33,4 @@ normative lines it rewrote).
 | §24 | [history/0.1.4.md](history/0.1.4.md) | 0.1.4 — community reports after 0.1.3 |
 | §25 | [history/0.2.0.md](history/0.2.0.md) | 0.2.0 |
 | §26 | [history/0.2.1.md](history/0.2.1.md) | 0.2.1 — after the 0.2.0 release |
+| §27 | [history/0.2.2.md](history/0.2.2.md) | 0.2.2 — after the 0.2.1 release (2026-10-09) |
@@ -41,6 +41,14 @@ server/match/

 ## 1. Match flow

+The INFO_CHECK page supports a host-proposed setup reroll (remake convenience proposal, GitHub #77 / PR #408).
+Every human must agree; AI and spectators do not vote, and one human rerolls immediately. Voting pauses the
+confirmation deadline and queued all-ready transition. Refusal, host cancellation, disconnect or departure restores
+the previous remaining time and ready states. Success redraws every random setup field and rebuilds shared / DIY
+stock, resets confirmation and its full deadline, and preserves the room, personal settings (including 潜能 / 练度)
+and the AI-picks-last option. Single-human confirmation remains untimed. `match/setupVote.js` owns the state;
+`Lobby.rerollSetup` checks host permission. The wire fields and stale-request checks are in DESIGN §8.1.
+
 ```
 LOBBY → INFO_CHECK (co-op 25 s; solo and single-human matches untimed; all humans confirmed ⇒ next) → BAND_DRAFT → BATTLE_CHECK (3 s)
 → for r = 1..lastRound (+ hidden):
@@ -175,18 +183,18 @@ A `choice:<effectId>` registry handler overrides the default application (§2.4)

 ### 1.3 Disconnects, AI takeover
 * Disconnected human: the seat keeps playing its last lineup; drafts auto-resolve at their deadlines, prep auto-readies at
-  the deadline (the temp pieces due at that prep sold/destroyed). Nothing is bought for them. A battle the human was authority of goes to the server
+  the deadline (an open 教鞭 choice is picked at random first, §2.5; then the temp pieces due at that prep are sold/destroyed). Nothing is bought for them. A battle the human was authority of goes to the server
  (normal / 联防: re-simulated from t = 0) or, on a boss field, to the partner's replica (DESIGN §14). The session stays
  resumable for 10 min (net.js `reconnectWindowMs`); a **solo** run's for the official `constants.singleReconnectTime`
  (86 400 s = 24 h, lobby.js `soloResumeWindowMs`) — nothing in a solo run is timed, so it simply waits. `onReconnect`
  resends `m.public`, `m.private` and the `b.start` of the field the player is on / watching.
-* `g.autoplay { on }` ("AI 托管"): the bot plays the seat (drafts, buying, placement, ready) until turned off.
+* `g.autoplay { on }` ("AI 托管"): the bot plays the seat (drafts, buying, placement, an open 教鞭 choice, ready) until turned off.
 * `onLeave` (quit / reconnect window expired): 中途退出 counts as elimination (research 00-INDEX §3, 01 §9, 06 §7 /
  §10.3): every copy the seat holds returns to the shared pool at once; the seat leaves the round loop and the Final
  Assault pairing (re-planned when it quits before the boss fight) and the boss pool (bloodPoint per player alive when
  the fight starts — a departed seat no longer counts, DESIGN §25.13.4); its
  running normal battle is force-ended; a pending band pick
-  becomes the default band and a 机变 turn passes on. Status `left`, LP 0, rounds passed = the rounds it had survived.
+  becomes the default band, a 机变 turn passes on and an open 教鞭 choice goes with the seat (`eliminate`). Status `left`, LP 0, rounds passed = the rounds it had survived.
  When no human is left at all the match ends immediately (`reason: 'abandoned'`); when only eliminated spectators are
  left it ends as `'eliminated'`.

@@ -263,7 +271,8 @@ like any owned unit — AI 托管 too; the buy loop's bench shed, which sells by
 never sells a piece gained since the bot's last prep ended, `rememberOwned`), 身份牌 / 通讯机 / 寻呼模块 on a focus
 member; 画卷 copies the most
 valuable deployed operator; 教鞭 / “神秘顾客” are used after a perfect battle and otherwise kept in the hand — a bot drops
-a kept one only when it needs the hand slot (“神秘顾客” then pays its fund), a human's seat under AI 托管 never. 机变: a
+a kept one only when it needs the hand slot (“神秘顾客” then pays its fund), a human's seat under AI 托管 never; 教鞭's three
+cards are scored like a 机变 bounty draft (§2.5). 机变: a
 bounty card is scored by its expected payout minus its expected leaks × the value of an LP (2 + 20 / LP), the kill
 chance from the exposure model below for that one enemy against the own board as it stands (its HP × the round's
 multiplier, DEF / RES, speed, route kind) — an expected-value comparison: it prefers the card with the best payout
@@ -275,7 +284,9 @@ its enemies; an exposure model (time on each tile × DPS against the round's arm
 unit's cover is the range it is deployed with under the player's loadout, `rangeRec` = `attackRangeGrid`, the grid the
 server's `summonRange` and the card use —, blocker hold time, anti-air only on flying routes) is maximized greedily
 (blockers, then damage dealers, then healers) over the server's deploy map (no 深水区, PlayerState.deployMap); a
-tactician's 援军 goes on a tile of the tactician's attack range (`PlayerState.summonRange`, the server's own rule; player
+blocker whose range is its own tile only (range 0-1, the one grid (0,0): 角峰, 古米, 泡泡, 折桠, 菲莱, 蛇屠箱, 塞雷娅, 余) takes a free
+tile of the enemy road first — it blocks and hits nothing off it — and only when the road has none free any legal tile (0.2.2, PR #339 by
+@IceCodeNew); a tactician's 援军 goes on a tile of the tactician's attack range (`PlayerState.summonRange`, the server's own rule; player
 report #9 after 0.1.0), and a tile the server's `g.move` refuses is skipped for the next best one (the planner reads
 `board.js legalTiles` on the deploy map; `g.move` is the judge). Boss
 rounds: the player's boss-field template (`Match.bossWaves`) is mapped onto the own board (rows −7, the right player
@@ -425,6 +436,7 @@ Writes (all validated, never throw on bad input, never make funds / pools negati
 | `setDeviceActive(alias, on)` / `setTileOverride(r, c, 'melee'\|'ranged'\|'none')` | terrain changes of this player's board (legality + battle input `deviceOverrides`) |
 | `addEffect(ref)` / `removeEffect(id)` / `setEffectCounter(id, v)` | EffectRefs `{ id, key?, name, desc, iconKind, iconId, counter?, battle=true, params?, data?, hidden? }` — shown in `m.private.effects`, passed to battles as `playerEffects` when `battle` |
 | `addBounty(card)` | a bounty (choices.json cards.bounty shape) on the next battles |
+| `offerBountyChoice(cards, sourceItemId)` | 教鞭 only: open a PERSONAL choice of up to three of `cards` (cards.bounty entries) for this player — shuffled once with the meta rng, kept in `m.private.personalChoice` until the owner confirms one (`g.choice { idx, choiceId }`) or the prep's deadline / an AI seat picks (§2.5); returns the intent result (`{ ok }`, or `{ error, detail }` for a phase that is not PREP, a Ready player, a choice already open or no card — before any rng draw or id, so the Art stays in the hand) |
 | `toast(text, kind)` / `ticker(text)` / `giftTicker(fromName, chessId)` | messages |
 | `teammates()` / `player(playerId)` | ctx objects of other alive players (team effects) |

@@ -458,20 +470,47 @@ bilibili BV1vzyVBuEN9 ≈ 8:24 / BV1Qkw1zMEoR ≈ 7:25, pointed out in PR #2 —
 is not consumed: PRTS 下半 记录 备注 "生效时，原干员销毁，获得一名高一阶的随机初始干员（最高六阶）", PRTS 卫戍协议/帮助 "佩戴的
 装备无法手动卸除，在失去该干员（干员出售、销毁、合并等）…时自动卸除", players re-inject it every round; player feedback after
 0.1.0), 画卷 (copy the operator in range with its
-items), 教鞭 / “神秘顾客” (a random bounty is added).
+items — the copied items arrive unequipped in the hand, PRTS 画卷 备注 "获得的装备为未装备状态"), 教鞭 (a personal choice of three 战术特训 cards), “神秘顾客” (a random bounty is added).

-**教鞭 / “神秘顾客” stay a random bounty (deliberate).** The official Arts open a personal 悬赏 choice
-("选择一项（特殊）悬赏任务进行挑战", `choice_event hunter_band_1`). The server applies a random bounty instead: content
-(server/sim/content/items/meta.js) draws 3 cards and takes one at random — 教鞭 from the 20 战术特训 cards (payout
-`perfect`: extra enemies, the card's coins when the own phase is perfect; PRTS 下半 记录 §法术 教鞭 "于3个战术特训的悬赏
-任务中选择一项", §机变阶段 "※以下悬赏任务仅由法术教鞭生成", 杜宾 加练！ "若自身战斗完美作战可获得资金"; user playtest #6
-item 4 review), “神秘顾客” (granted by nothing in act2) from the band-bounty family `enemyeffect_b_*` [ASSUMED] — each
-limited to enemies the mode can field (the built-in fallback: a random tier ≤ II bounty of `cards.bounty`), and adds it
-with `ctx.addBounty`. A personal choice overlay would need its own phase / protocol message outside SP_DRAFT (a
-second, simultaneous draft in co-op, with timers and AI takeover) for a rarely used Art, while the random pick keeps the
-risk / reward the item is about. The
-bounty then behaves like any other (next battles, 联防 payouts, Final Assault spawns). 神秘顾客's destroy clause (+1
-fund, the Art passes to the next alive player) is content too (`onDestroy`).
+**教鞭 offers a personal choice of three 战术特训 cards (0.2.2); “神秘顾客” stays a random bounty.** The official Arts open a
+personal 悬赏 choice ("选择一项（特殊）悬赏任务进行挑战", `choice_event hunter_band_1`, choiceType `PERSONAL_CHOOSE`). 教鞭
+(PRTS 下半 记录 §法术 教鞭 "使用后销毁，于3个战术特训的悬赏任务中选择一项"; §机变阶段 "※以下悬赏任务仅由法术教鞭生成"; 杜宾
+加练！ "<教鞭>：使用后为下场战斗添加额外敌人，若自身战斗完美作战可获得资金"; user playtest #6 item 4 review) offers three of
+the 20 战术特训 cards (payout `perfect`: extra enemies, the card's coins when the own phase is perfect), each limited to
+enemies the mode can field. Until 0.2.1 the server took one of the three at random — a deliberate simplification: a
+personal choice looked like it needed a phase of its own, a protocol message outside SP_DRAFT, a timer and an AI
+takeover for a rarely used Art. The owner decided on 2026-10-08 to follow the official choice (PR #353 by @Sukvii), and
+it fits in the existing parts:
+
+* **Offer.** Content (server/sim/content/items/meta.js) calls `ctx.offerBountyChoice(cards, itemId)` (`Match.offerBountyChoice`,
+  match/spDraft.js): the eligible cards are shuffled once with the meta rng and up to three different ones are kept in
+  `ps.personalChoice { id, round, sourceItemId, cards }`. The Art is used up (and counts for the two per round) when the
+  offer is made; nothing is applied yet. Refused — the Art stays in the hand, no rng draw, no id — outside the player's
+  PREP, for a Ready player, while an offer is already open and when no card is left to offer.
+* **Private.** Only the owner sees it: `m.private.personalChoice` (re-sent on a reconnect or a refresh, so the same cards
+  and id come back); never `m.public`, a scout's view or a spectator. Players have offers of their own, at the same time.
+* **Confirm.** `g.choice { idx, choiceId }` (`Match.pickPersonalChoice`) validates the owner, the id, the round and the
+  index, then adds the card with `addBounty` and closes the offer; a request without a `choiceId` is the public 机变 draft
+  as ever, and a stale, foreign or repeated id never falls through to it (`BAD_TARGET`). The client shows the overlay of
+  the 机变 draft (`ChoiceOverlay`, `personal`: the same two taps, no order, no turns) over a locked prep.
+* **No phase, no timer.** The offer lives inside PREP. Ready is refused while it is open (`setReady`: `BAD_TARGET`;
+  `m.private.canReady` false), and the prep's own deadline resolves it: in a timed co-op prep `prepDeadline` takes one of
+  the cards at random (the meta rng, like a 机变 turn that runs out) before the temp pieces and Ready; a solo or
+  single-human prep is untimed and has no timer for it, and a disconnected player waits for the deadline like for any
+  other prep action. A seat the engine plays — an AI teammate or a human under AI 托管 — picks like the bots do; a seat that
+  left is eliminated (`onLeave` → `eliminate`), which drops its open offer: nobody picks for it.
+* **The bots.** `bot.js` scores the three cards like a 机变 bounty draft (`botPickCard` → `bountyScore`: the expected
+  payout minus the expected leaks × the value of an LP, against the own board; the bots' seeded rng only breaks ties), so
+  the choice is deterministic for a seed: at the start of the bot's prep (an offer left by a human who turned 托管 on),
+  right after it cast the Art, and before Ready. When a step of the bot's prep fails, the scheduler's fallback readies the
+  seat with the random pick instead — no seat is ever left un-ready by an offer.
+* The chosen bounty is a bounty like any other: next battles (a multi-round card lasts two, §1.2), 联防 payouts, the Final
+  Assault / Hidden Core spawns — and in the boss rounds its perfect-battle coins (§3 Waves).
+
+“神秘顾客” (granted by nothing in act2, no PRTS list of its cards) remains random: three cards are drawn from the band-bounty
+family `enemyeffect_b_*` [ASSUMED], limited to enemies the mode can field, and one is added with `ctx.addBounty` (the
+built-in fallback: a random tier ≤ II bounty of `cards.bounty`). Its destroy clause (+1 fund, the Art passes to the next
+alive player) is content too (`onDestroy`).
 EffectRefs: `effect:builtin_round_coin`, `effect:builtin_gift`, `effect:builtin_next_buy_golden_item` (整备),
 `effect:builtin_next_buy_elite` (升华). An eliminated player gets no dispatch, except an EffectRef whose handler sets
 `afterElimination: true` (its `onRoundStart` still runs: `EffectDispatcher.dispatchEliminated`, called by
@@ -508,7 +547,7 @@ Any `choice:` handler whose EffectRef reuses its own key must guard like this (o
 * **Shop**: `shopSlots[level]` chess slots + item slot(s), copy-weighted rolls over remaining pool copies of unbanned,
  visible chess with tier ≤ level; item slot: tier by the same shares, uniform item within the tier. Level-up price =
  base per mode, −1 each round start (floor 0), reset to the next base after upgrading; an upgrade opens the new level's
-  extra slots at once with new cards (the cards shown stay); `MAX_LEVEL` at 6. One freeze
+  extra slots at once, empty (the cards shown stay; the empty slots fill on the next roll); `MAX_LEVEL` at 6. One freeze
  toggle freezes every unsold slot until the next round start; a manual refresh rerolls everything (new slots stay
  frozen). Unfrozen slots are cleared at combat start. Slot positions are stable (frozen slots keep their index).
 * **Pool**: copies 12/14/18/16/8/5 (缪尔赛思 4); a normal piece holds 1 copy, an elite 3; displays never reserve copies;
@@ -619,7 +658,16 @@ Any `choice:` handler whose EffectRef reuses its own key must guard like this (o
  `enemyScale[r]` × the tuning layer §3.1, bounties, boss templates, 联防 routing). Bounties with battles left (every one
  that lasts more than one battle; an official multi-round card lasts two, §1.2) also spawn in the Final Assault /
  Hidden Core, on the owner's half of the boss field (a route ending at its goal); the boss battle then uses up one of
-  the bounty's battles and its kill coins go to pending funds.
+  the bounty's battles and its kill coins go to pending funds — with, for every perfect-payout card (战术特训), its coins
+  as after a normal battle ("若各自行动阶段就达成完美作战，获得N资金"): once per player, when the team won (the first end the
+  server registered: the leader's pool empty), the result is a real one (not the stand-in of a field that never
+  reported) and the player's own field let no counted enemy through [ASSUMED: the text has no boss-battle exception, and
+  the card's enemies do come in the boss battle] — and, for a field a client reported (the boss path never re-simulates
+  it), the result agrees with what its authority reported while it fought: LP it asked the team to pay must have been
+  split in `b.progress.leaksBy` (per player, the enemy leaks; the rest is the leader's own "扣除目标生命" effects), and a
+  player's leaked list must show at least its reported leak LP (`Match._bossLeaksAgree`), so a result cannot hide a charged
+  leak or hand it to the other seat; a server-run field is the truth itself. They are spent in the Hidden Core's prep like
+  the kill coins.
 * **Combat time limit**: data `combatTimeLimit` (the level's `maxPlayTime`) counts REAL seconds of the forced 2×
  battle; the Battle / 联防 limit is `gd.combatTimeLimit(r)` = 2 × that in game seconds (`config.combatTimeScale`,
  default 2). Read as game seconds the rounds' own spawn schedules would not fit (R2's last flyer spawns at 43 s of
@@ -826,9 +874,11 @@ round was over. The official 1 s `broadcastBeginDelay` is not modelled.
  units > active bond > Σ active layers > standing > seat; the ranks marked 存疑 in PRTS). Unite enemies re-enter with
  their original stats on the official 联防 spawn timing (waves.js `buildUniteWave`, DATA.md §15 #22).
 * A manual refresh while frozen keeps the new slots frozen until the next round start.
-* A level-up opens the new level's extra slots at once, each with a new card drawn at the new level, and keeps the
-  cards shown (0.2.0, community report item 19; the official tutorial's 「升级后将出现更多的商品栏位」); the new card
-  follows the freeze toggle like a manual refresh's cards.
+* A level-up opens the new level's extra slots at once — empty, with no card drawn into them (official footage, GitHub
+  #332: bilibili BV1AXwuzdEys 1:39, a 1→2 upgrade with the new slot visible and empty; the official texts only ever say
+  「升级后将出现更多的商品栏位」/「增加刷新栏位」) — and keeps the cards shown. The empty slots fill on the next roll (a manual
+  refresh or the round start), like every unfrozen slot; the shop draws an empty slot as a bare frame, never 已招募. 0.2.0
+  drew a card into each new slot (community report item 19, one uncorroborated remark): reverted in 0.2.2.
 * Buying a second copy of an equipped normal item merges into the golden item in the hand (not equipped).
 * Promotions by effects (升华, 博士投影) keep the equipment; merges return it; 突变细胞's transformation returns it (the
  cell included) before its new operator is gained into the 整备区 — the carrier's tile is left empty (official footage,
@@ -6,6 +6,8 @@

 界面语言：默认中文；标题画面右上角或「设置」里可切换 **中文 / English / 日本語 / 한국어 / 繁體中文**（会记住选择，不用刷新）。日语、韩语、繁体中文的游戏文本取自官方日服 / 韩服 / 台服数据，界面文字暂未翻译——日语和韩语显示英文界面，繁体中文显示简体中文界面，欢迎社区补全（docs/I18N.md「Adding a language」）。英文的干员、技能、盟约、装备、敌人等游戏文本取自官方英文服数据，界面文字全部有英文；切到英文后仍显示中文的只有玩家名字（人机的名字也是）、多轮悬赏改写过的「接下来两场作战」说明，以及官方英文服还没有的游戏文本（部分自选干员和较新的模组）；「玩法说明」的教程图是官方的中文图片，标题有英文（docs/I18N.md）。

+干员语音（官方台词）：「设置」里的「干员语音」调音量，「语音语言」选 **中文 / 日本語**（默认中文，和界面语言无关，会记住选择；日语缺了哪一句就播那一句的中文）。点选干员时——作战区或整备区里的干员、商店和晋升奖励里的干员卡片——会说一句官方的「选中干员」语音，休整期也一样；连续点选时新的一句接替正在说的那句，但不会打断正在播放的部署、技能等作战语音。部署（每场第一个部署的干员说「行动出发」）、技能、接敌和结算语音只在作战中播放。替补干员和原型干员（预备干员、Sharp、Mechanist 等）、盟约·辅助干员和召唤物没有官方语音，点选时不出声。
+
 ## 1. 一局的流程

 在「确认本局信息」页，房主可以点击「重刷本局」。所有真人玩家（包括房主；AI 与观战者不参与）一致同意后，重新随机战场、敌方领袖、隐秘核心、特训敌人及完整回合安排、禁用盟约与干员，所有玩家重新确认。房间号、连接、干员调配、干员持有和自选编队配置保留；固定战场与模式固定禁用遵循原规则，随机结果可能重复。投票期间暂停确认倒计时与阶段跳转；有人拒绝、断线、退出，或房主取消时，保留原本信息并恢复剩余确认时间。只有一名真人玩家时，点击即可刷新。
@@ -19,11 +21,12 @@
 | 阶段 | 说明 |
 |---|---|
 | 确认本局信息 | 本局地图（标准模拟固定为「战场#01」；险境从 8 张地图中随机，绝境 / 终极从不含战场#01 的 7 张中随机）、敌方派系（三种特训敌人类型，如折射、飞行：和官方一样，每种占全局 15 个回合位中的 3 个，第 14 回合领袖战和第 15 回合隐秘核心也各占一个位，所以险境 / 绝境 / 终极第 1–13 回合中有 7–9 波是这三种类型、每种最多 3 波，其余各波是「特异」类型；每波只有一种类型）、**被禁用的盟约及其干员**、领袖预告。所有人确认后继续；同盟模拟最多 25 秒；只有你一名玩家时（独立模拟，或同盟模拟只带 AI 队友）不限时。 |
-| 选择策略 | 40 名策略（分队长）决定**初始目标生命值（20–45）**和一条永久规则。同盟：随机顺序轮流选，**每人 30 秒**，右上角和「决策中」显示同一个倒计时；AI 队友立即选完；队友已选的不能重复，每人可**跳过一次**（排到最后）；超时自动选你**当前选中**的策略（已被队友选走则选华法琳）。独立：自由选择、不限时；同盟模拟只有你一名玩家时也不限时。选的时候可以点左侧决策顺序下方的「查看禁用盟约与干员」，再看一遍「确认本局信息」里的盟约（置灰、禁用人数）和本局禁用干员：只能查看，倒计时照常走，底部也显示轮到谁和剩余秒数；点「关闭」、点窗口外面或按 Esc 关闭，轮到下一位时会自动关闭。 |
-| 机变 | 部分回合开始前出现：同盟共 6 张卡轮流各取 1 张（首位 30 秒，其余 16 秒；只有你一名玩家时不限时），独立 3 选 1。卡片包括装备、资金、干员、盟约层数、悬赏等；**悬赏决策**为你的作战添加悬赏敌人，卡上用蓝字标出持续多久：「下场作战」/「接下来两场作战」的只出现一场或两场，之后不再出现；官方的「多轮悬赏」（原文写「之后的每场作战」）在这里也按「接下来两场作战」处理，只出现两场（如教鞭的法术大师A2·多轮战术特训）。各回合的悬赏决策按官方对局截图来出：第 3 回合是 6 张「接下来两场作战」的悬赏（3 张 1 资金、2 张 2 资金、1 张 3 资金）；第 9 回合是各首领的「下场作战」悬赏和「源石虫·特训」；绝境 / 终极的第 11 回合多半是悬赏决策（也可能是机密商店或战术决策），其中的悬赏通常有 1 只特异巨兽，其余是各阵营的「下场战斗」悬赏。击倒悬赏敌人的人（联防中的队友也算）获得资金；右侧效果栏会写「还剩 N 场作战」。**机密商店**的 6 件装备都免费，和官方一样可能出现两件相同的装备（各算一张卡，可以分别被选走）；第 11 回合（绝境 / 终极）的机密商店按官方：2 件 VI 阶、至少 1 件 V 阶和 1 枚盟约之币，标准 / 险境较早回合的机密商店各阶装备都可能出现。**战术决策**也和官方一样可能出现两张相同的卡（官方截图里出现过两张「补给」；各算一张卡，可以分别被选走，每张只生效一次，两人各选一张全队卡则效果叠加）；第 11 回合（绝境 / 终极）的战术决策按官方截图只出盟誓、驰援、列装 / 财富 / 补给 / 整备 / 升华等增益卡，不出削弱敌人的卡和战场演变卡。每张卡左上是图标和名称，下面写着完整效果（与详情卡相同的官方描述，手机上也显示）；已被选走的卡变暗，右上角显示选走它的玩家头像（悬停可看名字）。选卡要**点两次**：第一次只选中（金框 +「确认选择 · 再次点击」），再点同一张或点右上的「确认选择」才生效；点别的卡换选，按 Esc 或点空白处取消。 |
-| 休整期 | 购买、摆阵、配装备、冻结 / 刷新、升级调度中心；同盟有倒计时（65–215 秒，随回合增加）；独立模拟、以及只有你一名玩家（只带 AI 队友）的同盟模拟不限时。所有人「准备就绪」即提前开战，全员就绪前可取消。 |
+| 选择策略 | 40 名策略（分队长）决定**初始目标生命值（20–45）**和一条永久规则。同盟：随机顺序轮流选，**每人 30 秒**，右上角和「决策中」显示同一个倒计时；AI 队友立即选完；队友已选的不能重复，每人可**跳过一次**（排到最后）；创建者可以在同盟等待室打开「**AI 队友最后选择**」（默认关闭）：选择策略和机变轮选时所有博士都排在 AI 队友前面（博士之间、AI 之间仍是随机顺序，AI 托管或掉线的博士也算博士），跳过时排到其他还没选的博士之后、AI 之前（没有其他博士时排到最后）；超时自动选你**当前选中**的策略（已被队友选走则选华法琳）。独立：自由选择、不限时；同盟模拟只有你一名玩家时也不限时。选的时候可以点左侧决策顺序下方的「查看禁用盟约与干员」，再看一遍「确认本局信息」里的盟约（置灰、禁用人数）和本局禁用干员：只能查看，倒计时照常走，底部也显示轮到谁和剩余秒数；点「关闭」、点窗口外面或按 Esc 关闭，轮到下一位时会自动关闭。 |
+| 机变 | 部分回合开始前出现：同盟共 6 张卡轮流各取 1 张（随机顺序，开启「AI 队友最后选择」时博士先选；首位 30 秒，其余 16 秒；只有你一名玩家时不限时），独立 3 选 1。卡片包括装备、资金、干员、盟约层数、悬赏等；**悬赏决策**为你的作战添加悬赏敌人，卡上用蓝字标出持续多久：「下场作战」/「接下来两场作战」的只出现一场或两场，之后不再出现；官方的「多轮悬赏」（原文写「之后的每场作战」）在这里也按「接下来两场作战」处理，只出现两场（如教鞭的法术大师A2·多轮战术特训）。各回合的悬赏决策按官方对局截图来出：第 3 回合是 6 张「接下来两场作战」的悬赏（3 张 1 资金、2 张 2 资金、1 张 3 资金）；第 9 回合是各首领的「下场作战」悬赏和「源石虫·特训」；绝境 / 终极的第 11 回合多半是悬赏决策（也可能是机密商店或战术决策），其中的悬赏通常有 1 只特异巨兽，其余是各阵营的「下场战斗」悬赏。击倒悬赏敌人的人（联防中的队友也算）获得资金；右侧效果栏会写「还剩 N 场作战」。**机密商店**的 6 件装备都免费，和官方一样可能出现两件相同的装备（各算一张卡，可以分别被选走）；第 11 回合（绝境 / 终极）的机密商店按官方：2 件 VI 阶、至少 1 件 V 阶和 1 枚盟约之币，标准 / 险境较早回合的机密商店各阶装备都可能出现。**战术决策**也和官方一样可能出现两张相同的卡（官方截图里出现过两张「补给」；各算一张卡，可以分别被选走，每张只生效一次，两人各选一张全队卡则效果叠加）；第 11 回合（绝境 / 终极）的战术决策按官方截图只出盟誓、驰援、列装 / 财富 / 补给 / 整备 / 升华等增益卡，不出削弱敌人的卡和战场演变卡。每张卡左上是图标和名称，下面写着完整效果（与详情卡相同的官方描述，手机上也显示）；已被选走的卡变暗，右上角显示选走它的玩家头像（悬停可看名字）。选卡要**点两次**：第一次只选中（金框 +「确认选择 · 再次点击」），再点同一张或点右上的「确认选择」才生效；点别的卡换选，按 Esc 或点空白处取消。 |
+| 休整期 | 购买、摆阵、配装备、冻结 / 刷新、升级调度中心；同盟有倒计时（65–215 秒，随回合增加）；独立模拟、以及只有你一名玩家（只带 AI 队友）的同盟模拟不限时。所有人「准备就绪」即提前开战，全员就绪前可取消；点「准备就绪」时商店自动收起、棋盘放大到全场，取消准备时商店再展开（资金旁的「收起」和商店标签「展开商店」随时可以手动切换）。 |
 - **特殊地形点一下就有说明**：战场上的**活性源石**、**沼泽**、**排气格栅**、**深水区**，以及**红门 / 蓝门 / 传送点**，点一下就会弹出它自己的说明卡——名称、机制，以及**这一关自己的数值**（比如战场#04 的活性源石：每秒 70 点真实伤害、攻击力 +20%、攻击速度 +20、持续 300 秒）。休整期和作战期都能点。普通地板 / 道路 / 高台点了不会弹东西，所以点一下仍然只是「取消选中、关掉卡片」。
-| 作战 | 每人在自己的战场各自作战（战斗在你的浏览器里模拟，2 倍速）。干员按**逐列从左到右、每列从上到下**的顺序（官方「从上到下>从左到右」：最左一列最先）自动部署并消耗部署费用（初始 10，+1/秒）。时限 45–115 秒（实际秒数，即屏幕上的倒计时），时间到仍存活的敌人算漏怪。有敌人进入蓝门时，顶栏的目标生命值会**立即**显示扣除后的数值（红字 + 「−N」，每回合至多 10 点），结算时确定；联防阶段标注「联防中」，队友拦下的部分最后不扣。左侧队伍栏同样显示队友的待扣生命值。被击倒的干员**留在倒下的格子上**（倒地姿势、略微变灰），之后也在这一格再部署（与官方一样；倒在其他干员或召唤物初始位置上的，会回到自己的初始位置躺下，召唤物已经消失也一样；唯一的例外是乌尔比安：三技能把他移到船锚处期间被击倒时，他回到自己的部署位躺下、再部署——这是项目维护者 2026-10-07 应社区反馈有意做出的改动，与官方「原地倒下」不同）；因自身效果强制退场的干员（史尔特尔「余烬」结束、耀骑士临光 2 技能结束等）同样倒在原地、在原地再部署；倒地干员所在的格子**不会被占用**，突袭盟约的再部署、援军、召唤物都会避开它。头上的圆圈是再部署倒计时：绿色圆环 + 秒数；变成黄色「DP」表示部署费用不够。血条下方的一行是**元素损伤**（与原版一致：元素图标 + 白条）：干员为带图标的彩色圆盘，敌人为小号纯色圆盘，右侧白条是剩余元素值，像血条一样从右往左减少，白条耗尽的那一下就是爆条（被大额元素损伤一次打掉一截时白条也会一次少一截，例如掠海漂移体每击约为 25%–50%）；爆条后进入爆发冷却（灼燃 / 神经 / 侵蚀 10 秒，敌方侵蚀 8 秒，凋亡 15 秒），这时白条变成**元素颜色**（神经青绿、侵蚀蓝、灼燃橙红、凋亡灰紫）并在冷却期间慢慢回满、图标闪烁——彩色条表示已经爆条、正在承受爆发效果，不是还剩的元素值；期间所有元素都不会累积也无法被治疗，结束后清零。元素爆发造成的是**无来源伤害**，不吃干员的增伤、穿透和装备加成（例如深巡打海嗣时灼燃爆发固定 7000），击杀仍记在爆条的干员名下；元素脆弱只提高元素伤害，不加快元素条；脆弱不影响元素伤害。已被击杀的敌人不会再被附加元素损伤。 |
+- **场地装置点一下也有说明**：**阻隔工事**（挡路：地面敌人绕着走，无路可走才撞碎它；作战中说明卡上有它的实时血量，被摧毁后点不出来）、**“双眼皮”炮台**（自动法术攻击、命中附加脆弱，按当前层数最高的盟约强化；还没被机械援助或机变卡启用的炮台不显示、点不出来）、**射击台**（远程位可部署的高台位）和**源石流发生装置**（吹风方向，以及对干员攻击力、敌人移动速度的加减）点一下同样弹出说明卡，数值都取自所在关卡的装置数据，装置名称按当前语言显示官方译名；被机变卡移除的箱子同样不显示。休整期和作战期都能点（GitHub #228，PR #229，感谢 @rickylxw）。
+| 作战 | 每人在自己的战场各自作战（战斗在你的浏览器里模拟，2 倍速）。干员按**逐列从左到右、每列从上到下**的顺序（官方「从上到下>从左到右」：最左一列最先）自动部署并消耗部署费用（初始 10，+1/秒）。时限 45–115 秒（实际秒数，即屏幕上的倒计时），时间到仍存活的敌人算漏怪。顶栏中间的胶囊是官方的「已解决 / 本回合敌人数」：打死一个、漏掉一个都算已解决（开局 0/3，漏一个 1/3，打死一个 2/3，打死会分裂的那只 3/3）；分裂出来的小怪、召唤物和领袖召唤的敌人不进分子也不进分母，但它们漏掉时仍然各自扣目标生命值、也会破坏完美作战（4 只磨砻全被打死、分裂出的 8 只全漏进蓝门：胶囊 4/4，扣 8 点），所以会召唤的领袖关胶囊也能填满。有敌人进入蓝门时，顶栏的目标生命值会**立即**显示扣除后的数值（红字 + 「−N」，每回合至多 10 点），结算时确定；联防阶段标注「联防中」，队友拦下的部分最后不扣。左侧队伍栏同样显示队友的待扣生命值。被击倒的干员**留在倒下的格子上**（倒地姿势、略微变灰），之后也在这一格再部署（与官方一样；倒在其他干员或召唤物初始位置上的，会回到自己的初始位置躺下，召唤物已经消失也一样；唯一的例外是乌尔比安：三技能把他移到船锚处期间被击倒时，他回到自己的部署位躺下、再部署——这是项目维护者 2026-10-07 应社区反馈有意做出的改动，与官方「原地倒下」不同）；因自身效果强制退场的干员（史尔特尔「余烬」结束、耀骑士临光 2 技能结束等）同样倒在原地、在原地再部署；倒地干员所在的格子**不会被占用**，突袭盟约的再部署、援军、召唤物都会避开它。头上的圆圈是再部署倒计时：绿色圆环 + 秒数；变成黄色「DP」表示部署费用不够。血条下方的一行是**元素损伤**（与原版一致：元素图标 + 白条）：干员为带图标的彩色圆盘，敌人为小号纯色圆盘，右侧白条是剩余元素值，像血条一样从右往左减少，白条耗尽的那一下就是爆条（被大额元素损伤一次打掉一截时白条也会一次少一截，例如掠海漂移体每击约为 25%–50%）；爆条后进入爆发冷却（灼燃 / 神经 / 侵蚀 10 秒，敌方侵蚀 8 秒，凋亡 15 秒），这时白条变成**元素颜色**（神经青绿、侵蚀蓝、灼燃橙红、凋亡灰紫）并在冷却期间慢慢回满、图标闪烁——彩色条表示已经爆条、正在承受爆发效果，不是还剩的元素值；期间所有元素都不会累积也无法被治疗，结束后清零。元素爆发造成的是**无来源伤害**，不吃干员的增伤、穿透和装备加成（例如深巡打海嗣时灼燃爆发固定 7000），击杀仍记在爆条的干员名下；元素脆弱只提高元素伤害，不加快元素条；脆弱不影响元素伤害。已被击杀的敌人不会再被附加元素损伤。 |
 | 联防 | 见第 5 节。 |
 | 结算 | 扣 LP、发放悬赏资金、结算层数；LP ≤ 0 的玩家淘汰（其干员回到公共卡池）。结算开始时屏幕正中跳出官方的「**作战结束**」框（约 3 秒后自己消失，不挡操作），下面一行只写**你自己**这一回合被扣的目标生命值：没扣是「**全员无伤！**」，扣了是「**生命值减少 −N**」；同时播官方的战斗结束音效（扣了、联防里没扣、普通回合没扣各一种）。联防回合见第 5 节。最终攻势和隐秘核心没有这一框（之后直接是结算称号或隐秘核心）；已淘汰的玩家和观战者不弹框。 |

@@ -48,9 +51,9 @@
 | 招募价格（按阶级 I–VI） | 2 / 3 / 3 / 3 / 4 / 4 |
 | 刷新 | 1 资金 |
 | 出售干员 | 任何干员（含精锐）都只返还 **1** 资金 |
-| 装备 / 法术 | 调度中心出售除特殊装备外的所有装备：4 把特殊维式重锤（战栗 / 坚固 / 加速 / 灼燃）只来自维多利亚盟约（每 25 层）和洛洛的定制品，突变细胞只来自昆图斯的策略。装备不能出售，整备区里的只能「销毁」（不返还）。已配发的装备锁定在干员身上：只能被第三件装备替换（弹出替换窗口，选择要替换的一件，它会被销毁；盟约之币、博士投影等装备即销毁的道具也一样，先销毁选中的那件再生效，干员留下一个空栏位，精锐干员不能使用博士投影，此时什么都不销毁），或在干员出售 / 合成精锐时退回整备区。**突变细胞**不会用完：战斗结束后装备者被销毁（部署在场上的话，原来的格子空出来，剩余可放置角色加 1），突变细胞和它的其他装备先退回整备区，可以再次配发给任意干员；随后获得一名高一阶（最高 VI）的随机普通干员（装备者是精锐也一样），和其他新获得的干员一样进入整备区（满了进入临时整备区），需要自己重新部署；如果它凑齐了合成所需的同名干员，照常合成精锐。**坚固维式重锤**首次受到致命伤害时锁血 8 秒，**每次部署一次**：被击倒再部署（或突袭跳跃再部署、被 M3茧甲 / 埃芒加德复活）后又能锁一次，娜仁图亚策略借来的锤子也一样；借用到期时正在锁血的会锁满 8 秒；站在阿戈尔干员身前被开场吞噬、自带源石溶剂掉血时也会用掉这一次 |
-| 冻结 | 免费，保留当前所有未买的格子到下一回合；冻结状态下仍可主动刷新 |
-| 升级调度中心（1→2 … 5→6） | 险境及以上 5 / 8 / 11 / 12 / 13；同盟标准 5 / 8 / 10 / 11 / 11；独立标准 1 / 1 / 5 / 8 / 10。**每过一回合价格 −1**（最低 0），升级后按下一级的价格重新计算。升级到干员栏位更多的等级（险境及以上 2 级 4 个、4 级 5 个；标准 2 级 4 个、6 级 5 个）时，**新栏位立刻刷出一名新干员**，已有的卡片不变 |
+| 装备 / 法术 | 调度中心出售除特殊装备外的所有装备：4 把特殊维式重锤（战栗 / 坚固 / 加速 / 灼燃）只来自维多利亚盟约（每 25 层）和洛洛的定制品，突变细胞只来自昆图斯的策略。装备不能出售，整备区里的只能「销毁」（不返还）。已配发的装备锁定在干员身上：只能被第三件装备替换（弹出替换窗口，选择要替换的一件，它会被销毁；盟约之币、博士投影等装备即销毁的道具也一样，先销毁选中的那件再生效，干员留下一个空栏位，精锐干员不能使用博士投影，此时什么都不销毁），或在干员出售 / 合成精锐时退回整备区。**突变细胞**不会用完：战斗结束后装备者被销毁（部署在场上的话，原来的格子空出来，剩余可放置角色加 1），突变细胞和它的其他装备先退回整备区，可以再次配发给任意干员；随后获得一名高一阶（最高 VI）的随机普通干员（装备者是精锐也一样），和其他新获得的干员一样进入整备区（满了进入临时整备区），需要自己重新部署；如果它凑齐了合成所需的同名干员，照常合成精锐。**坚固维式重锤**首次受到致命伤害时锁血 8 秒，**每次部署一次**：被击倒再部署（或突袭跳跃再部署、被 M3茧甲 / 埃芒加德复活）后又能锁一次，娜仁图亚策略借来的锤子也一样；借用到期时正在锁血的会锁满 8 秒；站在阿戈尔干员身前被开场吞噬、自带源石溶剂掉血时也会用掉这一次。**教鞭**（杜宾的策略每 2 个回合给一枚）拖到场上使用后即销毁，休整期里弹出 3 张战术特训悬赏（只有你自己看得到，刷新页面或断线重连后还是这 3 张），点一张再点一次确认，选中的那张成为接下来的悬赏；选定之前不能点「准备就绪」。同盟模式的休整期倒计时结束时还没选，会随机选一张；独立模拟和只有一名玩家时休整期不限时，所以不会自动选。AI 队友和 AI 托管的座位按悬赏评分自己选。战术特训的赏金要「各自行动阶段就达成完美作战」才给（最终攻势、隐秘核心里是你这一侧没有漏怪且全队获胜）；“神秘顾客”仍然随机给一个悬赏 |
+| 冻结 | 免费，保留当前所有未买的格子到下一回合；冻结状态下仍可主动刷新。策略「梓兰」的每次主动刷新会单独冻结送来的那张同名干员：只有这一张卡显示冰霜（上方的冻结按钮不亮），没有再刷新就留到下一回合，再手动刷新会连它一起重抽 |
+| 升级调度中心（1→2 … 5→6） | 险境及以上 5 / 8 / 11 / 12 / 13；同盟标准 5 / 8 / 10 / 11 / 11；独立标准 1 / 1 / 5 / 8 / 10。**每过一回合价格 −1**（最低 0），升级后按下一级的价格重新计算。升级到干员栏位更多的等级（险境及以上 2 级 4 个、4 级 5 个；标准 2 级 4 个、6 级 5 个）时，**新栏位立刻出现，但空着不补干员**（官方如此：升级只开位，新位在下一次刷新——手动刷新或回合开始的自动刷新——时才会出现干员；空栏位只画一个空框，不写「已招募」），已有的卡片不变 |
 | 其他收入 | 悬赏敌人、盟约「远见」「奇迹」、装备「骑士储蓄罐」、部分策略和特质 |

 调度中心等级决定商店能出现的**最高阶级**和格子数（险境及以上：干员 3 / 4 / 4 / 5 / 5 / 5 格 + 装备 1 格）。
@@ -62,8 +65,9 @@
 - **位置上限**：棋盘最多部署 **8** 名；整备区 **10** 格。整备区满时**不能购买或领取**任何干员、装备（包括能凑齐 3 名合成精锐的那一张，和官方一样），先出售、部署或配发腾出格子；新获得的单位（合成的精锐、奖励、退回的装备等）进入**临时整备区**（整备区前面的 5 格），整备区一空出格子，临时整备区的单位就**自动补进去**（从右往左）。临时整备区有单位时**不能准备就绪**，棋盘上会用红色虚线框标出这一排并说明规则：把它们放进整备区或战场、配发或使用，或者出售 / 销毁。休整期中进入临时整备区的单位，若到**本休整期结束**时仍留在那里就会被自动处理（干员返还卡池、装备销毁）；作战、结算期间或准备就绪之后才进入的（例如战斗奖励、结算时合成的精锐），会保留到**下一个休整期**结束，你一定有机会处理它们。
 - **晋升精锐**：3 名同名干员（无论是否部署）自动合成**精锐**（等级、技能、模组大幅提升；风丸只需 2 名），精锐不会再合成。如果被合成的干员里有**已经部署在战场上**的，精锐会直接出现在**那名干员的位置**（保持原来的朝向，不多占部署位；有两三名都在场上时，取最先部署的那名——最左一列优先，同一列从上到下）；否则精锐进入整备区（满了进入临时整备区）。原先配发的装备都退回整备区，需要重新配发；被合成干员的召唤物随之移除，场上的精锐会把自己的召唤物送进整备区。点选一张能合成的商店卡片时，精锐将出现的那一格会亮起金色边框，同时打开的详情卡里也会写明去向（「可晋升：精锐干员将出现在作战区原位置 / 将进入整备区」；详情卡会自动停靠在不遮挡这一格的一侧）。在回合开始或结算时因策略、奖励等完成的合成，精锐出现时同样会播放金色的晋升特效。
 - **晋级奖励**：每次晋升后获得一次特殊招募：从 3 名 **高一阶**（最高 VI）、**互不相同**的随机干员中**免费**挑 1 名（该阶剩余干员不足 3 名时由低一阶补足）。它不能刷新、不能冻结、本回合结束失效。
- **干员持有（补位）**：大厅、房间和「确认本局信息」里的「干员调配」按钮打开的界面有两个标签页：「干员调配」选择每名干员携带的技能和精锐的模组；「**干员持有**」可以把你没有的干员**下掉**（点一下卡片切换「持有 / 未持有」）。和官方一样只有 55 张「普通」棋子能下掉（界面按阶级列出本赛季商店里的 53 名，每张写着由谁替补、用几技能）；74 张「预设」棋子（多是 1–2 阶的常见干员）无论是否持有都由本人上场，不在其中。默认全部持有。下掉的干员和官方一样**显示为替补干员**：商店和晋升奖励的卡片、整备区和临时整备区里的单位、作战区、结算界面的阵容，都是替补干员的头像、名字和模型（例如银灰显示为 Sharp），带一个小小的「替补」标记；详情卡显示替补干员的立绘、名字、技能、天赋和模组，名字下面写着「银灰的替补」。**盟约、特质、阶级、价格、合成仍按原干员**（卡片和详情卡上的盟约、特质就是原干员的），作战时的属性、技能、天赋、模组、攻击范围、近战位 / 远程位都是替补干员的（例如银灰 → Sharp 的 2 技能「亮剑」）。同一名替补干员可能替补好几名干员（Sharp 替补忍冬、银灰、隐德来希等），它们按各自的原干员分别合成：看阶级和盟约区分，点开详情卡能看到替补的是谁。同盟模拟中队友查看你的整备区和作战区、你的盟约成员时，看到的也是替补干员。替补的技能和模组由棋子固定，「干员调配」对它不生效；替补干员没有召唤物（下掉缪尔赛思就没有流形）。这是**局外设置**：下一局开始时生效，对局中修改只保存到下一局；同盟模拟中只影响你自己的棋子，队友的棋子按队友的设置。没有官方的好友借用：想用本人上场，改回「持有」即可。「全部持有」一键恢复，「导出 / 导入」可以把设置带到别的浏览器；大厅的「干员调配」按钮上会显示「替补 N」提醒你有几名干员被下掉。
- **自选编队**：同一个界面的第三个标签页「**自选编队**」有 4 个名额：5 阶两个、6 阶两个。点名额打开选择列表，每个名额可以放一名**你持有的 6★ 干员**（棋子池里没有的那些，标「已持有」：3 个技能任选其一，精锐的模组任选其一或不装备；只在集成战略或生息演算里生效的模组不能选：ISW-A、SO-A / SO-B、RA-A）或一名**原型干员**（标「原型」：Sharp、Mechanist 等 9 名 6★，5 阶另有 6 名 4★ 预备干员；技能和模组与补位时相同，不能更改）。同一名原型干员可以同时放进一个 5 阶和一个 6 阶名额；你持有的干员只能放一个名额；同一阶的两个名额不能放同一人。列表只列出已经做好战斗逻辑的干员（原型干员全部可选，持有的 6★ 随战斗逻辑完成陆续加入）。对局中，自选干员**只出现在你自己的商店里**：调度中心升到 5 级后才会刷出 5 阶名额的干员，6 级后才会刷出 6 阶的；刷新概率和同阶干员一样，每个名额有自己的库存（5 阶 8 张、6 阶 5 张，不与队友共用），价格与同阶干员相同，晋升奖励里也可能出现；效果、奖励或机变卡「获得一名随机干员」时，也可能抽到你自己的自选干员（按那个效果自己的阶级规则，不看调度中心等级）。三张合成精锐，精锐时所选模组生效（5 阶 1 级、6 阶 3 级）。自选干员**没有特质**，盟约按干员的实际阵营分配（推进之王是维多利亚，煌同时是炎和维多利亚，没有对应阵营的是协防干员），和普通棋子一样计入盟约人数；本局禁用了它全部的盟约时，它不会出现在商店里（协防不会被禁用），开局的「确认本局信息」、选策略时的「本局信息」和对局中的「本局信息」都会在「本局禁用干员」下写明是哪一名。商店卡片、整备区、作战区和详情卡上都标着「自选」，详情卡显示所选的技能和模组。这同样是**局外设置**：下一局开始时生效，对局中修改只保存到下一局；AI 队友不使用自选；你开启 AI 托管时，AI 会优先买你的自选干员，并在快能刷出自选名额时提早一回合升级调度中心；信标不会把自选干员送给队友。「导出 / 导入」「全部清空」可用，大厅的「干员调配」按钮上会显示「自选 N」。
+- **干员调配（技能、模组、潜能与练度）**：大厅、房间和「确认本局信息」里的「干员调配」按钮打开的界面有三个标签页：「干员调配」「干员持有」「自选编队」。「干员调配」把干员列成一张表，每名干员一行，各列是**技能**、精锐的**模组**和**潜能 · 练度**：直接点行里的技能图标、模组图标（「不装备」在最前，模组只在精锐时生效）或改两个下拉框就设好了；点头像或名字在右边打开详情，依次是潜能与练度、特质、技能（普通 Lv.4 / 精锐 Lv.7 两档数值可切换）、局内数值和模组（窄屏——宽度 1000 像素以下，例如横屏的手机——详情盖住列表，点「干员列表」返回）。**潜能**（1–6）和**练度**（未精英化 / 精英阶段1 / 精英阶段2 / 精英阶段2-60级）按干员设置，同一名干员的普通、精锐和自选共用，默认潜能 6、精英阶段2-60级（维护者 2026-10-08 的决定：默认满潜、满加成）。官方规则：潜能取你自己持有的这名干员的潜能；练度是调度手册「加成情况」里的**自持有加成**——未精英化 +0%，精英1 攻击、防御 +5%，精英2 攻击 +10%、防御和生命 +5%，精英2 60 级三项 +10%，单独乘算，不和其他「攻击力 +x%」相加；未持有的特许干员在官方按潜能 1、没有加成，想照官方打就把它设成潜能 1、未精英化。补位的替补干员和自选的原型干员没有潜能和练度，设置对它们不生效。商店卡片、详情卡和「局内数值」都按你的设置显示；同盟模拟中队友的干员按队友的设置。和技能、模组一样是**局外设置**：「确认本局信息」结束前修改对本局生效，之后只保存到下一局。「恢复默认」「全部恢复默认」同时重置潜能与练度，「导出 / 导入」会带上它们，「干员调配」按钮上的数字也算上改过潜能或练度的干员。
+- **干员持有（补位）**：同一个界面的第二个标签页「**干员持有**」可以把你没有的干员**下掉**（点一下卡片切换「持有 / 未持有」）。和官方一样只有 55 张「普通」棋子能下掉（界面按阶级列出本赛季商店里的 53 名，每张写着由谁替补、用几技能）；74 张「预设」棋子（多是 1–2 阶的常见干员）无论是否持有都由本人上场，不在其中。默认全部持有。下掉的干员和官方一样**显示为替补干员**：商店和晋升奖励的卡片、整备区和临时整备区里的单位、作战区、结算界面的阵容，都是替补干员的头像、名字和模型（例如银灰显示为 Sharp），带一个小小的「替补」标记；详情卡显示替补干员的立绘、名字、技能、天赋和模组，名字下面写着「银灰的替补」。**盟约、特质、阶级、价格、合成仍按原干员**（卡片和详情卡上的盟约、特质就是原干员的），作战时的属性、技能、天赋、模组、攻击范围、近战位 / 远程位都是替补干员的（例如银灰 → Sharp 的 2 技能「亮剑」）。同一名替补干员可能替补好几名干员（Sharp 替补忍冬、银灰、隐德来希等），它们按各自的原干员分别合成：看阶级和盟约区分，点开详情卡能看到替补的是谁。同盟模拟中队友查看你的整备区和作战区、你的盟约成员时，看到的也是替补干员。替补的技能和模组由棋子固定，「干员调配」对它不生效；替补干员没有召唤物（下掉缪尔赛思就没有流形）。这是**局外设置**：下一局开始时生效，对局中修改只保存到下一局；同盟模拟中只影响你自己的棋子，队友的棋子按队友的设置。没有官方的好友借用：想用本人上场，改回「持有」即可。「全部持有」一键恢复，「导出 / 导入」可以把设置带到别的浏览器；大厅的「干员调配」按钮上会显示「替补 N」提醒你有几名干员被下掉。
+- **自选编队**：同一个界面的第三个标签页「**自选编队**」有 4 个名额：5 阶两个、6 阶两个。点名额打开选择列表，每个名额可以放一名**你持有的 6★ 干员**（棋子池里没有的那些，标「已持有」：3 个技能任选其一，精锐的模组任选其一或不装备；只在集成战略或生息演算里生效的模组不能选：ISW-A、SO-A / SO-B、RA-A）或一名**原型干员**（标「原型」：Sharp、Mechanist 等 9 名 6★，5 阶另有 6 名 4★ 预备干员；技能和模组与补位时相同，不能更改）。同一名原型干员可以同时放进一个 5 阶和一个 6 阶名额；你持有的干员只能放一个名额；同一阶的两个名额不能放同一人。列表只列出已经做好战斗逻辑的干员（原型干员全部可选，持有的 6★ 随战斗逻辑完成陆续加入）。对局中，自选干员**只出现在你自己的商店里**：调度中心升到 5 级后才会刷出 5 阶名额的干员，6 级后才会刷出 6 阶的；刷新概率和同阶干员一样，每个名额有自己的库存（5 阶 8 张、6 阶 5 张，不与队友共用），价格与同阶干员相同，晋升奖励里也可能出现；效果、奖励或机变卡「获得一名随机干员」时，也可能抽到你自己的自选干员（按那个效果自己的阶级规则，不看调度中心等级）。三张合成精锐，精锐时所选模组生效（5 阶 1 级、6 阶 3 级）。你持有的干员的潜能与练度在名额卡片上设置（默认潜能 6、精英阶段2-60级，规则同「干员调配」）；原型干员没有潜能和练度。自选干员**没有特质**，盟约按干员的实际阵营分配（推进之王是维多利亚，煌同时是炎和维多利亚，没有对应阵营的是协防干员），和普通棋子一样计入盟约人数；本局禁用了它全部的盟约时，它不会出现在商店里（协防不会被禁用），开局的「确认本局信息」、选策略时的「本局信息」和对局中的「本局信息」都会在「本局禁用干员」下写明是哪一名。商店卡片、整备区、作战区和详情卡上都标着「自选」，详情卡显示所选的技能和模组。这同样是**局外设置**：下一局开始时生效，对局中修改只保存到下一局；AI 队友不使用自选；你开启 AI 托管时，AI 会优先买你的自选干员，并在快能刷出自选名额时提早一回合升级调度中心；信标不会把自选干员送给队友。「导出 / 导入」「全部清空」可用，大厅的「干员调配」按钮上会显示「自选 N」。
 - 每回合商店会自动刷新未冻结的格子。
 - **「本回合首次主动刷新」按每名干员各自计算**（拉普兰德）：她在本回合遇到的第一次主动刷新生效，在整备区里也算——先刷新、再买下拉普兰德，下一次刷新她仍会叠层；本回合新合成的精锐算作一名新的拉普兰德——即使合成用的几名本回合已经生效，精锐在本回合的下一次主动刷新仍会 +8。

@@ -72,16 +76,17 @@
 - 从整备区把干员拖到棋盘：近战位 / 远程位高亮为可放置格。**特性写着「可以放置于远程位」的近战干员（钩索师、推击手：歌蕾蒂娅、崖心、见行者，以及自选编队的温蒂）也能放到高台上**，普通和精锐都可以，装什么模组或者不装模组都一样；其他近战干员只能放在地面。站在高台上时照常攻击，但不阻挡敌人。**水里不能部署**：战场#08（涨潮控制）中间的那条水池（深水区）和官方一样拒绝部署，任何干员和召唤物都放不上去；突袭盟约的干员再部署、战术家的援军自动选战术点时也只会落在岸上。松手后出现**方向轮盘**，向上 / 右 / 下 / 左滑动，棋盘上会用橙色条纹预览旋转后的攻击范围，松手确认；松在中心或点「✕ 点击取消」取消。
 - 想只改朝向：把干员拖回它自己所在的格子再选方向。朝向会一直保留到下一回合。
 - **点选按地上的方格**：点哪个方格就选中站在那个方格上的单位，点空方格不选任何单位（点到后一格干员画出来的头部，算作点了后面那一格）。作战中友方同样按方格；敌人在格子之间走动，点它的身体或脚下附近（0.6 格内）即可，飞行敌人点它的身体。
- 拖动时模型一直在指针 / 手指下（脚在指针下方约半格，鼠标和触屏一样），**指针所在的格子**就是落点并高亮；装备拖到哪个格子，就配发给那个格子上的干员；在商店栏上松手则放回原位。
- 点击干员显示攻击范围和底部按钮（「撤退」回到整备区、「出售 +1」）；右键 / 长按查看详细信息：头像旁是盟约，下面第一项就是干员的**特质**（如「休整期结束时 …」），之后是特性、属性、攻击范围、技能、模组、装备、天赋。属性是**实时数值**：作战中每秒刷新 4 次（当前生命、攻击、防御、法抗、攻击间隔、阻挡），休整期显示棋盘上的干员**开战时**的数值（已算入装备、盟约层数、特质、策略与机变效果）；高于基础值为绿色、低于为红色并标出差值（攻击间隔变短算增益），鼠标悬停可看基础值。基础值是官方数据在该阶级的等级、技能等级（和精锐的模组）下、**满潜能**、0 信赖的数值：所有干员（精锐、补位、自选也一样）都带上全部潜能加成——部署费用、再部署时间、攻击力 / 生命 / 防御 / 法抗 / 攻速，以及「天赋效果增强」，召唤物的天赋也按召唤者的满潜能（维护者 2026-10-07 的决定，GitHub #252：官方赛事视频里的刺玫正是满潜能的攻击 435、费用 15）。
+- 拖动时**指针 / 手指所在的格子**就是落点并高亮。这一格可以放下（棋盘或整备区）时，干员模型直接站在这一格上，和官方部署一样：手指在格子里移动时模型不动，移到下一个可放下的格子就站到那一格；指针不在可放下的格子上时（例如在商店栏上），模型跟在指针 / 手指下（脚在指针下方约半格）。鼠标和触屏一样。装备拖到哪个格子，就配发给那个格子上的干员；在商店栏上松手则放回原位。
+- 点击干员显示攻击范围和底部按钮（「撤退」回到整备区、「出售 +1」）；右键 / 长按查看详细信息。作战中、查看队友的阵地时，点开场上我方干员或召唤物（包括队友的）的详情卡，棋盘上也会画出它**当前的攻击范围**（按它现在的位置、朝向和技能，例如开着扩大攻击范围的技能时就是扩大后的范围），关掉详情卡、拖动或选方向时消失；敌人、整备区和临时整备区里的单位不画。详情卡里：头像旁是盟约，下面第一项就是干员的**特质**（如「休整期结束时 …」），之后是特性、属性、攻击范围、技能、模组、装备、天赋。属性是**实时数值**：作战中每秒刷新 4 次（当前生命、攻击、防御、法抗、攻击间隔、阻挡），休整期显示棋盘上的干员**开战时**的数值（已算入装备、盟约层数、特质、策略与机变效果）；高于基础值为绿色、低于为红色并标出差值（攻击间隔变短算增益），鼠标悬停可看基础值。基础值是官方数据在该阶级的等级、技能等级（和精锐的模组）下、按你在「干员调配」里给这名干员设的**潜能和练度**（默认潜能 6、精英阶段2-60级）、0 信赖的数值：潜能加成包括部署费用、再部署时间、攻击力 / 生命 / 防御 / 法抗 / 攻速，以及「天赋效果增强」，召唤物的天赋按召唤者的潜能；练度把攻击、防御、生命上限单独乘上自持有加成（精英阶段2-60级 ×1.1；召唤物不乘）。补位的替补干员和原型干员没有潜能档，也没有自持有加成（0.2.2；0.2.1 是所有干员一律满潜能、没有自持有加成，GitHub #252）。
 - **召唤物手动摆放**：把带可部署召唤物的干员放上棋盘后，它的召唤物会进入整备区（赫默的医疗探机、巫恋的诅咒娃娃、凯瑟琳的 2 个支援装置、浊心斯卡蒂的海嗣、伺夜的狼群、缪尔赛思的流形；同一干员的召唤物叠在一格，不占部署位），像干员一样拖到棋盘上并选方向。凯瑟琳的支援装置给**它面朝的那一格**上的干员加屏障，所以要放在干员旁边、朝向干员。**战术家的援军**（伺夜的狼群、缪尔赛思的流形，也就是选战术点）和官方一样**只能放在召唤者的攻击范围内**：拖动时只有这些格子会亮起；之后在原地给战术家换朝向时，不在新攻击范围里的援军会退回整备区（还在范围里的保持不动；整备区和临时整备区都满、援军无处可退时，这次换朝向不生效并提示「整备区已满」）。撤退或出售干员时它的召唤物一起移除；**移动干员到别的格子时（包括把召唤物拖到干员身上交换位置），它已摆放的召唤物会退回整备区**，需要重新摆放。整备区满时退回的召唤物进入临时整备区，休整期结束还没处理的会在下一回合重新回到整备区。
- 作战中：海嗣、狼群、流形、支援装置开战时就在摆放的位置部署；医疗探机、诅咒娃娃开战时也在摆放的位置免费部署一次（持续 10 / 15 秒，和官方一样不需要先开技能），之后所属干员每次发动技能时再次在原位出现；所属干员被击倒、再部署回场时，技能给的那一个如果还没出现，会立刻出现。巫恋退场时诅咒娃娃随之撤退，赫默退场时已出现的医疗探机不受影响。医疗探机、诅咒娃娃、海嗣（以及自选干员的工程蓄水炮、沙地兽、打字机、本能的召唤）这类不会受到攻击、到时间就离场的召唤物不受任何伤害（活性源石也一样）、也不会被治疗，它们的血条就是倒计时：随时间减少，空了就离场。医疗探机、诅咒娃娃、支援装置、海嗣没有摆放就不会出现；狼群、流形（战术家的援军）没有摆放时会在战术点出现。狼群最后一只狼影被击倒后进入战术点形态：暂时离开战斗（不阻挡、不攻击、不会被攻击，格子仍然占着），经过狼影的恢复时间（25 秒）后在原地满血、带着 1 只狼影回来，之后重新开始每 25 秒加 1 只狼影；伺夜退场时狼群随之撤退，伺夜再部署回场时狼群重新带着 2 只狼影出现。
- **阻挡**看接触：干员阻挡中心距离约 0.7 格以内、与它接触的敌人（地面干员挡不住飞行和近地悬浮的敌人）。敌人卡在两名干员之间时，阻挡它的干员倒下、撤退或被晕眩后，另一名仍与它接触、还有空余阻挡数的干员会**立刻接替阻挡**；都没有空位时敌人才会走掉。被冻结或晕眩的敌人同样会被接触到的干员挡住：干员再部署到它身边就会阻挡它，隐匿的敌人也随之破隐（沉睡、浮空、恐惧的敌人仍然挡不住）。迎面走来的敌人会停在干员格子前面。**阻挡了就一定能打到**：任何干员（包括站在近战位的远程干员和召唤物）总能攻击自己阻挡的敌人，不管它在不在攻击范围内、在干员哪一侧，并且优先攻击它（医疗干员除外：咒愈师以外的医疗挡住敌人时照常治疗队友，不会攻击它；「停止攻击」类技能期间也除外；技能原文另有限制的也按原文：普罗旺斯二技能开启时不以生命值高于 80% 的敌人为目标，寒檀二技能的冰凌随机落在攻击范围内的地块上）。
+- 作战中：海嗣、狼群、流形、支援装置开战时就在摆放的位置部署；医疗探机、诅咒娃娃开战时也在摆放的位置免费部署一次（持续 10 / 15 秒，和官方一样不需要先开技能），之后所属干员每次发动技能时再次在原位出现；所属干员被击倒、再部署回场时，技能给的那一个如果还没出现，会立刻出现。巫恋退场时诅咒娃娃随之撤退，赫默退场时已出现的医疗探机不受影响。医疗探机、诅咒娃娃、海嗣（以及自选干员的工程蓄水炮、沙地兽、打字机、本能的召唤）这类不会受到攻击、到时间就离场的召唤物不受任何伤害（活性源石也一样）、也不会被治疗，它们的血条就是倒计时：随时间减少，空了就离场。琳琅诗怀雅二技能的香槟炸弹也不会掉血（活性源石和各种伤害都不会）、不会被治疗，不会因为生命值而消失，只在「见面礼」炸到碰到它的第一个地面敌人后消失（维护者 2026-10-08 的决定）；隐匿（没有被反隐、也没有被阻挡）或沉睡的敌人选不中，不会触发它。医疗探机、诅咒娃娃、支援装置、海嗣没有摆放就不会出现；狼群、流形（战术家的援军）没有摆放时会在战术点出现。狼群最后一只狼影被击倒后进入战术点形态：暂时离开战斗（不阻挡、不攻击、不会被攻击，格子仍然占着），经过狼影的恢复时间（25 秒）后在原地满血、带着 1 只狼影回来，之后重新开始每 25 秒加 1 只狼影；伺夜退场时狼群随之撤退，伺夜再部署回场时狼群重新带着 2 只狼影出现。
+- **阻挡**看接触：干员阻挡中心距离约 0.7 格以内、与它接触的敌人（地面干员挡不住飞行和近地悬浮的敌人）。敌人卡在两名干员之间时，阻挡它的干员倒下、撤退或被晕眩后，另一名仍与它接触、还有空余阻挡数的干员会**立刻接替阻挡**；都没有空位时敌人才会走掉。被冻结或晕眩的敌人同样会被接触到的干员挡住：干员再部署到它身边就会阻挡它，隐匿的敌人也随之破隐（沉睡、浮空、恐惧的敌人仍然挡不住）。迎面走来的敌人会停在干员格子前面。**阻挡了就一定能打到**：任何干员（包括站在近战位的远程干员和召唤物）总能攻击自己阻挡的敌人，不管它在不在攻击范围内、在干员哪一侧，并且优先攻击它——只有速射手例外：攻击范围内有飞行敌人时先打飞行敌人（特性「优先攻击空中单位」，维护者 2026-10-08 的决定，GitHub #220）（医疗干员除外：咒愈师以外的医疗挡住敌人时照常治疗队友，不会攻击它；「停止攻击」类技能期间也除外；技能原文另有限制的也按原文：普罗旺斯二技能开启时不以生命值高于 80% 的敌人为目标，寒檀二技能的冰凌随机落在攻击范围内的地块上）。
 - 开战时干员按**逐列从左到右、每列从上到下**的顺序部署（最左一列最先，与阿戈尔吞噬「更靠左和靠上」的顺序相同），召唤物在所有干员之后。被阻挡的敌人先打阻挡它的干员；没被阻挡的远程敌人先打嘲讽等级高的，同级时打最后部署的（召唤物、最右一列靠下的干员和刚再部署的干员更容易挨打）。隐匿的干员只会被它阻挡的敌人攻击，没被它挡住的敌人的溅射、爆炸、范围技能、留下的伤害区域（包括假想敌：蚀裂被击倒后喷出的毒雾）以及深池伙友卫队、扎罗的减攻速光环也落不到它身上（和官方一样；写明「无视无法选择」的效果如污染秽蚀照常能打到，寒霜的减攻速光环和“墓碑”这类地图效果照常生效）；**迷彩**（如忍冬三技能期间击倒敌人后、寒芒克洛丝一技能）同样只会被它阻挡的敌人攻击，也显示为半透明，但躲不开溅射、爆炸和范围伤害（同时打范围内所有单位的普通攻击——如“斩胄之剑”“破胄之锤”“灵幛”——打不到它），也不算隐匿（不触发叙拉古等「隐匿时」效果）。叙拉古的隐匿按时（32 秒 + 每层 0.4 秒）结束；忍冬的迷彩按官方持续到她下一次开技能。**蒂比起飞**期间只挡飞行敌人，地面敌人不会选她为目标（普通攻击、溅射、技能范围、燃烧区域和要选中目标的减益都落不到她身上）；飞行和悬浮敌人、写明「无视无法选择」的效果（如污染秽蚀）照常能打到她；地面敌人的光环和全场效果（如“墓碑”让全场攻击力、防御力减半）、她攻击敌人时招来的反制（如重装侦察兵的暴露）照常生效，起飞前已挂上的持续伤害也照常跳。
- **技能自动释放**按官方「技能策略」：一般的手动技能在技能就绪、即将攻击 / 治疗且初始攻击范围内有敌人（治疗技能：有受伤的队友）时释放；**带独立技能范围**的技能（如德克萨斯「剑雨」、凛御银灰「御敌的锋锐」、锏「归于宁静」，不含「攻击范围扩大」类）在**技能范围内出现敌人**时就释放，不必先攻击；**开启后攻击范围比平时大**的手动技能（如莫斯提马「序时之匙」、银灰「真银斩」、史尔特尔「黄昏」、白面鸮「脑啡肽」）在**开启后的攻击范围内出现敌人**（治疗技能：有受伤的队友）时就释放，不必等敌人走进平时的攻击范围，这是本作按玩家反馈有意与官方不同的地方；**重装干员**的手动技能在受到伤害时释放（例外：深巡、雷蛇的二技能，号角的二、三技能，灰毫的一、二技能按玩家反馈改为攻击范围内有敌人时就释放（深巡二技能按开启后更远的攻击范围判断），余的二技能「厚礼上宾」改为周围两格（技能范围）内有敌人时就释放，这是本作有意与官方不同的地方；深巡一技能仍要受到伤害）；执旗手 / 战术家 / 吟游者技能就绪即释放，解放者 / 阵法术师在初始攻击范围内有敌人时释放（开启后攻击范围扩大的技能——薄绿「风语」、蜜蜡「沙暴扩散」、卡涅利安「食噬之印」、玛恩纳「未宽解的悲哀」，以及安洁莉娜「秘杖·反重力模式」——同样按开启后的攻击范围判断）。**自动触发**的技能按其本身规则：只作用于队友的（如引星棘刺「度算浪波」、浊心斯卡蒂「同葬无光之愿」、纯烬艾雅法拉「无声润物」）技力满就释放，不用等敌人或伤员；伺夜「领袖的呼唤」「领袖的馈赠」只在场上有狼群时释放：狼群处于战术点形态时，「领袖的呼唤」也会释放并让它立刻带着 1 只狼影回到场上（狼影未满时改为加 1 只狼影，已满时把狼群的生命值回满），「领袖的馈赠」也会释放（立即获得费用，强化留给回来的狼群的下一次攻击）；又如古米「备用军粮」在身边（技能范围内）有受伤的队友时就会发动，下一次攻击改为治疗该队友（不用先接敌；她的二技能「食粮烹制」是手动技能，按重装规则在受到伤害时释放）；塞雷娅「急救」在她即将攻击时、身边有生命不高于一半的队友才发动，这次攻击直接改为治疗；新约能天使二技能的护盾按官方给攻击范围内**仇恨最高**的队友（嘲讽等级高的优先，同级时给最后部署的），不一定是站在最前面挨打的那个。手动技能的自动释放有 3 秒冷却：开战部署后先等 3 秒，之后同一干员两次自动释放至少间隔 3 秒（可充能的技能不会同时放完）。
- **推开 / 拉拽**按官方「力度 − 重量」决定距离：例如野鬃二技能（中等力度）把重量 0 / 1 / 2 / 3 的敌人推开约 2.1 / 1.7 / 0.4 / 0.1 格，重量 4 以上推不动——轻型敌人反而推得更远。野鬃二技能沿她的朝向推开：正前方 45° 以内按上面的距离，身后、侧面或贴身的敌人只被轻推一下（约 0.4 格以内）；琳琅诗怀雅三技能的金币则从她身边向外推开，落在她前方 2–4 格的范围（官方「范围2-4」）内的地面敌人和她阻挡的敌人身上。攻击速度最低为 20（沼泽、寒霜等减速叠满时也不会更慢）。
- **生命回复不是治疗**（与官方一致）：吟游者的特性（魔王、浊心斯卡蒂及其海嗣）、调香师的天赋、瑕光二技能、铃兰三技能、锡人炼金单元给友军的「每秒恢复生命」是提高对方的**生命回复速度**，不是治疗：收割者 / 不屈者 / 武者（如隐德来希、圣约送葬人、折桠）这类「无法被友方治疗」的干员和处于禁疗的干员也能吃到，治疗量加成（如「受到的治疗效果提升」）也不会放大它；铃兰三技能开启后的第一秒没有回复。直接治疗（医疗干员的治疗、技能的「恢复…生命」）仍然对这些干员无效。
+- **技能自动释放**按官方「技能策略」：一般的手动技能在技能就绪、即将攻击 / 治疗且初始攻击范围内有敌人（治疗技能：有受伤的队友）时释放；**带独立技能范围**的技能（如德克萨斯「剑雨」、凛御银灰「御敌的锋锐」、锏「归于宁静」，不含「攻击范围扩大」类）在**技能范围内出现敌人**时就释放，不必先攻击；**开启后攻击范围比平时大**的手动技能（如莫斯提马「序时之匙」、银灰「真银斩」、史尔特尔「黄昏」、白面鸮「脑啡肽」）在**开启后的攻击范围内出现敌人**（治疗技能：有受伤的队友）时就释放，不必等敌人走进平时的攻击范围，这是本作按玩家反馈有意与官方不同的地方；**重装干员**的手动技能在受到伤害时释放（例外：深巡、雷蛇的二技能，号角的二、三技能，灰毫的一、二技能按玩家反馈改为攻击范围内有敌人时就释放（深巡二技能按开启后更远的攻击范围判断），余的二技能「厚礼上宾」改为周围两格（技能范围）内有敌人时就释放，这是本作有意与官方不同的地方；深巡一技能仍要受到伤害）；执旗手 / 战术家 / 吟游者技能就绪即释放，解放者 / 阵法术师在初始攻击范围内有敌人时释放（开启后攻击范围扩大的技能——薄绿「风语」、蜜蜡「沙暴扩散」、卡涅利安「食噬之印」、玛恩纳「未宽解的悲哀」，以及安洁莉娜「秘杖·反重力模式」——同样按开启后的攻击范围判断）。**自动触发**的技能按其本身规则：只作用于队友的（如引星棘刺「度算浪波」、浊心斯卡蒂「同葬无光之愿」、纯烬艾雅法拉「无声润物」）技力满就释放，不用等敌人或伤员；立即生效、本身不是一次攻击的技能（如度算浪波、先锋的回费技能）两次释放之间至少隔一个攻击间隔（释放本身不占用普通攻击），攻速越快放得越勤，迅捷返还的技力再多也不会每一帧都放（GitHub #298）；伺夜「领袖的呼唤」「领袖的馈赠」只在场上有狼群时释放：狼群处于战术点形态时，「领袖的呼唤」也会释放并让它立刻带着 1 只狼影回到场上（狼影未满时改为加 1 只狼影，已满时把狼群的生命值回满），「领袖的馈赠」也会释放（立即获得费用，强化留给回来的狼群的下一次攻击）；又如古米「备用军粮」在身边（技能范围内）有受伤的队友时就会发动，下一次攻击改为治疗该队友（不用先接敌；她的二技能「食粮烹制」是手动技能，按重装规则在受到伤害时释放）；塞雷娅「急救」在她即将攻击时、身边有生命不高于一半的队友才发动，这次攻击直接改为治疗；新约能天使二技能的护盾按官方给攻击范围内**仇恨最高**的队友（嘲讽等级高的优先，同级时给最后部署的），不一定是站在最前面挨打的那个。手动技能的自动释放有 3 秒冷却：开战部署后先等 3 秒，之后同一干员两次自动释放至少间隔 3 秒（可充能的技能不会同时放完）。白铁的「铁钳号·原型机」能被我方干员攻击（不受伤害）：它在攻击范围 / 技能范围内时，能作用到它的技能会像遇到敌人一样自动开启并消耗弹药（项目维护者 2026-10-08 按官方体验确认；白铁™多功能平台不会被攻击）；技能效果只会挑选敌人、打不到它的技能（如德克萨斯「剑雨」）不会为它开启，留给真正的敌人。
+- **推开 / 拉拽**按官方「力度 − 重量」决定距离：例如野鬃二技能（中等力度）把重量 0 / 1 / 2 / 3 的敌人推开约 2.1 / 1.7 / 0.4 / 0.1 格，重量 4 以上推不动——轻型敌人反而推得更远。野鬃二技能沿她的朝向推开：正前方 45° 以内按上面的距离，身后、侧面或贴身的敌人只被轻推一下（约 0.4 格以内）；琳琅诗怀雅三技能的金币则从她身边向外推开，落在她前方 2–4 格的范围（官方「范围2-4」）内的地面敌人和她阻挡的敌人身上。被推开、拉动的敌人会进入官方的**失衡**状态，这段时间里不走路、也不发动普通攻击：推开按力度减重量的档位持续 0.2–约 1.17 秒（中等力度推重量 1 的敌人为 0.8 秒），拉拽持续 1 秒（力度比重量低 2 档时 0.5 秒，拉到身前后也一样），失衡期间也不会放技能（例如弑君者的闪现）或执行脚本移动（例如枯朽之种俯冲），弧光锋卫则在失衡期间每 0.066 秒受到 400 点真实伤害；推不动的空中单位（静态刚体）被推时停顿 0.1 秒、被拉时停顿整个拉力作用时间；浮空、重生和路线上的瞬移出现都会立刻解除失衡。画面上的滑行也用同样的时间（默认 2 倍速下现实时间减半）。攻击速度最低为 20（沼泽、寒霜等减速叠满时也不会更慢）。
+- **生命回复不是治疗**（与官方一致）：吟游者的特性（魔王、浊心斯卡蒂及其海嗣）、调香师的天赋、瑕光二技能、铃兰三技能、锡人和引星棘刺（一、二技能）的炼金单元给友军的「每秒恢复生命」是提高对方的**生命回复速度**，不是治疗：收割者 / 不屈者 / 武者（如隐德来希、圣约送葬人、折桠）这类「无法被友方治疗」的干员和处于禁疗的干员也能吃到，治疗量加成（如「受到的治疗效果提升」）也不会放大它；铃兰三技能开启后的第一秒没有回复。直接治疗（医疗干员的治疗、技能的「恢复…生命」）仍然对这些干员无效。
 - **元素损伤条**在单位自己的血条正下方：左边是元素图标，白条是剩余的元素值，只有真正爆发时才会耗尽；爆发后条变成元素的颜色、图标闪烁，在冷却期间慢慢回满（这段时间不会再积累元素损伤）。
+- **弹药、狼影和「我执」**也画在血条上（0.2.2）：弹药类技能（如隐现的二技能，14 发）开启时，血条下方的技力条换成**分格的黄色弹药条**——一格一发，打一发少一格，补弹时回来；这类技能开启期间脚下不再有持续的金色光环。伺夜的狼群在血条下方有一排**菱形**，亮的是剩下的狼影，暗的是已用掉的空位（最多 3 个），进入战术点形态时消失。斩业星熊进入「我执」后，绿色血条停在 1 点，超出的伤害涨成血条上的**红条**，红条满了才会倒下。这些只是显示，不改变任何战斗数值。
 - **敌人**：远程敌人的射程按官方从干员的碰撞体积（半径 0.25 格）算起，比数据上的攻击范围半径远 0.25 格——例如萨卡兹枯朽战车（射程 2.2）能打到中心 2.45 格以内、站在低地的干员。没被阻挡的远程敌人每次攻击都会站定、播完攻击动作，两次攻击之间才继续前进。它的第一下攻击就是「秽蚀轰击」（在目标处留下半径 1.7、持续 10 秒的污染区：低地每秒 50、高台 25 点真实伤害，同名不叠加），被阻挡时普通攻击伤害翻倍。“萨科塔之翼”等血量第一次低于一半时会在原地小范围乱飞 5 秒（官方叫「恐惧」）；叙拉古、妮芙等干员造成的恐惧会让敌人朝远离干员的方向逃跑。隐匿的敌人只有被阻挡（或被揭示）时才能被选为攻击目标，溅射、技能的圆形范围伤害也打不到没被挡住的隐匿敌人（和官方一样）；阻挡解除后 3 秒内它仍然可以被选中（官方页面注明 0 秒或 1 秒的敌人按页面），之后才重新隐匿：深池逐火战士每次被击倒后先重生 1 秒，然后变成继续往前走的隐匿「余烬」（需要 5 次伤害击倒，护卫的「火灰」10 次），一般要用干员挡住它才能选中（挡住时它不再半透明；被阻挡时被击倒的，重生结束后约 2 秒内也能用远程或范围伤害补刀），10 秒内没打掉就满血站起来；敌人重生时身上的减速、恐惧等状态会被清空、技能冷却从头开始，被恐惧的逐火战士变成余烬后也能被挡住；转译基底·α 原始形态受到的伤害全部无效，受到第 4 次物理伤害、第 4 次法术伤害或被阻挡时花 2 秒变成寻仇者、特战术师或幽灵，之后才能被击倒。
 - **复活**（策略「埃芒加德」的前 3 次、M3茧甲）与官方一致：算作一次被击倒，紧接着在倒下的位置免费立即再部署——干员「被击倒时」和「部署时」的效果都会触发（例如砾的不屈 / 卡西米尔层数、瑕光的层数、卡西米尔盟约的部署攻击力），技力和身上的增益像普通再部署一样重置。
 - **策略「鸭爵」（“神秘顾客”）**：从第 5 回合起，你和队友的每场作战（**最终攻势和隐秘核心也算**，联防不另外替换）有 0–2 名地面敌人（不含领袖及其部件、悬赏敌人）被替换为特殊版本的鸭爵、高普尼克、流泪小子或圆仔，出现在被替换敌人的时间和路线上（两人同场的领袖战里，各自只替换走向自己保护目标点的敌人）；击倒者获得 1 资金（联防中的队友也算），漏进保护目标点照常扣 1 点目标生命值。鸭爵和圆仔**不攻击、无法被阻挡**：鸭爵受到伤害后移动速度 +300%；圆仔始终面向我方干员较多的一侧，来自正面的物理 / 法术伤害 −80%。流泪小子隐匿（被阻挡时才能被选中），攻击使目标晕眩 3.5 秒，被晕眩的干员挡不住它（官方规则）；它被放开后要过 1 秒才重新隐匿（官方是 0 秒，本模式有意放宽，维护者 2026-10-07 的决定）；高普尼克免疫晕眩、冻结和浮空。
@@ -112,8 +117,8 @@
 - **附加盟约**（精准、迅捷、灵巧、奥术、坚守、助力、远见、奇迹、投资人、突袭、不屈、调和、协防干员、独行、绝技）：多为 2 / 3 名激活；「独行」需要场上**恰好 1 名**，「绝技」按场上的精锐数量计算，没有势力盟约的干员算「协防干员」。
 - **本局禁用**：「标准模拟」与官方相同，关闭拉特兰、阿戈尔、卡西米尔、灵巧、奥术、奇迹、投资人、突袭、独行、绝技这 10 个盟约，它们不会激活。你有这些盟约的成员时（奇迹、投资人也算整备区里的成员），盟约栏末尾显示**灰色带 ✕ 的盟约图标和「本局禁用」**，干员卡、盟约标签和盟约详情也这样标注。「远见」「奇迹」「投资人」计入整备区，盟约详情的「成员」人数也一样计入整备区里的成员（临时整备区不算）。
 - **调和**：激活时，场上已有成员的核心盟约人数各 +1（只加 1 人，不论有几名调和干员）。盟约详情会写「含调和 +1」，成员列表上方列出带来这 1 人的调和干员（如缪尔赛思）。
- **炎佑**：场上 6 名以上「炎」干员时召唤炎佑（9 名时两只）。它追踪全场仇恨最高的敌人并悬停在其上方，技能「祛恶之焰」锁定 1 个目标**持续喷火**（最长 20 秒，每秒对目标周围的敌人造成法术伤害并附带灼燃，喷火时不普攻，目标死亡或被沉默时结束）；场上暂时没有敌人时停在原地等待。它持有官方的「孤立」：医疗干员不会治疗它，友方的增益、光环和天赋也不会选中它（敌人照常可以攻击它）。
- **层数**：盟约通过各种效果叠加层数，层数越高效果越强（例如「炎」攻击力随层数提升、「远见」每 10 层给资金）。**层数整局保留，但只有盟约激活时才生效**，多数情况下也只有激活的盟约能叠层。每个盟约最多 **999 层**（与官方相同，取自官方客户端的常量）：到 999 后再获得的层数直接作废，按层数发放的奖励（如「远见」的资金）也随之停止。作战中获得的层数在盟约栏里**实时**显示，作战结束后一直保留到结算（联防期间也不会退回去）。「调和」「协防干员」「独行」「绝技」与官方一样**不显示层数**（盟约栏、详情和结算卡片都不显示），但叠层的特质 / 策略 / 装备照样给它们加层（PRTS 盟约记录：「部分盟约不会显示叠加层数，但是叠加层数的特质/策略/装备等效果仍然对其生效」），这些层数也计入激活层数总和；它们的效果本身与层数无关。
+- **炎佑**：场上 6 名以上「炎」干员时召唤炎佑（9 名时两只）。它追踪全场仇恨最高的敌人并悬停在其上方，技能「祛恶之焰」锁定 1 个目标**持续喷火**（最长 20 秒，每秒对目标周围的敌人造成法术伤害并附带灼燃，喷火时不普攻，目标死亡或被沉默时结束）；场上暂时没有敌人时停在原地等待。它持有官方的「孤立」：医疗干员不会治疗它，友方的增益、光环和天赋也不会选中它（敌人照常可以攻击它）。它是飞行单位：标注「不可对空」的敌人效果（高能源石虫 / 冰爆源石虫 / 卷心籽的死亡爆炸、水遁忍者的漩涡、鼎沸、“萨科塔昂首”的祈祷邀约、“斩胄之剑” / “破胄之锤”的攻击等）不会波及它。
+- **层数**：盟约通过各种效果叠加层数，层数越高效果越强（例如「炎」攻击力随层数提升、「远见」每 10 层给资金）。**层数整局保留，但只有盟约激活时才生效**，多数情况下也只有激活的盟约能叠层。每个盟约最多 **999 层**（与官方相同，取自官方客户端的常量）：到 999 后再获得的层数直接作废，按层数发放的奖励（如「远见」的资金）也随之停止。作战中获得的层数在盟约栏里**实时**显示，作战结束后一直保留到结算（联防期间也不会退回去）。「调和」「协防干员」「独行」「绝技」与官方一样**不显示层数**（盟约栏、详情和结算卡片都不显示），但叠层的特质 / 策略 / 装备照样给它们加层（PRTS 盟约记录：「部分盟约不会显示叠加层数，但是叠加层数的特质/策略/装备等效果仍然对其生效」），这些层数也计入激活层数总和——只有策略「余」的【文火慢炖】不给这四个盟约加层（第 8 回合只激活了其中一个时什么也不加；维护者 2026-10-08 的决定）；它们的效果本身与层数无关。
 - **特质在哪里生效**：「进入休整期时」「休整期结束时」「刷新时」的特质（详情卡的第一项）只在干员部署在作战区时生效；文字里写着「此干员在整备区时也有效」的拉普兰德、调香师、波登可，以及激活【炎】/【投资人】时的琳琅诗怀雅，在整备区里也生效（临时整备区不算），与官方数据一致；「获得时」「售出时」的特质在获得、售出的那一刻生效，不论干员在哪里。「远见」「奇迹」「投资人」的「整备区的…干员也可用于激活盟约」只是把整备区的成员算进人数，不会让赫默等干员的特质在整备区生效。写着「使已激活的…」的特质要求盟约当时已经激活：例如「叙拉古」只数作战区里不同的叙拉古干员，整备区里的拉普兰德不算在这 3 名里；拉普兰德每回合只在她经历的第一次主动刷新时叠层（免费刷新也算），那次刷新时叙拉古没有激活，这一回合就不会再叠。
 - **加成怎么叠**：盟约、策略、装备、机变卡和特质给的「攻击力 / 防御力 / 生命值 +X%」与技能的「攻击力 +X%」**相加**后再乘到属性上（官方：直接乘算），例如精准 +130%、独行 +60%、一件 +40% 的装备合计 +230%，而不是 ×2.3×1.6×1.4。「伤害提升至 X%」「受到的伤害 +X%」（维多利亚、谢拉格、奥术、脆弱等）这类伤害倍率之间仍然相乘；但同名效果只算一个——同一战场两名玩家都开了「奥术」时，敌人身上只挂一层奥术易伤，取两人中较高的那一个（不再相乘）；两名「灵知」的坚冰、两名「莱恩哈特」/「缄默德克萨斯」的法抗削减也一样只算最强的一个。「坚守」的反伤计入被打的那名成员（个人对领袖伤害统计里能看到）。
 - 每局会随机**禁用**若干盟约（标准模拟较少），它们的干员不会出现在商店；开局的「确认本局信息」会列出，选择策略时（「查看禁用盟约与干员」）和对局中（左上角 🔍）也能再查看。
@@ -134,8 +139,8 @@
 - 左下角「交流」打开表情面板：6 套主题 × 6 个表情，左右滑动（或方向键）换主题。表情以气泡形式显示在你头像旁 3 秒，发送冷却 1 秒。
 - 休整期、或自己的作战结束后（显示「作战结束，等待队友完成作战」），点左侧队友头像 →「前往查看」观看他的战场；自己的作战进行中不能观战。联防和最终攻势中可以用 ‹ › 切换左半 / 全景 / 右半。
 - 被淘汰后可以自由观看任何战场，而且**不用每回合重选**：你上次点「前往查看」的是谁，之后每个阶段（回合开始、机变、休整期、作战、最终攻势）都会自动跟到他的阵地或战场；没选过或他也被淘汰时，跟随座位顺序第一位还在场的博士（包括 AI 队友）。阶段中途换人立即生效，之后就跟随新选的人。
- **观战席**（本作新增，原版没有）：每个同盟房间另有 2 个观战席。在大厅输入同盟密钥后点「观战」进入（模拟进行中也可以），不占博士席位、不算参战人数，也不能购买、部署、准备或发送表情；观看方式与被淘汰的玩家相同——自动跟随你上次查看的博士（没查看过则是座位顺序第一位还在场的博士）：休整期显示他的阵地，作战时观看他的战场，点左侧成员头像可切换查看任何人。断线重连会回到观战席；创建者可以在等待室用观战者名字旁的 ✕ 将其移出；观战者在等待室有空位时可以点「入座」成为博士。
- **观看谁，盟约栏就显示谁的盟约**：查看队友的阵地或战场时（休整期、作战结束后、被淘汰后自动观战），左上的盟约栏换成**该队友的盟约、人数和层数**，前面有琥珀色的「👁 队友名」标签，点开的盟约详情也写着「队友名 的盟约」（成员一栏只标出他在场上的干员，整备区看不到）；作战中层数实时增长（与他自己看到的一致）。联防和最终攻势里跟随 ‹ ›：镜头在队友的半场时显示他的盟约，在你自己的半场时显示你自己的；「全景」时，你在这个战场上作战就显示你自己的，你不在这个战场上（漏怪后观看联防、被淘汰后观战）则显示你点「前往查看」选中的那名队友，没选过就显示该战场的第一名玩家（1 号联防者 / 这一组的第一位），**不会显示你自己的**。在这样的战场上点某名队友的头像 →「前往查看」，镜头会直接移到他的半场，‹ › 上的「👁 队友名」与盟约栏的标签一致；‹ › 旁边有「返回战场」（被淘汰时没有）。盟约栏换人时，从盟约栏打开的盟约详情会自动关闭。点开场上某个干员的详情卡时，卡上的盟约（以及点它打开的盟约详情）属于**这个干员的主人**，与盟约栏当前显示谁无关——例如在「全景」点队友的干员，看到的是「队友名 的盟约」及他的人数和层数；在队友的半场点你自己的干员，看到的是你自己的。回到自己的阵地 / 战场后恢复成你自己的盟约。作战结束后观看队友时刷新页面 / 断线重连，仍会回到正在观看的队友战场，「👁 队友名」和「返回战场」都还在。已被淘汰的队友不再显示盟约（也无法被查看）。
+- **观战席**（本作新增，原版没有）：每个同盟房间另有 2 个观战席。在大厅输入同盟密钥后点「观战」进入（模拟进行中也可以），不占博士席位、不算参战人数，也不能购买、部署、准备或发送表情；观看方式与被淘汰的玩家相同——自动跟随你上次查看的博士（没查看过则是座位顺序第一位还在场的博士）：休整期显示他的阵地，作战时观看他的战场，点左侧成员头像可切换查看任何人。断线重连会回到观战席；创建者可以在等待室用观战者名字旁的 ✕ 将其移出；模拟开始后，顶栏延迟旁会出现一个「👁 人数」的小胶囊（没有观战者时不显示），点开是观战者名单（离线的有标记），创建者可以在名单里用 ✕ 把观战者移出，被移出的人立刻回到大厅、不再收到战场画面（观战席上限仍是 2 个，不能调整）；观战者在等待室有空位时可以点「入座」成为博士。
+- **观看谁，盟约栏就显示谁的盟约**：查看队友的阵地或战场时（休整期、作战结束后、被淘汰后自动观战），左上的盟约栏换成**该队友的盟约、人数和层数**，前面有琥珀色的「👁 队友名」标签，点开的盟约详情也写着「队友名 的盟约」（成员一栏标出他在场上的干员；休整期查看时他整备区、临时整备区里的干员也会列出、标为「整备区」，和他自己看到的一样计数：投资人、远见、奇迹这类含整备区的盟约把整备区的算进成员人数，临时整备区的哪个盟约都不算；作战中只列出场上的）；作战中层数实时增长（与他自己看到的一致）。联防和最终攻势里跟随 ‹ ›：镜头在队友的半场时显示他的盟约，在你自己的半场时显示你自己的；「全景」时，你在这个战场上作战就显示你自己的，你不在这个战场上（漏怪后观看联防、被淘汰后观战）则显示你点「前往查看」选中的那名队友，没选过就显示该战场的第一名玩家（1 号联防者 / 这一组的第一位），**不会显示你自己的**。在这样的战场上点某名队友的头像 →「前往查看」，镜头会直接移到他的半场，‹ › 上的「👁 队友名」与盟约栏的标签一致；‹ › 旁边有「返回战场」（被淘汰时没有）。盟约栏换人时，从盟约栏打开的盟约详情会自动关闭。点开场上某个干员的详情卡时，卡上的盟约（以及点它打开的盟约详情）属于**这个干员的主人**，与盟约栏当前显示谁无关——例如在「全景」点队友的干员，看到的是「队友名 的盟约」及他的人数和层数；在队友的半场点你自己的干员，看到的是你自己的。回到自己的阵地 / 战场后恢复成你自己的盟约。作战结束后观看队友时刷新页面 / 断线重连，仍会回到正在观看的队友战场，「👁 队友名」和「返回战场」都还在。已被淘汰的队友不再显示盟约（也无法被查看）。

 ## 9. 结算称号

@@ -169,6 +174,25 @@
 | `Space` | 准备就绪 / 取消准备；独立模拟的作战中暂停 / 继续 |
 | `Esc` | 取消 / 关闭：依次关闭表情面板、敌人预览、盟约浮层、详情卡、本局信息 / 敌方情报（固定，不能修改） |

-刷新、冻结、升级、撤退、出售和准备只在休整期有效，撤退 / 出售要先点选一名干员。方向轮盘（方向键预览、`Enter` 确认、`Esc` 取消）、「玩法说明」和「干员调配」的翻页键是这些界面自己的按键，不能修改。
+刷新、冻结、升级、撤退、出售和准备只在休整期有效，撤退 / 出售要先点选一名干员。方向轮盘（方向键预览、`Enter` 确认、`Esc` 取消；按 Tab 先到轮盘上的「点击取消」，在它上面按 `Enter` 是取消，Tab 到其他按钮上按 `Enter` 没有反应）、「玩法说明」和「干员调配」的翻页键是这些界面自己的按键，不能修改。

-**修改快捷键**：标题画面右上角或对局左下角的「设置」→「快捷键」。点一下要改的按键（只用键盘时按 Tab 移到它上面再按 Enter），出现「按下新按键…」后按新的键；按 `Esc` 取消，设置窗口不会关闭。可以用字母、数字、空格、标点符号键和 Delete / Insert / Home / End / PageUp / PageDown；`Esc`、`Tab`、`Enter`、方向键和 F1–F12 留给界面使用，也不能搭配 Ctrl / Alt / ⌘。新按键已被另一个快捷键使用时两者互换，列表下方会写明（例如把「撤退选中干员」改成 `X`，「出售选中干员」就换成原来的 `Q`）。「恢复默认」改回上表的按键。修改立即生效，商店栏、准备按钮和单位底部「撤退 / 出售」按钮上显示的按键也跟着变；设置和音量等一起保存在当前浏览器里（换浏览器或清除网站数据后回到默认）。按键按键帽上印的字符识别（AZERTY 等键盘按印着的字母），俄文等非拉丁布局按键位识别。手机等纯触屏设备要连接键盘才能用快捷键，设置里的快捷键列表默认折叠。
+**修改快捷键**：标题画面的「设置」按钮、大厅和房间顶栏的「设置」（玩法说明左边）或对局左下角的「设置」→「快捷键」。点一下要改的按键（只用键盘时按 Tab 移到它上面再按 Enter），出现「按下新按键…」后按新的键；按 `Esc` 取消，设置窗口不会关闭。可以用字母、数字、空格、标点符号键和 Delete / Insert / Home / End / PageUp / PageDown；`Esc`、`Tab`、`Enter`、方向键和 F1–F12 留给界面使用，也不能搭配 Ctrl / Alt / ⌘。新按键已被另一个快捷键使用时两者互换，列表下方会写明（例如把「撤退选中干员」改成 `X`，「出售选中干员」就换成原来的 `Q`）。「恢复默认」改回上表的按键。修改立即生效，商店栏、准备按钮和单位底部「撤退 / 出售」按钮上显示的按键也跟着变；设置和音量等一起保存在当前浏览器里（换浏览器或清除网站数据后回到默认）。按键按键帽上印的字符识别（AZERTY 等键盘按印着的字母），俄文等非拉丁布局按键位识别。手机等纯触屏设备要连接键盘才能用快捷键，设置里的快捷键列表默认折叠。
+
+## 12. 反馈问题
+
+遇到报错、卡住或结果不对时，打开「设置」→「问题反馈」，点「复制诊断信息」，再到 [GitHub issue](https://github.com/sganggs/Stronghold-Protocol/issues) 里描述发生了什么并粘贴进去。诊断信息包括游戏版本、浏览器和设备、当前所在的阶段与回合，以及打开页面以来记录到的错误（最多保留最近 40 条）。战斗中复制时，默认附上这场战斗的数据，开发者可以用它重现这场战斗；不想附上可以关掉「附上本场战斗」。诊断信息只在你的浏览器里生成，不会自动上传；你和其他玩家的名字、房间号会被替换成 `<me>`、`<player2>`、`<room>` 这样的占位符。剪贴板不可用时（部分 App 内置浏览器），内容会显示在下方的文本框里，手动全选复制即可。
+
+## 13. 统计（本机）
+
+标题画面右上角、大厅和等待室顶栏的「统计」打开**统计数据**页：这台设备上你参与过的对局的记录。记录只存在这个浏览器里（`localStorage`），不会上传到服务器，服务器也不保存任何对局历史；换浏览器、换设备或清除网站数据后，记录不会跟过去。要搬家，用页面顶部的「导出」（一个 JSON 文件）和另一台设备上的「导入」（按记录合并，重复的跳过）。
+
+| 页签 | 内容 |
+|---|---|
+| 总览 | 总局数、胜场、胜率、到达 / 通关隐秘核心的次数、最远回合、累计时长；各难度一行；六种结算称号各拿过几次 |
+| 策略 | 每个策略用了几局、通过几局（你自己撑到最后、队伍获胜才算通过）和通过率 |
+| 战斗累计 | 击倒敌人、领袖伤害、造成伤害等结算页上的数字，在打到结算的对局里累加 |
+| 对局记录 | 每局一行，最新的在上面（一次显示 50 行，「显示更多」继续）；点一行回看那一局的结算页（最新的 30 局可以回看，更早的只留汇总，点不开） |
+
+**什么算一局**：打到结算页的对局，以及中途退出（「放弃模拟」）前至少清完 1 个回合的对局，计入以上所有统计——和官方一样，中途退出按淘汰处理，通过的回合数就是退出时已经通过的回合。**进去就退、没清完第一个回合就退的对局不计入统计**（不算一局，不影响胜率和策略通过率），但会留在「对局记录」里，标着「中途退出」和「不计入统计」。已经被淘汰后再退出的对局算（你输了这一局）。下面这些不会记录：观战；被服务器结束的对局（服务器重启、同盟解散）；关掉页面后没有再回来的对局。
+
+**存储与隐私**：最多保留最近 1000 局，所有记录加起来不超过约 0.8 MB（超出时最旧的先丢）。只有最新的 30 局保留完整的阵容、盟约和队友信息，更早的压缩成你自己的一行汇总。同盟对局的记录里有**队友的代号**，它们保存在本机，也会出现在你导出的文件里；不会发给任何人。想删掉就点「清空」（会先确认），或清除这个网站的数据。读不出来的存档（损坏的数据）会另存为备份并重新开始记录，不会让页面出错；由更新版本导出的文件会被拒绝导入，也不会被旧版本覆盖。
@@ -3,7 +3,9 @@
 Audience: **content authors** (kits, bonds, garrisons, items, bands, enemies, bosses, devices, choices) and the
 **match owner** who drives `Battle`. The normative contract is DESIGN.md §5 ([design/engine.md](design/engine.md));
 this file documents the concrete implementation, every hook and helper, the SkillSpec schema with worked examples,
-the profession defaults and the test harness. Everything here is deterministic: the only randomness is `battle.rng()`.
+the profession defaults and the test harness. Everything here is deterministic: the only randomness is `battle.rng()`,
+and the only floating-point functions are correctly rounded ones (`detmath.js` below), so a battle gives the same bits in
+every browser and on the server.

 ```
 server/sim/
@@ -13,6 +15,8 @@ server/sim/
                   tiles (relocation, bodies, tactical points), displacement, economy (DP, layers, coins), events
  constants.js     TICK, MOVE_SCALE, ATTACK_PAUSE, element numbers, tuning knobs
  rng.js           mulberry32 PRNG (+ int/range/chance/pick/shuffle/weighted)
+  detmath.js       hypot / powi / sin / cos / atan2 with the same bits on every engine — the sim never calls the
+                   implementation-approximated Math functions (hypot, sin, cos, atan2, pow, ** …; lint + test enforce it)
  grid.js          stage grid, tile semantics, 8-dir A* (no corner cutting), obstacles
  units.js         Unit model + stat aggregation
  buffs.js         Buff model, mod keys, status catalogue
@@ -180,10 +184,20 @@ by a `battleStart` handler covers it like a later knock-out (机变 征召 −50

 `SpawnSpec = { time, enemyKey, routeIndex, count=1, interval=0, mods:{hpMul,atkMul,defMul,resMul,speedMul}, sourcePlayerId?,
 bounty?:{coins, ownerPlayerId}, tag?:'boss'|'part'|'escort'|'bounty', ownerPlayerId?, pos?:[r,c], route?:RouteSpec, countInTotal? }`.
+A spec entry the battle queues (`_queueSpawn`) is `inTotal` when it counts (`countInTotal`, `tag`, data `notCountInTotal`):
+that is the HUD capsule's own set (§1.3); everything content spawns at runtime (a split, a summon — a boss's included —,
+a part) is not, even when it counts for LP (`counted`).
 `RouteSpec` accepts data/waves.json routes (`{motion, start, end, checkpoints:[[r,c]…], steps:[{t:'move',p},{t:'wait',s},{t:'disappear'},{t:'appear',p}]}`)
 and research routes (`{m, s, e, cp:[['MOVE',r,c]…]}`). `spawnsFromTemplate(waveEntry, {mods})` (simdata.js) converts a
 template into `{ routes, spawns, maxPlayTime, overrides, extraRoutes }` (non-spawn `action` entries are skipped; `unharmful`
-and `tag:'part'` spawns don't count in `total`).
+and `tag:'part'` spawns don't count in `total`). The boss templates' 传送门 reads as "reappear on the far side" (GitHub
+#336, PR #337): four official routes write the crossing as `disappear → wait → move` to the exit with no `appear` (their
+twins spell the `appear` out), so normalizeRoute re-inserts the exit `appear` ahead of that move — otherwise the enemy
+walks the rest of its route hidden and leaks unseen (test/sim/portal-appear.test.js). A route that ENDS on an entrance
+(`tile_telin`; the boss / Hidden Core circuits end on [1,3] / [1,17]) does not leak there: the enemy vanishes, comes
+out of the far exit (`tile_telout`; both entrances feed [5,10], the pairing of every explicit `disappear`/`appear` pair
+of the data) after 3 s hidden inside — the wait of the explicit crossings (85 of 95) — and walks on to the blue door
+on the entrance's side, where it leaks (ai.js portalPickup; [ASSUMED] the pairing and the 3 s).

 A `bounty` pays `coins` once, when the enemy really dies (not a knock-out it survives; a leak pays nothing), to
 `Battle._bountyPayee`: the player of the operator or summon that dealt the blow, if that player is in the battle; any
@@ -194,13 +208,19 @@ Leader parts (`tag:'part'`) pass damage to their leader with `loseHp(leader, sha
 (无来源, credited to the attacker's `bossDamage`): `PART_TRANSFER` 1 for 斩胄之剑 / 破胄之锤 (`BLADE_TRANSFER`, the same
 constant) and 碎铳之簧 (PRTS "受到伤害时令假想敌：胄/铳受到等量的无来源生命流失"; DESIGN §20.10, §20.13). Content may replace a part mid-battle: every 剑/锤 sortie (content/bosses.js `kitBlade`)
 ends by spawning a new 初始模式 copy on its level branch route (`left_hand_origin` / `right_hand_origin`) with the old
-HP, then `kill(old, null)` — uncounted, no bounty; the client sees a `die` and a `spawn`.
+HP, then `kill(old, null)` — uncounted, no bounty; the client sees a `die` and a `spawn`. The Hidden Core 铳's 【末日布道】
+puts every 碎铳之簧 into 追逐模式 for its whole 5 s gain (`dog_duration`; invulnerable, no ordinary attacks): the spring
+keeps following the casting gun after reaching it (PR #347; until 0.2.1 arrival or 1 tile ended the chase), and each
+gun casts on its own data timer (20 s after it spawns, then every 45 s), so on a pair field the later call retargets
+every spring to that gun for a new 5 s (PRTS: 「持续召唤场上所有“碎铳之簧”向自身移动」).

 WALK legs pathfind on the stage grid inside the rect with the official flow field (grid.js: 4-direction SPFA from the
 destination, crates cost 1000, then Bresenham line-of-sight smoothing — research 08 §3.4). The official route stays
 unless 0.1.0's road-over-floor preference route (the fewest non-blockable tiles among equal-length chains, a line of
 sight that never covers floor — diagonal-step corners included — its grid route does not walk; user playtest #2 item 2)
-crosses strictly fewer non-blockable tiles (floor / gate lanes); "crosses" = passes through the tile's interior, a
+crosses strictly fewer non-blockable tiles (floor / gate lanes) and no more 深水区 (GitHub #375: on 战场#08(下半) the
+patrolling 铳 / 卢西恩 waded through two water tiles to dodge one floor tile; the official upper road crosses one);
+"crosses" = passes through the tile's interior, a
 corner touch does not count (`grid.js` `segmentTiles`; community report D5 after 0.1.0: 战场#04's lower-gate enemies cut
 diagonally from row 9 into row 10 as officially), and on equal counts the official route stays.
 test/sim/pathing-official.test.js lists the 21 of 154 stage routes that still differ from the official ones (战场#01's
@@ -257,7 +277,9 @@ included), summons (流形's melee copy) and a ranged operator standing on a mel
 enemies it blocks, in range or not, whatever its facing, and targets them first ("可以选择且优先选择阻挡单位", PRTS 选择器;
 the user's rule after playtest #6, "阻挡了就一定要能打到": officially the collision pushes a blocked enemy to its blocker's
 front — `Battle.blockedTargets`, used by `ai.js acquireTargets` and the skills' DEFAULT trigger, and
-`sortEnemyTargets`; "自身这格内" rules name the blocked enemies separately — 瑕光 S2, PRTS 备注). A kit that picks its own
+`sortEnemyTargets`; one exception, the owner's decision of 2026-10-08 (GitHub #220): the 速射手 air priority —
+priority 'fly', trait 优先攻击空中单位 — comes before its own blocked enemy, so a 速狙 on a melee tile shoots a flyer in its
+range first, PRTS 索敌的概念 「阻挡（近战限定）→特殊优先级→…」; every other priority keeps the blocked enemy first; "自身这格内" rules name the blocked enemies separately — 瑕光 S2, PRTS 备注). A kit that picks its own
 targets from the range (`beforeAttack`) adds `battle.blockedTargets` to its candidates (深靛, 迷迭香 S3, 荒芜拉普兰德 S2,
 佩佩 S2, 灵知 S3, 远牙 S3); only skill texts that exclude targets keep their rule (普罗旺斯 S2: none above 80 % HP; 寒檀
 S2: icicles on random tiles of her range). This replaces the
@@ -358,6 +380,10 @@ one; the sourceless 毒雾 of 假想敌：蚀裂 skips a 隐匿 one but still re
 范围内的所有单位持续视为受到源石污染区影响，技力自然回复速度倍率-80%，每0.5秒受到50真实持续伤害（同类效果取最高）") is the same
 kind of damage — 无来源 true (like the terrain it stands for [ASSUMED]), the chimera credited — not a 流失 (player report
 D1 audit): radius 1.2, a tick every 0.5 s, one tick per unit per 0.5 s however many chimeras reach it (`mem.chimeraAt`).
+Every enemy effect PRTS marks "不可对空" skips a flying ally (the 炎佑 dragon, `isFlying`): the 死亡爆炸 of 高能源石虫 / 冰爆源石虫 /
+卷心籽 (archetypes.js `deathBoom`, `noAir` by default), 水遁忍者's 漩涡形态, 鼎沸's pulses, “萨科塔昂首”'s 祈祷邀约, the attack of
+“斩胄之剑” / “破胄之锤”, the splash and blast of 烹泉 / 沏虹, 庞贝's blast and 集团军重型火炮's shell and 燃烧区域 (the first six
+reached it until 0.2.1 — a community report).

 **Knock-outs that are not deaths** (`content/enemies/archetypes.js`; player reports after 0.1.0): a `killed` ability that keeps the
 enemy alive hides the knock-out from every later `kill` handler, the kill count, kill credit and the bounty — they all
@@ -406,13 +432,32 @@ recorded for the player whose half contains the goal it reached, with `sourcePla
 ### 1.3 BattleResult

 ```js
-{ time, reason: 'cleared'|'timeout'|'forced', killed, total, errors, unspawned?: [{enemyKey,time,tag}], bossHpLeft?,
-  perPlayer: { [playerId]: { killed, total, perfect,
+{ time, reason: 'cleared'|'timeout'|'forced', killed, total, resolved, errors, unspawned?: [{enemyKey,time,tag}], bossHpLeft?,
+  perPlayer: { [playerId]: { killed, total, resolved, killedInTotal, leakedInTotal, perfect,
      leaked: [{ enemyKey, mods, lpr, sourcePlayerId, tag, counted, boss?, spawned }],
      layerGains: {bondId: n}, coins, damageDealt, bossDamage, healingDone, deaths,
      unitsEnd: [{ uid, id, defId, hpPct, sp, skillActive, alive }],   // operators + board summon pieces; sp = skill.spTotal
      unitStats: [{ id, uid, defId, name, kind, dmg, kills, heal, taken, attacks }] } } }
 ```
+- **Two sets of counters, and they never cross.** `counted` drives the official LP / 完美作战 / `killed` reading (master's,
+  untouched): every counted enemy — a runtime split child, a summon, a boss's summon included — is in it. `inTotal` (the
+  new flag, `spawns.js`) is the HUD capsule's own set (DESIGN §14 顶栏胶囊): only the enemies the field itself scheduled
+  (`_queueSpawn`, and only those that `counted` — `countInTotal: false`, `tag: 'boss'` / `'part'`, data `notCountInTotal`
+  stay out) are `inTotal`. Everything a runtime `spawnEnemy` makes defaults to `inTotal: false` (an explicit
+  `inTotal: true` is the content back door); a 变身 / 重生 (`setForm`, a revive) is the same enemy and keeps its flag.
+- `total` counts the `inTotal` enemies (the capsule's denominator), `killedInTotal` / `leakedInTotal` (battle level and per
+  player) count those that were knocked down or leaked, and `resolved = min(total, killedInTotal + leakedInTotal)` is the
+  capsule's numerator (official: 开局 0/3 → 漏一个 1/3 → 打死一个 2/3 → 打死会分裂的 3/3). A leaked split child or summon
+  moves none of them, while `leakedCount` / `leaked[]` still charge its LP and break 完美作战. The per-player numbers are
+  attributed like their `counted` twins — a knock-out to the enemy's owner (as `total`), a leak to the player whose half it
+  reached (`_recordLeak`) — so in a two-half field (the 联防 lane, col 18 → col 2; the boss pair's crossing routes) an enemy
+  that spawns on one half and leaks on the other is in one player's `total` and in the other's `leakedInTotal`, and each
+  player's own `resolved` clamps it away (0 and 0 for an enemy the field resolved). Only the battle-level numbers are exact:
+  the match never sums the per-player `resolved` — `validateClientResult` keeps the result's own and `m.public` publishes
+  it (design/network.md 顶栏胶囊).
+- `killed` therefore may **exceed** `total` (it counts runtime children, they do not): 4 磨砻 killed plus one of their 8
+  木制瑞印 reads `killed` 5 with `total` 4 and a capsule of 4/4. `spec.js compactResult` and the match's validation bound
+  `killed` by the spec's content-spawn allowance (`4·spawns + 100`) instead of by `total`.
 - `cleared`: no spawns pending and no enemy alive (or the shared boss pool reached 0 — a pool never holds less than
  `BOSS_POOL_MIN_HP` (1): the hit that would leave less takes the rest and the browser's `LocalBossPool` reads less as
  0, so float dust of the pool sync can never keep a leader alive; user playtest #6). A boss/hidden battle **with a
@@ -520,6 +565,16 @@ Aggregation: `ATK/DEF/maxHp = (base + Σflat) × (1 + Σpct) × Πmul`, for ATK
 percentages: `ATK = ((base + ΣatkFlat) × (1 + ΣatkPct) + ΣatkFinal) × ΠatkMul`; `res = clamp((base + ΣresFlat) × ΠresMul, 0, 100)`;
 `aspd = clamp(base + Σaspd, 20, 600)` (floor 20: PRTS 数值范围 ATTACK_SPEED 默认下限; user playtest #6); `interval = bat × (1 + ΣbatPct) × 100 / aspd`; `moveSpeed = (base + ΣmoveFlat) × ΠmoveMul`;
 tiles/s = `moveSpeed × MOVE_SCALE (0.5)`. A maxHp change keeps the HP ratio. Elite stats (module included) come from data.
+The attack cooldown `atkCd` starts at `interval` and counts down by `TICK`; what is left within 1e-9 of 0 is 0
+(`ai.js attackCountdown`, the tolerance of skill `timeLeft`, buff intervals and `every()`; PR #402 by @Cloudnyco, 0.2.2),
+so an interval of a whole number of ticks takes exactly that many: 1 s = 30 ticks, 3 s = 90 (floating point alone left
+2.1e-16 after thirty steps of 1/30, and every such attack came a tick late — 31, 91). The same countdown runs the 双眼皮
+turrets (`content/devices.js`) and 寒檀's S2 icicles; the drones of 澄闪 / 荒芜拉普兰德 and 电弧's 赛柯 volleys compared
+with 1e-9 already. Content clocks that count up and restart from 0 (the boss self-timers of `content/bosses.js`, kit auras)
+are not covered. Official (PRTS 作战机制/sandbox 「帧对齐机制对攻速的影响」): the time unit is one frame, 1/30 s, and since
+the 2019-12-24 update a non-integer frame count is rounded (四舍五入; before, 进一取整); an exact one stays. Here a
+non-integer count still ends on the tick that crosses 0 — ⌈30 × interval⌉ ticks (1.2 s = 36, 1.25 s = 38; 能天使's
+0.678 s is 21 here, 20 frames officially): an older difference from the official rounding, not changed by this rule.

 **Which bucket (PRTS 游戏数据基础 属性基本公式 / 作战机制, DESIGN §20.10).** `Σpct` is the official **直接乘算** class — its
 values are summed (`A = (A₀ + D_p)(1 + D_t)`, D_t = t₁ + … + tₙ): a skill's or talent's "攻击力+X%" **and** every "+X%"
@@ -527,8 +582,9 @@ ATK / DEF / max HP bonus of the 卫戍 systems — 盟约, 策略 (bands), 装
 卫戍协议：盟约 下半/PRTS盟约记录 "盟约效果，策略效果，装备效果提供的属性加成均为直接乘算"). Content builds those with
 `content/support directMods({ atk, def, hp })` (constants.js `DIRECT_BONUS_STACKING` 'add'; 'multiply' = the v2.5
 per-source ×(1 + x), which compounded with layers: user report after playtest #6). `Πmul` is for 最终乘算 / "提升至X%"
-effects (炎佑 ×1.5 at 9 炎, 虚弱, 停顿 …) and the char_attribute_mul 特质 ("攻击力和生命值+20%", a rune on the base
-attributes). `atkFinal` is the **最终加算** (PRTS `A_f = F_t[(A + D_p)(1 + D_t) + F_p]`: added after the 直接乘算, inside
+effects (炎佑 ×1.5 at 9 炎, 虚弱, 停顿 …), the char_attribute_mul 特质 ("攻击力和生命值+20%", a rune on the base
+attributes) and the unit's 练度 `cultMul` (自持有, the same char_attribute_mul key, 0.2.2: ATK / DEF / max HP of the owned
+operator's tier, §12). `atkFinal` is the **最终加算** (PRTS `A_f = F_t[(A + D_p)(1 + D_t) + F_p]`: added after the 直接乘算, inside
 the 最终乘算): 阿戈尔's devoured base ATK ("基础攻击力（最终加算）", DESIGN §24.7) — a skill's ATK +% does not scale it.
 Damage multipliers (`dmgDealtMul`, "伤害提升至X%"; `*TakenMul`, 脆弱 / "受到的伤害+X%") multiply each other
 (PRTS 游戏数据基础 "同种倍率间叠乘"), same-named statuses keep the strongest — catalogue statuses (§3) and the content
@@ -592,7 +648,9 @@ and the 浮空 weight rule, then `statusApplied { source, target, status, durati
 = the target carried no buff of that status before (a refresh / a weaker "取最高" application is not an entry: "进入…时"). Effects follow the official term table
 (`gamedata_const.termDescriptionDict`, `ba.*`). Same-key statuses refresh to the longer duration, except the
 "同名效果取最高" ones marked *strongest* below: the strongest value wins, a weaker application never overrides it and,
-if it outlasts it, resumes when the strong one expires (pass `refresh` to opt out); `stackAs` = the strength such an
+if it outlasts it, resumes when the strong one expires (pass `refresh` to opt out) — every such waiting application keeps
+its own end, and they resume strongest first (`buff.data.tail` is a chain ordered by strength: of 30 % / 25 % / 20 %
+applied one after another, the 25 % follows the 30 %, then the 20 % — GitHub #342, PR #348); `stackAs` = the strength such an
 application competes with instead of its `value`, which stays its effect (Raidian S3's 虚弱: PRTS 备注 "在叠加时视为90%…的
 虚弱（仅影响叠加优先级，不影响实际效果）").
 `battle.applyStrongest(target, key, { duration, value, mods: (v) => mods, source })` gives a content effect that is not a
@@ -739,7 +797,8 @@ selection (`injuredAlliesInKeys`) takes it — 凯尔希 on her Mon3tr (PRTS "
 describes as raising the target's 生命回复速度 are hpRegen buffs, never `heal` calls, so 禁疗 and 无法被友方治疗 (`noHeal`:
 收割者 / 不屈者 / 武者) do not stop them: the 吟游者 trait (professions.js `bardRegen`, 分支特性信息 吟游者; 魔王's 微尘 ×1.5
 through its `bardRegen` hook; 浊心斯卡蒂 and her 海嗣), 调香师's 熏衣草, 瑕光 S2, 铃兰 S3 (none in its first second, refreshed
-every second), 锡人's 炼金单元 (GitHub #96 / #137); `overheal`
+every second), 锡人's 炼金单元 (GitHub #96 / #137) and 引星棘刺's S1 / S2 炼金单元 (the same wording and blackboard key as
+锡人 S2; a community report, 0.2.2); `overheal`
 turns the excess into an `overheal` shield. `battle.loseHp(target, amount, { source, from, tags, silent, sourceless })` = HP
 loss ignoring DEF/RES/shields/dodge (流失); `sourceless: true` makes it 无来源 ("受到等量的无来源生命流失": hooks see no source,
 `source` keeps the credit — stats and the per-player shared-pool tally), as does a 无来源 `from`. A 流失 skips the damage
@@ -829,7 +888,7 @@ instance) and skip `'counter'` / `'reflect'` damage. When the guard trips, the l
 | `addBuff(unit, buff)`, `removeBuff(unit, key)` | §3 |
 | `spawnToken(ownerUnit | playerId, tokenId, row, col, { def, stats, hp, duration, untargetable, dir, kit, force, anySource })` | field tiles; def from data/tokens.json `variants[ownerChessId]` for the owner unit's selected skill / module (`tokenDef`); `dir` defaults to the owner unit's (else the player's: RIGHT, mirrored side LEFT; a legacy `facing` ±1 is still read); returns the token or null (tile busy; or the owner runs a **non-default** skill that does not produce the token — `producesToken` — unless `anySource`: kit install hooks written for the default skill run under every skill). `spawnDevice(key, row, col, { …, dir })` likewise |
 | `tokenDef(tokenId, ownerUnit | chessId)`, `producesToken(ownerUnit, tokenId)` | the token def a summon of that owner gets — `getToken(id, owner.defId, owner.def.loadout)`, exact even when two players of one field give the same chess different loadouts (prefer it over an id-only `battle.data.getToken(id, unit.defId)` for summon stats / blackboards); whether the owner's loadout makes the token (DATA.md §14 `sources` has 'skill' or 'talent'; true when the data does not tell: no own variant, player-owned summons) |
-| `spawnEnemy(enemyKey, { routeIndex, route, pos, mods, tag, sourcePlayerId, ownerPlayerId, bounty, countInTotal, def })` | returns the enemy, or null past `MAX_ALIVE_ENEMIES`; `def` = inline record (enemies.json shape or normalised) for keys missing from data |
+| `spawnEnemy(enemyKey, { routeIndex, route, pos, mods, tag, sourcePlayerId, ownerPlayerId, bounty, countInTotal, inTotal, def })` | returns the enemy, or null past `MAX_ALIVE_ENEMIES`; `def` = inline record (enemies.json shape or normalised) for keys missing from data. A runtime spawn is never `inTotal` (§1.3 — the HUD capsule's own set) unless content passes `inTotal: true` explicitly |
 | `spawnDevice(key, row, col, { hp, obstacle, blockCnt, name, def, res, atk, bat, aspd })`, `setObstacle(r, c, on)` | obstacles re-path enemies; spawnDevice returns null outside the rect, on a living unit or on a knocked-out operator's tile |
 | `isReservedTile(r, c)` | true when a living unit stands there or it is the rest tile (`restTile`) of an ally piece that has not deployed yet / waits to redeploy — the tile a knocked-out operator lies on, else the home tile — every automatic picker (the 突袭 landing tile, tactical points, summon / device tiles) must skip these (`findTacticalPoint` does) |
 | `downOn(r, c, except?)`, `restTile(u)` | the down operator (`isDown`: knocked out or forced out) lying on a tile (no ally deploys / moves there: `_deploy`, `spawnDevice`, `relocate`); the tile a withdrawn ally comes back on — its `body` tile when down, else its home (§1) |
@@ -837,7 +896,7 @@ instance) and skip `'counter'` / `'reflect'` damage. When the guard trips, the l
 | `allySelectable(ally, by)`, `alliesFor(by, ownerId?)` | may an ability of ally `by` select `ally` — never a 孤立 unit (炎佑, 从不混淆的方向) but `by` itself (PRTS 选择器: "若掩码中孤立为1，且选择器的阵营与目标为友好关系，则不可选中"); `allies()` without the 孤立 ones — content picks ally targets (buffs, auras, 全场 talents, heal picks) through these, `alliesInGrid` too |
 | `unitsInGrid(unit, grid, {side, extend})`, `alliesInGrid(unit)`, `enemiesInRadius(x, y, r, centre?)`, `foesInRadius(x, y, r, centre?)`, `alliesInRadius(x, y, r, ownerId?)` | grid offsets are relative to facing RIGHT, rotated by `unit.dir`; enemies by their body (§2 hit areas: a huge enemy on every tile it occupies / within `r` of its rectangle; `centre` = splash around a target, a 中点判定 by position; `foesInRadius` = the enemies an operator's area effect can select — no untargetable enemy, none whose 隐匿 is on (`targeting.js enemyStealthed`: not blocked, not revealed, not within 3 s of its last block), PRTS 作战机制 §AOE伤害判定 — while enemy-side auras / heals and collisions keep `enemiesInRadius`) |
 | `enemiesInKeys(keys, attacker, profile)`, `blockedTargets(unit, profile)` | targetable enemies whose body is on the tiles (a huge one listed once); the enemies a unit blocks — always selectable by it, a ranged operator on a melee tile included (§1.2 Blocking) |
-| `setAllyTarget(unit, on)`, `allyTargetsInKeys(keys, attacker)`, `isAllyTarget(unit)` | an **ally target**: an ally unit our attacks select like an enemy — 白铁's 铁钳号·原型机, a summon of the enemy camp ("可被我方干员攻击但不受伤害", PRTS; kits/ops/op-ironmn.js). The attack loop (ai.js `acquireTargets`) appends the registered ones standing on the attacker's range after every enemy (their 嘲讽等级 −2 puts them last: one target ⇒ an enemy first; all in range / more targets ⇒ them too) and ranged attacks fly to them; heals, skills that pick their own victims and enemies never select them (the kit gives it `untargetable` against enemies and cancels the damage it takes in its `hit` hook). None registered ⇒ the loop is unchanged |
+| `setAllyTarget(unit, on)`, `allyTargetsInKeys(keys, attacker)`, `isAllyTarget(unit)` | an **ally target**: an ally unit our attacks select like an enemy — 白铁's 铁钳号·原型机, a summon of the enemy camp ("可被我方干员攻击但不受伤害", PRTS; kits/ops/op-ironmn.js). The attack loop (ai.js `acquireTargets`) appends the registered ones standing on the attacker's range after every enemy (their 嘲讽等级 −2 puts them last: one target ⇒ an enemy first; all in range / more targets ⇒ them too) and ranged attacks fly to them; heals, skills that pick their own victims and enemies never select them (the kit gives it `untargetable` against enemies and cancels the damage it takes in its `hit` hook). A skill that acts on them counts the registered ones like enemies for its automatic start (skills.js `_allyTargetIn` / `allyTargetsOk`, every enemy condition of the trigger rules — the owner's rule of 2026-10-08): one that acts through the unit's own attacks, or whose kit flags `allyTargets: true` because its selectors take them too (`allyTargetsInKeys` / `allyTargetsInRadius`: 蕾缪安 S3's locks, 艾丽妮 S3, 陈 S2, 引星棘刺 S2); [ASSUMED] a skill that picks its victims among enemies only (or whose kit says `allyTargets: false`) is not opened by them and waits for an enemy. None registered ⇒ the loop is unchanged |
 | `allies(ownerId?)`, `aliveEnemies()`, `unitAt(r, c)`, `unitById(id)`, `tileInfo(r, c)`, `lowestHpAllyInRange(unit)` | |
 | `addLayers(playerId, bondId, n, reason, {source})`, `addCoins(playerId, n)` | layers are a no-op when `flags.layerGainsEnabled` is false (unite/boss); a gain adds at most the room left under `BOND_LAYER_CAP` (999, shared/constants.js `layerGainRoom`: the client's `AddBondCount` min(L + n, 999)) on the live copy — or, without one, on the battle's own gains — and returns what it added (0 at the cap: no hook, no event) |
 | `getPlayer(playerId)` | `{ playerId, seat, side, colOffset, mirror, dir (default unit direction: RIGHT, mirrored side LEFT), facing (its sign), bonds (live copy, layers updated by addLayers), bandId, playerEffects, lpForBoss, dp, units }` |
@@ -846,12 +905,12 @@ instance) and skip `'counter'` / `'reflect'` damage. When the guard trips, the l
 | `onFieldBoard(r, c)` | whether a field tile lies on the board of a player of this field — `onOwnBoard` of any of `players` (the players whose units the battle holds; an eliminated or absent teammate is not one): both halves of the two-helper 联防 field and of a Final Assault / Hidden Core pair field, the own half only for a lone 联防 helper and on a solo boss field, never a boss field's hand / 临时整备区 rows; the 突袭 landing takes only such tiles (DESIGN §25.18, §26.1) |
 | `addDp(playerId, n)`, `retreat(unit, {reason, permanent, dying})`, `relocate(unit, r, c)` | `relocate` only changes the tile (state kept, no event) |
 | `moveRedeploy(unit, r, c, { clearSp })` | a 【移动】 (PRTS 术语释义: "不退场，以当前血量在目标位置部署", a special retreat + redeploy): `relocate`'s checks (false = refused, nothing changed), then a new deployment (`deploySeq` / `aggroSeq` / `deployedAt`) and `deploy {initial:false, move:true}` (deploy effects fire again); no `die` / `death`, timer or cost (不屈 / 阿戈尔's revive never see it); HP, buffs and a running skill are kept (owner's deviation, DESIGN §22.3); `clearSp` empties the SP before the deploy handlers run — 乌尔比安 S3's move (kits/ops/chess_char_5_05-ulpia.js) and 【返回】 (`clearSp`; also the 从不混淆的方向 fallback, content/tokens.js) |
-| `redeploy(unit, { free=true, tile, keepSp })` | immediate (re)deployment of a dead/retreated ally (full HP, `deploy {initial:false}`); `free: false` pays `base.cost` DP (refused without it); without `tile` it lands on the unit's rest tile (`restTile`: where a knocked-out operator lies, else home); `tile: [r, c]` lands on that tile once (home unchanged; refused when off-rect, occupied or a knocked-out operator's tile, no fallback); `keepSp` keeps SP/charges (保留技力), restored before `deploy` fires — 突袭 raids, 阿戈尔 / 不屈 revives where the unit lies |
+| `redeploy(unit, { free=true, tile, keepSp })` | immediate (re)deployment of a dead/retreated ally (full HP, `deploy {initial:false}`); `free: false` pays `base.cost` DP (refused without it); without `tile` it lands on the unit's rest tile (`restTile`: where a knocked-out operator lies, else home); `tile: [r, c]` lands on that tile once (home unchanged; refused when off-rect, occupied or a knocked-out operator's tile, no fallback); `keepSp` keeps SP/charges (保留技力), restored before `deploy` fires — but not for a skill that costs no SP, which the reset re-arms as at every deployment (伊内丝 S3 casts again on a 突袭 landing) — 突袭 raids, 阿戈尔 / 不屈 revives where the unit lies |
 | `refreshRange(unit)`, `setExtraRange(unit, keys)`, `rangeChanged(unit)` | rebuild the ranges after changing `unit.rangeGrid` (流形 copies); extra targetable tiles (absolute keys; merged into every later rebuild until set again; `null` clears; never in `baseRangeKeys`): 蕾缪安 wanted, 维娜 S3 |
-| `push(enemy, force, {from, dir, fixed, fixedAngle, inward, effect})`, `pull(enemy, force, {to, center, stop})`, `pullToFront(enemy, unit, force)`, `forceLevel(enemy, force)`, `pushDistance(enemy, force, {effect})` | the official 位移 (PRTS 游戏数据基础 §重量公式 / 推与拉; user playtest #6 item 14): 受力等级 = 力度 (微小力 −1, 小力 0, 中力 1, 较大力 2, 大力 3 …) − current 重量等级 (massLevel, 失重 counts). Push distance per level (`constants.js PUSH_TILES`, PRTS 推与拉's 弹道 column): ≤ −3 → 0, −2 → 0.12, −1 → 0.44, 0 → 1.7, 1 → 2.14, 2 → 2.96, ≥ 3 → 3.53 tiles; `effect` = a 特效 push (`PUSH_TILES_EFFECT`: −2 → 0.085, −1 → 0.374, 0 → 1.562, 1 → 1.987, 2 → 2.773, ≥ 3 → 3.331 — 见行者 S1 / S2, `PUSH_EFFECT_SKILLS`; every other pusher uses the 弹道 column [ASSUMED]); radial (away from `from`; the client's buff template `knockback[relative]`: 莫斯提马 S3, 山 S3, 琳琅诗怀雅 S3 — also PRTS 备注 "推开效果为径向推动") unless `dir` (directional; template `knockback[dir]` = Knockback {`_useSourceDirection` true, `_decreaseForceLevelWhenNotInDirection` 2}: 推击手, 野鬃 S2 "往攻击方向" (charpack char_496_wildmn: buff `wildmn_s_2[force]`) — > 45° off or < 0.25 tile ⇒ radial and level −2; `fixed` waives both (圣聆初雪 S1 朝部署方向, `KnockBackWithCharacterDirection`), `fixedAngle` only the angle (见行者 S2)); `inward` = a push towards `from` (薄绿 S2), stopping at the 急停 radius. Pull: level ≥ 0 → to `to` / the 急停 radius around `center`, −1 → 35 % of the way, −2 → 0.03, ≤ −3 → 0; `pullToFront` aims at the 拉力起点 0.5 tile ahead of the unit with the 急停 radius 0.6708 around it (never moves an enemy the unit itself blocks — nor does a pull towards an ally's centre, e.g. the 流形 S3 pulse) A 静态刚体 (data `staticBody`: every air unit of the mode except “炎佑”, plus 昆图斯 — PRTS 特殊机制 "可以进入失衡状态…但物理层面上无法产生任何速度或移动", "与单位的行动方式无关") moves 0 from every source while the skills that reach it still hit it (薄绿 S2, 锏 S3, the 钩索师 …; player report after 0.1.0) |
-| `displace(enemy, {x, y}, tiles)` | the raw mover behind push / pull: along passable tiles (by `motion`: FLY in the rect, else ground-passable), no weight rule; nothing for 失衡免疫 (`noDisplace`: 近地悬浮 incl. 喷气人's 飞行模式, 浮空, the 胄 parts, 守墓石像's statue and flight), 静态刚体 (`def.staticBody`) and leaders (`_displaceable`); unblocks + re-paths |
+| `push(enemy, force, {from, dir, fixed, fixedAngle, inward, effect})`, `pull(enemy, force, {to, center, stop})`, `pullToFront(enemy, unit, force)`, `forceLevel(enemy, force)`, `pushDistance(enemy, force, {effect})` | the official 位移 (PRTS 游戏数据基础 §重量公式 / 推与拉; user playtest #6 item 14): 受力等级 = 力度 (微小力 −1, 小力 0, 中力 1, 较大力 2, 大力 3 …) − current 重量等级 (massLevel, 失重 counts). Push distance per level (`constants.js PUSH_TILES`, PRTS 推与拉's 弹道 column): ≤ −3 → 0, −2 → 0.12, −1 → 0.44, 0 → 1.7, 1 → 2.14, 2 → 2.96, ≥ 3 → 3.53 tiles; `effect` = a 特效 push (`PUSH_TILES_EFFECT`: −2 → 0.085, −1 → 0.374, 0 → 1.562, 1 → 1.987, 2 → 2.773, ≥ 3 → 3.331 — 见行者 S1 / S2, `PUSH_EFFECT_SKILLS`; every other pusher uses the 弹道 column [ASSUMED]); radial (away from `from`; the client's buff template `knockback[relative]`: 莫斯提马 S3, 山 S3, 琳琅诗怀雅 S3 — also PRTS 备注 "推开效果为径向推动") unless `dir` (directional; template `knockback[dir]` = Knockback {`_useSourceDirection` true, `_decreaseForceLevelWhenNotInDirection` 2}: 推击手, 野鬃 S2 "往攻击方向" (charpack char_496_wildmn: buff `wildmn_s_2[force]`) — > 45° off or < 0.25 tile ⇒ radial and level −2; `fixed` waives both (圣聆初雪 S1 朝部署方向, `KnockBackWithCharacterDirection`), `fixedAngle` only the angle (见行者 S2)); `inward` = a push towards `from` (薄绿 S2), stopping at the 急停 radius. Pull: level ≥ 0 → to `to` / the 急停 radius around `center`, −1 → 35 % of the way, −2 → 0.03, ≤ −3 → 0; `pullToFront` aims at the 拉力起点 0.5 tile ahead of the unit with the 急停 radius 0.6708 around it (never moves an enemy the unit itself blocks — nor does a pull towards an ally's centre, e.g. the 流形 S3 pulse) A 静态刚体 (data `staticBody`: every air unit of the mode except “炎佑”, plus 昆图斯 — PRTS 特殊机制 "可以进入失衡状态…但物理层面上无法产生任何速度或移动", "与单位的行动方式无关") moves 0 from every source while the skills that reach it still hit it (薄绿 S2, 锏 S3, the 钩索师 …; player report after 0.1.0). **失衡 (UNBALANCE)** (PR #392's idea, its timing re-done from the sources): the move stays instant, then the enemy is held in the 失衡 state — a state machine, not a status (PRTS 失衡位移机制; 异常效果 「失衡…不属于异常效果」: no stun, no immunity, SP untouched), `enemy.unbalanceUntil`, read by `ai.js updateEnemy`: no walking (恐惧 / 诱导 included — no fear exemption) and no normal attack started, a swing short of its frame cut, the attack clip's stand ended; and no skill (失衡免疫 「失衡期间无法自主移动、发动攻击、使用技能」): content skills wait (`enemies/helpers.js canCast(e, sil, battle)`, `unbalancedNow`), so do blinks (`blinkForward`) and scripted moves (枯朽之种's dive, the 碎铳之簧 / 铳 charges), and 乌顶巨角卢鲁's clash waits for its end. Left running on purpose: checks the state does not stop (异常效果: 水遁忍者's damage — 「仍然能正常造成伤害」), passive auras and terrain / periodic effects (no skills), death / damage-taken triggers, skill cooldowns (only the cast waits) — and 弧光锋卫's bleed, which is the state's own (天赋 「处于失衡状态时，每0.066s受到400点无来源真实持续伤害」: every 0.066 s for as long as it lasts — until 0.2.2 one hit per tile moved [ASSUMED]). Length, game seconds: a push — the 位移时间 of its 受力等级, PRTS 游戏数据基础 推力-位移近似对应表 (`constants.js PUSH_UNBALANCE`: −2 → 0.2, −1 → 0.4, 0 → 0.8, 1 → 0.9, 2 → 32/30, ≥ 3 → 35/30), the level push() computes (the −2 correction included), kept when the 特效 column or a wall shortens the slide ([ASSUMED] for the wall; a wall at the first step → the 0.1 s floor [ASSUMED]); a pull — its force window, 推与拉 「作用时间默认为 1 s；若本次受力等级 < −1，作用时间改为 0.5 s」 (`PULL_UNBALANCE` / `_WEAK`), to its end after the 急停 too (「目标将仍保持失衡状态至拉力作用时间结束为止」), none at ≤ −3; a 静态刚体 hit by a force > 0 — no movement, the state while the force acts (特殊机制 静态刚体: 「在失去施力后立刻」 out, 「0.1 秒保底」): a push 0.1 s (`UNBALANCE_MIN`; a push's force is instant), a pull its window; 失衡免疫 and leaders never. 浮空 ends it (术语释义 浮空), so do a 重生 (重生 / 逐火's 余烬) and a route APPEAR relocation — forced state switches (失衡位移机制). The end compares with 1e-9 so a float-summed 35/30 s keeps its 35 frames. The `displace` fx carries the length (`dur`): the client's slide takes as long (render/units.js slideTo). PR #392's 0.6387·√tiles (from the page's μ = 0.5 单位假设) and iLine truncated
+| `displace(enemy, {x, y}, tiles, {dur})` | the raw mover behind push / pull: along passable tiles (by `motion`: FLY in the rect, else ground-passable), no weight rule, no 失衡 state of its own (push / pull add it; `dur` only rides on its `displace` fx); nothing for 失衡免疫 (`noDisplace`: 近地悬浮 incl. 喷气人's 飞行模式, 浮空, the 胄 parts, 守墓石像's statue and flight), 静态刚体 (`def.staticBody`) and leaders (`_displaceable`); unblocks + re-paths |
 | `after(seconds, fn, {owner})`, `every(seconds, fn, {owner, immediate})` | return `{cancel()}`; fn(battle, sched) |
-| `fx(kind, params)`, `rng()` (+ `rng.int/range/chance/pick/shuffle/weighted`) | never use Math.random |
+| `fx(kind, params)`, `rng()` (+ `rng.int/range/chance/pick/shuffle/weighted`) | never use Math.random; for distances, angles and powers use `hypot`, `sin`, `cos`, `atan2`, `powi` from `server/sim/detmath.js`, never `Math.hypot` / `sin` / `cos` / `atan2` / `pow` / `**` (their last bits differ between browsers) |
 | `forceAttack(unit, targets?, { noAmmo })` | an immediate attack with the current profile (hooks, attack SP, a running ammo skill's bullet); `noAmmo: true` = spends no bullet and emits no `ammoUsed` (圣约送葬人 extra attack). Returns true when it attacked |
 | `findTacticalPoint(unit)`, `groundPathTiles()` | tactical point (战术点, tactician 援军 / talent tokens): a free (`isReservedTile`) walkable, standable (`canStand`: never the 深水区) tile of the initial range **on an enemy ground path first** (`groundPathTiles`: grid paths of every non-FLY route, cached per grid version), then nearest (Chebyshev, rows break ties), then the lowest tile key |
 | `effectiveProfile(unit)`, `reduceElement(unit, amount, el?)` | |
@@ -869,7 +928,9 @@ instance) and skip `'counter'` / `'reflect'` damage. When the guard trips, the l
  PRTS 异常效果 晕眩 names no SP effect — community report #18). Attack-type SP: attacks made
  by the skill (the pending "next attack" of an instant/charge skill, every shot of a timed skill including the one that
  ends it) recover nothing, so a cost-N skill fires every **N+1** attacks (AK). Charges (`maxChargeTime > 1`):
-  SP fills to cost → +1 charge (SP restarts) until charges are full (then SP stays full).
+  SP fills to cost → +1 charge (SP restarts) until charges are full (then SP stays full). SP within 1e-9 of the cost is
+  the cost (`gainSp`, `_normalize`; a tiny negative remainder is 0): the float sum of the 1/30-SP time gains can stop
+  2.5e-14 short of it — a 10-SP skill at 1 SP/s took 301 ticks, now 300 (PR #402, 0.2.2).
 - Triggers (`skill.trigger.rule` in data — the official 技能策略, PRTS 卫戍协议/帮助 §作战阶段 技能操作, resolved by
  `tools/build-data.mjs resolveTrigger`: charId rows by skill index; the class rows (重装 / 执旗手 / 战术家 / 吟游者 / 解放者 /
  阵法术师) for **every MANUAL skill** of the class; SKILL_RANGE for a MANUAL skill with a 技能范围 of its own; ACTIVE_RANGE
@@ -909,13 +970,22 @@ instance) and skip `'counter'` / `'reflect'` damage. When the guard trips, the l
  `GDGLOW_SKILL_2` — "全场存在可选目标时释放技能": a targetable enemy anywhere (a heal skill: an ally that needs healing),
  every tick; `NEVER` (alias `MANUAL`) — never auto-casts: the kit calls `unit.skill.activate(reason)` itself;
  `MLYSS_WTRMAN` and any unknown rule behave like DEFAULT. Units that never attack (bard, phalanx, librator) check
-  DEFAULT every tick. Silence blocks activation. The automatic operations cool down (`constants.js AUTO_OP_COOLDOWN` 3 s,
+  DEFAULT every tick. In every enemy condition above a registered ally target (`setAllyTarget`: 白铁's 铁钳号·原型机, an
+  enemy-camp summon) counts like an enemy for a skill that acts on it (`_allyTargetIn`, `allyTargetsOk` — the owner's rule
+  of 2026-10-08: with the device alone in range such a skill opens and spends itself, its bullets included, on it): one
+  acting through the unit's own attacks, or flagged `allyTargets` by its kit; [ASSUMED] one that picks enemies only waits.
+  Silence blocks activation. The automatic operations cool down (`constants.js AUTO_OP_COOLDOWN` 3 s,
  "自动操作具有3s冷却，在完成一次操作或作战开始时部署的单位将进入冷却"): a MANUAL skill is auto-cast no sooner than 3 s
  after its previous cast (a charged skill spends its charges 3 s apart) or after its battle-start deployment
  (`Battle._deploy` initial; `battle.flags.startOpCooldown`, default 3 — the test harness sets 0 unless told otherwise);
  AUTO skills are exempt; kits with their own automatic cast of a MANUAL skill check `skill.opCooling` (波登可 S1, 雪猎
  special bullets, 流形 copy). While a cast "next attack" (instant / charges with an attack override) waits for its
-  attack, no further charge is cast. `gainSp` is ignored while a duration/ammo/toggle skill
+  attack, no further charge is cast. An instant / charges skill with no attack override of its own (a throw, a heal, a
+  buff, a DP gain …) is cast from the tick (the rules above, the DEFAULT of units that never attack, content trigger
+  ranges) at most once per attack interval of its unit — `skill.nextCastAt`, set by such a cast, reset on deployment —:
+  attack speed paces it, and an SP refund that refills the bar at once (迅捷) no longer recasts it every tick (GitHub #298,
+  引星棘刺 S1; [ASSUMED] one attack interval: PRTS prints no 前后摇 for such casts). The cast is no attack: the unit's
+  attacks and heals keep their own rhythm, and its first cast comes as soon as it is ready (#124). `gainSp` is ignored while a duration/ammo/toggle skill
  runs (its bar shows the skill), whatever the reason, and — any reason but `'init'` — while the unit has the `noSp`
  flag (阻回: "停止并阻止任意形式的技力回复"; the operators' 凋亡 burst, §3).
 - Kinds: `duration` (mods for `duration` s), `ammo` (mods until `ammo` attacks were made, optional duration cap),
@@ -1292,7 +1362,9 @@ Unknown subprofessions fall back to the profession default (test `professions.te

 ## 9. Wire format (snapshot.js, DESIGN §8.2)

- `snapshot()` → `{ fieldId, t, units: [[id, x, y, hp, maxHp, sp, spMax, flags, anim]], dp, killed, total, dps?, boss?, down?, elem? }`.
+- `snapshot()` → `{ fieldId, t, units: [[id, x, y, hp, maxHp, sp, spMax, flags, anim]], dp, killed, total, resolved, dps?, boss?, down?, elem?, ammo?, wolves?, neg?, stand?, standCut? }`.
+  `killed` counts every `counted` knock-out (§1.3) and `total` only the enemies this field scheduled (`inTotal`), so it may
+  exceed it; `resolved = min(total, killedInTotal + leakedInTotal)` is the HUD capsule's numerator (DESIGN §14 顶栏胶囊).
  `sp/spMax` show remaining duration/ammo as a draining bar while a timed skill is active (ammo: `ammoLeft / ammoMax`, the
  activation's real total — 拉特兰's and 逃犯引渡手续's extra bullets included, community report #35). Units in DIE state stay 0.8 s.
  Active finite zero-SP duration skills display `timeLeft/duration` using `duration` as `spMax`; after end they show
@@ -1303,8 +1375,25 @@ Unknown subprofessions fall back to the profession default (test `professions.te
  respawn timer; `state` = constants.js
  `DOWN_STATE`: 0 counting, 1 timer done / DP short, 2 timer done / its tile taken — a safeguard: no ally deploys on a
  knocked-out operator's tile, §1; `row, col` = the tile it lies on and comes back on, `unit.body`); `elem: [[id, element, fill,
-  cooldownEnd, cooldown]]` (only when non-empty) = `elementView` of every unit with a gauge or a running 爆发冷却 (§3).
+  cooldownEnd, cooldown]]` (only when non-empty) = `elementView` of every unit with a gauge or a running 爆发冷却 (§3);
+  `ammo: [[id, left, magazine]]`, `wolves: [[id, left, max]]` and `neg: [[id, fill]]` (each only when non-empty; display only) = the HP-bar
+  readouts of `snapshot.js` `ammoView` (an ally's running ammo skill: whole rounds, the magazine = `ammoMax`), `wolfView` (伺夜's 狼群:
+  `mem.shadows` of `mem.wolfCapacity`) and `negView` (`unit.negFill()`, set by 斩业星熊's T1 业火: the 我执 pool as a share of its cap);
+  UnitInfo `ammoSkill` marks an ally whose skill is an ammo magazine (DESIGN §8.2).
  `fieldMeta()` lists the knocked-out operators too (a client joining mid-battle shows them; DESIGN §18.3).
+  `stand: [[id, until]]` (only when non-empty) = the game time each enemy's attack recovery ends (`atkStandUntil`: it
+  stands for the rest of its attack clip, §1.2 `attackStand`) — enemies alive, deployed, visible, neither feared nor stunned;
+  display metadata the battle never reads (PR #381). When `until` falls between two snapshots the renderer holds the older
+  position until then and interpolates the rest of the interval (HP / SP keep the whole interval); a newer snapshot that
+  moved before `until` is interpolated linearly; while the newest snapshot still holds a stand nothing is extrapolated
+  (the next one confirms the move). A snapshot without the field interpolates over the whole interval, a teleport or a
+  redeploy still snaps first, and nothing is inferred from the attack animation. `standCut: [[id, at]]` (only when
+  non-empty) = the latest game time each enemy's recovery was cut or ignored (a displacement — 失衡 —, fear, a stun, hiding,
+  leaving the field), one per enemy: an interval whose newer snapshot carries a cut inside it is interpolated linearly even
+  when the older stand ended inside it too, and an older cut leaves a later stand alone. It covers the enemies listed in
+  `units` (the 0.8 s death window included), never a hidden or unlisted one. The newest snapshot showing a death, a stun
+  (freeze, sleep, 浮空 included) or a block is not extrapolated until a later snapshot shows it moving again; the
+  extrapolated velocity uses the same moving span as the interpolation (the stand is not in its denominator).
 - `drainEvents()` tuples: `['spawn', UnitInfo]` (first appearance), `['deploy', id]` (every (re)deploy and 【移动】 — the client plays the deploy and its interpolation snaps instead of sliding), `['atk', src, tgt, projKind]`
  (`none|arrow|bolt|bomb|lob|orb|drone|enemy|boomerang|droneBomb|chain|chainHeal`; a boomerang's way back has no event — the
  renderer flies it back to the thrower at `BOOMERANG_RETURN_SPEED`; an enemy's `profile.shot` may name another kind,
@@ -1314,7 +1403,7 @@ Unknown subprofessions fall back to the profession default (test `professions.te
  `['fx', kind, x, y, extra]` (`hitCap` `{ id, n }`: a leader's hit cancelled by 限伤 — the renderer draws nothing;
  `extra.form` = the unit's model form from then on — an enemy's `content/enemies/helpers.js setForm`, a 傀儡师's 替身 — `shared/protocol.js fxForm`),
  `['layer', playerId, bondId, n]` (n = the layers actually added, capped at 999), `['bounty', playerId, coins]`.
- `UnitInfo = { id, kind, side, ownerId, defId, name, tier, golden, spine, avatar, x, y, facing, dir, maxHp, motion?, boss?, uid?, form?, skillIndex?, moduleId?, items?, standInFor?, diy? }` (`standInFor`: a 补位 stand-in's replaced operator charId — `spine` / `avatar` / `name` are the stand-in's, DATA.md §18; the prep scouting units of `Match.prepFieldMeta` carry it too, board and held pieces alike with the stand-in's art (the owner's recall of the official mode, 2026-10-06), and so do the m.result lineups; `diy`: a 自选 piece's pick `{ charId, skillIndex, uniEquipId }` — `defId` is its slot, `spine` / `avatar` / `name` the operator's; the prep scouting units of `Match.prepFieldMeta` and the m.result lineups carry it too (server/match/player/diy.js, the client composes the operator's card from it); `skillIndex`: an ally's equipped skill, DESIGN §16; `form`: the unit's current model form — `content/enemies/helpers.js setForm`: 掠海漂移体 `'crawl'`, 暴鸰 `'bombed'` after its drop, 转译基底·α's forms …; a 傀儡师 fighting as its 替身 `'doll'` — so a view built mid-battle from `fieldMeta()` starts on that clip set; `items`: an ally operator's equipped item ids — a 变形同构体 wearer is a member of the bond it grants on the client too, the bond popup and the detail card's chips)
+- `UnitInfo = { id, kind, side, ownerId, defId, name, tier, golden, spine, avatar, x, y, facing, dir, maxHp, motion?, boss?, uid?, form?, skillIndex?, moduleId?, items?, standInFor?, diy?, potential?, cultivate? }` (`potential` / `cultivate`: an ally operator's potential below 6 and its 练度 tier, 0.2.2 — a teammate's card composes its numbers from them) (`standInFor`: a 补位 stand-in's replaced operator charId — `spine` / `avatar` / `name` are the stand-in's, DATA.md §18; the prep scouting units of `Match.prepFieldMeta` carry it too, board and held pieces alike with the stand-in's art (the owner's recall of the official mode, 2026-10-06), and so do the m.result lineups; `diy`: a 自选 piece's pick `{ charId, skillIndex, uniEquipId }` — `defId` is its slot, `spine` / `avatar` / `name` the operator's; the prep scouting units of `Match.prepFieldMeta` and the m.result lineups carry it too (server/match/player/diy.js, the client composes the operator's card from it); `skillIndex`: an ally's equipped skill, DESIGN §16; `form`: the unit's current model form — `content/enemies/helpers.js setForm`: 掠海漂移体 `'crawl'`, 暴鸰 `'bombed'` after its drop, 转译基底·α's forms …; a 傀儡师 fighting as its 替身 `'doll'` — so a view built mid-battle from `fieldMeta()` starts on that clip set; `items`: an ally operator's equipped item ids — a 变形同构体 wearer is a member of the bond it grants on the client too, the bond popup and the detail card's chips)
  (`dir` = the unit direction, allies meaningful, enemies 'RIGHT'; `facing` = its horizontal sign for sprite flipping)
  (`spine`/`avatar` are asset ids from data).
 - flags: UF bits (blocked 1, stunned 2, frozen 4, stealth 8 — 隐匿 (an enemy's only while not blocked / revealed and not within 3 s of a block's end) or an ally's 迷彩 — skill 16, shield 32, invuln 64, cold 128, sleep 256, flying 512);
@@ -1401,8 +1490,14 @@ never displaced, §6), tags, abilities, skills, talent, raw`. Tokens use `varian
 `skcom_withdraw` token skills are ignored.
 Operator loadouts (DESIGN §16, DATA.md §2.2 / §14): `getChess(id, { skillIndex, moduleId })` = the def with the selected
 skill (`def.skill`, its bb / SP / trigger) and, for elites, the module's stats / trait / talents; `def.loadout` = the
-resolved loadout `{ skillIndex, moduleId, skillIsDefault, moduleIsDefault, isDefault }` (illegal choices fall back to the
-default; cached per loadout; the default loadout is the same object as `getChess(id)`). `getToken(id, ownerChessId,
+resolved loadout `{ skillIndex, moduleId, potential, skillIsDefault, moduleIsDefault, potentialIsDefault, isDefault }`
+(illegal choices fall back to the default; cached per loadout; the default loadout is the same object as `getChess(id)`).
+Potential (0.2.2, DATA.md §2.3, shared/potential.js): a loadout's `potential` (潜能 1–6; default 6 = the data's full
+potential) composes the record at it first (`loadoutRecord` → `atPotential`: stats, statsBase, talents, module talent
+changes); `getToken` composes the owner variant at the owner's `loadout.potential` (talents; a 自选 summon's deploy
+limit / holding / count) and `getDiy(slotId, pick, potential)` an owned pick's form (a prototype has none: `diyProto`);
+the kits that read a summon variant raw (content/tokens.js `ownVariant`, `kits/shared/summoner.js tokenStat`, 多萝西,
+麦克尼斯特) compose it the same way. `getToken(id, ownerChessId,
 ownerLoadout)` = the summon for that owner loadout (`bySkill` / `byModule` merged; `def.sources` ⊆ ['talent', 'skill',
 'display'] and `def.count` of that loadout — `[]` / `['display']`: the owner does not make it). Every Battle gets a
 per-battle view (`withUnitLoadouts`: id-only `getChess(id)` / `getToken(id, owner)` use the loadout the inputs give that
@@ -1414,7 +1509,17 @@ the source's `backups` (data/backups.json; `rawBackups()` follows the fallback c
 the replaced operator's, `loadout` the backup selection (`standIn: true`); its skill / module ignore the loadout's
 ("对于补位干员其技能不可更改", shared/standIn.js). A PRESET / DIY chess or a source without backups (a browser that did not
 fetch backups.json) gives the chess's own def. A PlayerBattleInput entry carries it as `standIn: true` (spec.js keeps
-nothing else); the per-battle view maps id-only lookups of that chess id to the stand-in.
+nothing else — no `potential` / `cultivate`: a stand-in has neither); the per-battle view maps id-only lookups of that
+chess id to the stand-in.
+潜能 / 练度 in a PlayerBattleInput operator entry (0.2.2; the owner's decision of 2026-10-08 — the player's 干员调配 settings,
+PlayerState.battleInput states both for every owned operator, bots the defaults 6 / 3): `potential` (1–6) → the def of
+`getChess(chessId, { …, potential })`; `cultivate` (0–3) → `unit.cultivate` and `unit.cultMul` = `data.cultivateMul(n)`,
+the effects.json CHAR_MAP `aceffect_char_<n+1>` `char_attribute_mul` (×ATK / ×DEF / ×max HP: 1 / 1.05·1.05·1 / 1.1·1.05·1.05
+/ 1.1 each), a multiplier of its own in the Πmul of §2 (never summed with the 直接乘算 percentages: PRTS 盟约记录
+「…与自持有干员的属性加成独立」); a stand-in or a prototype 自选 def never takes it, summons neither [ASSUMED: the official
+buff names the operator]. spec.js keeps each only when well-formed (never on a stand-in or a summon); absent = full
+potential and no 练度 (a raw unit: tests, tools). `unitStatsEntry` counts the 练度 in the unit's own numbers (`base`), as
+the record cards do; UnitInfo carries `potential` (below 6) and `cultivate`.
 自选 (DATA.md §18): `getChess(slotId, { diy })` = `getDiy(slotId, pick)` — shared/diy.js `diyRecordOf` of the DIY slot
 record (`_a` normal, `_b` elite) with the pick over this source, normalised: `id` / `baseId` / `golden` / `tier` of the
 slot, `bonds` the pick's derived ones, `raw.garrisonIds` `[]` (no 特质), every combat field of the operator at the slot's
@@ -1427,8 +1532,8 @@ id, `loadout` the selection with `diy` = the checked pick `{ charId, skillIndex,
 `diy` (spec.js keeps `charId` / `skillIndex` / `uniEquipId` when well-formed); Battle fields a DIY slot only with a legal
 pick (none ⇒ no unit), and each unit resolves its own pick — two players filling one slot id differently are exact.
 Assumptions taken here (documented choices): element burst numbers for necrosis/apoptosis; unspawned enemies at timeout
-are dropped (not leaks); geek drain is non-lethal; tactician reinforcement stats; displacement is instantaneous (the official 失衡 slide over ≈ 0.4–1.2 s is not
-modelled — only its distance, `push` / `pull`); pushes PRTS does not classify use the 弹道 distance column (`PUSH_TILES`);
+are dropped (not leaks); geek drain is non-lethal; tactician reinforcement stats; displacement is instantaneous, followed by the 失衡 hold (push: the 位移时间 table; pull: the force
+window — `push` / `pull`; a wall-shortened slide keeps its row time, a wall at the first step gets the 0.1 s floor); pushes PRTS does not classify use the 弹道 distance column (`PUSH_TILES`);
 platforms/mounds are ground obstacles that elevate operators; the generic kit maps 凋亡 to `apoptosis` (侵蚀 has no element key
 yet — enemy content must pick one); 抵抗 covers the control statuses of `RESIST_STATUSES` (the official term lists 晕眩/寒冷/
 冻结/恐惧/诱导…, the rest follow the operator kits that grant it) and caps at 0.95; tactical points prefer enemy path tiles.
@@ -22,14 +22,19 @@ view.on('pieceDragStart'|'pieceDrop'|'pieceDragEnd'|'pieceClick'|'pieceHover'|'t
   // area 'outside' lets the DOM UI decide: released off the grid or over DOM covering the canvas (the shop bar), mouse
   //   and touch alike — there is no drag-to-sell, the piece goes back
   // the drop target, its highlight and tileHover while dragging = the tile under the POINTER; an item dropped on a
-   //   tile equips the unit standing on it (§18.1)
+   //   tile equips the unit standing on it (§18.1); a dragged unit is drawn standing on that tile while it is a legal
+   //   target — a board tile or a bench slot (render/drag.js dragStandTile), the official deploy drag — and held under
+   //   the pointer elsewhere (DRAG_HOLD_TILES), an item plate on the pointer (§27.62)
 view.pieceScreenRect(uid)                    // the drawn body's rect { left, top, right, bottom, width, height, x, y } (tooltips / overlays)
 view.tileScreen(row, col)                    // { x, y, s, poly } client px of a tile top (direction wheel geometry)
 view.holdPiece(uid, {row,col}|null)          // keep a dropped piece on a tile while its direction is chosen
 view.setPieceDir(uid, dir)                   // show a prep piece facing UP|RIGHT|DOWN|LEFT (wheel preview / stored dir)
 view.resize() ; view.destroy()
 ```
-Visuals: oblique perspective board (tilted ~30°, trapezoid tiles), raised highland blocks with side faces, road/floor/forbidden styles per research 07 §7, pulsing red gates, blue objective, dashed hand pads. Units: Spine (Front/Back rule, feet anchored — flying units hover `FLY_HOVER` 1.3 tiles up, the client's 0.35 in character space, research 12, §25.13.5 — an enemy flyer from the road whatever tile it crosses (GitHub #277) —, shadow sprite on the ground), elite gold outline glow, tier chip, HP bar (green ally / red enemy), SP bar (skill ready glow), status icons, the element gauge row under the bars — icon + white bar of the remaining 元素值, element-coloured refill during a 爆发冷却 (§18.3, §20.8); enemies drawn × their official prefab factor (enemies.json `modelScale`, §20.4); an enemy's mode — the `form` of a sim fx (`shared/protocol.js fxForm`), or UnitInfo `form` for a view built later through `render/app/info.js renderInfo` — switches clip sets (`render/units.js FORMS`: 掠海漂移体 crawl, 暴鸰 bombed, 转译基底's forms, the 逐火 embers, the leaders' 重生, 守墓石像, the 孤岛风云 prisoners' warning and 解放; closing clips timed from the fx `dur`), a one-off cast (`PROJ[kind].once`) plays once, and a local-only model's web alias is tinted (`ALIAS_TINT`: 灼热 / 炽焰源石虫, §21.14); damage numbers (phys orange-white, arts purple, true white, heal green); projectiles drawn per kind, hit sparks and melee slashes, skill activation bursts and auras (§17.3); death fade for enemies, summons and devices — a knocked-out operator stays on its tile in its held Die pose under a redeploy ring (§18.3); leak flash at the objective. Fallback when Spine missing/failed: avatar in a rarity-coloured diamond with bob/lunge tweens — a failed or timed-out model is loaded again (bounded), and a view re-resolves its model and avatar when the asset manifest arrives late or the tab is shown again (§21.27). 60 fps target with ≤ 120 units. Picking (which unit a press / hover / dropped item is on) and the drag target follow one rule, `render/pick.js` — the tile under the pointer (§18.1); the Spine store never hands out a skeleton whose unload is in flight (§17.1).
+
+`createFieldView(host, { …, signal })` takes an `AbortSignal` (0.2.2, GitHub PR #302 by @siruimei07): an abort during the build — the Pixi load, the manifest, the fonts, the board art and the 3D board waits — destroys what the attempt made (the canvas, the ticker, the listeners, the observers) and rejects, a failure while it is built releases the same partial resources, and `view.destroy()` is idempotent. `ui/fieldHost.js mountFieldView(host, { signal })` aborts that signal when the engine fails, times out (12 s) or the owner unmounts, disposes a view a factory returns after that, and never clears the other children of `host` — the simplified view and a newer mount share it.
+
+Visuals: oblique perspective board (tilted ~30°, trapezoid tiles), raised highland blocks with side faces, road/floor/forbidden styles per research 07 §7, pulsing red gates, blue objective, dashed hand pads. Units: Spine (Front/Back rule, feet anchored — flying units hover `FLY_HOVER` 1.3 tiles up, the client's 0.35 in character space, research 12, §25.13.5 — an enemy flyer from the road whatever tile it crosses (GitHub #277) —, shadow sprite on the ground), elite gold outline glow, tier chip, HP bar (green ally / red enemy), SP bar (skill ready glow; while an ammo skill runs it is the magazine instead — a yellow cell per round, emptying from the right, `b.snap ammo`, and the skill draws no standing aura, UnitInfo `ammoSkill`), 斩业星熊's 我执 pool as a red bar over her drained HP bar (`b.snap neg`), 伺夜's 狼群 count as a row of diamonds under the bars, the lit ones the 狼影 left (`b.snap wolves`; 0.2.2, GitHub PR #303 / #378 / #383), status icons, the element gauge row under the bars — icon + white bar of the remaining 元素值, element-coloured refill during a 爆发冷却 (§18.3, §20.8); enemies drawn × their official prefab factor (enemies.json `modelScale`, §20.4); an enemy's mode — the `form` of a sim fx (`shared/protocol.js fxForm`), or UnitInfo `form` for a view built later through `render/app/info.js renderInfo` — switches clip sets (`render/units.js FORMS`: 掠海漂移体 crawl, 暴鸰 bombed, 转译基底's forms, the 逐火 embers, the leaders' 重生, 守墓石像, the 孤岛风云 prisoners' warning and 解放; closing clips timed from the fx `dur`), a one-off cast (`PROJ[kind].once`) plays once, and a local-only model's web alias is tinted (`ALIAS_TINT`: 灼热 / 炽焰源石虫, §21.14); damage numbers (phys orange-white, arts purple, true white, heal green); projectiles drawn per kind, hit sparks and melee slashes, skill activation bursts and auras (§17.3); death fade for enemies, summons and devices — a knocked-out operator stays on its tile in its held Die pose under a redeploy ring (§18.3); leak flash at the objective. Fallback when Spine missing/failed: avatar in a rarity-coloured diamond with bob/lunge tweens — a failed or timed-out model is loaded again (bounded), and a view re-resolves its model and avatar when the asset manifest arrives late or the tab is shown again (§21.27). 60 fps target with ≤ 120 units (a crowd draws its Spine units through the cached atlas, `render/impostor.js`, each refreshed every k-th frame: the refresh turns follow this frame's update order — `ctx.impostorSlot()`, a counter reset at the start of every frame — so a frame refreshes ⌊n/k⌋ or ⌈n/k⌉ of n units instead of whatever random per-unit phases gave, and a unit not refreshed for 2k frames is refreshed regardless; GitHub PR #334 by @Cloudnyco, issue #327 step 1 — its step 2, a time budget by priority, is not done). Picking (which unit a press / hover / dropped item is on) and the drag target follow one rule, `render/pick.js` — the tile under the pointer (§18.1); the Spine store never hands out a skeleton whose unload is in flight (§17.1).

 ---

@@ -37,15 +42,15 @@ Visuals: oblique perspective board (tilted ~30°, trapezoid tiles), raised highl

 Look & feel: research 06 §11–§12 (near-black green-grey panels, mint `#4ed8af` accents, gold prices, bracket frames, hexagon badges, bond discs, 7-segment countdown). Fonts: Noto Sans SC (Google Fonts with local fallback), Bender/Novecento self-hosted for numerals and Latin micro-labels.

-Screens: **Title** (season art backdrop, 开始 → nickname), **Lobby** (独立模拟 / 同盟模拟 → create or join; recent rooms), **Room** (4 seats, invite code + copy link, difficulty picker (host), add/remove AI, ready, start), **Briefing** (INFO_CHECK: stage, factions, disabled bonds, banned chess, boss silhouette, ready x/4), **Band draft** (grid of 40 bands with icon/name/LP/effect; turn order list; skip; 查看禁用盟约与干员 under the order list opens the briefing's bond rows, legend and banned operators again in a read-only dialog — one component, `ui/matchInfo.js`, §21.28), **Game** (Pixi field + DOM HUD: top bar with round/phase capsule/LP (live during the own battle, §17.5)/timer/ready; bond strip + popups; left team panel with LP/status/emote bubbles, click to watch (a boss round frames the viewer's pair in green, §15); bottom shop bar (level card, 3–5 chess cards, item card, funds, refresh, freeze, collapse); detail panel on click/hover (header with bonds, then 特质 — the operator's own effect — right under it, 特性, stats (live: the battle's own numbers / the start-of-battle ones in prep, §18.5), skill, module, equipped items, talents: §17.5); enemy preview drawer (next round enemies with counts, factions; its 本局信息 tab lists the briefing's disabled bonds and banned operators, §21.28); 机变 overlay (the official card: framed icon + name, the effect text — data `descRaw` — under it, the taker's round avatar at the top-right; two taps, §18.2, §20.7); merge reward overlay (the promotion reward, or a special offer — 凯瑟琳's 定向投放 … — as item / chess cards under its source's label, with the count of offers queued behind it, §21.2); phase banners; the round's result box at SETTLE (the official 「作战结束」 + 「全员无伤！」 / 「生命值减少 −N」 for the viewer's own LP — after a 联防 the server's charge, `m.public.uniteResult.losses` — with the matching BATTLEOVER sound; none in the boss rounds or for a viewer not in the round; GitHub #235, PR #112); ticker (by broadcast priority, highest first; a leader-damage line plays only in its round and a player's newer one replaces the older, §21.10); toasts; emote wheel; settings (volume, quality, show damage numbers)), **Result** (victory/defeat, rounds, per-player lineup, stats, titles, back to room).
+Screens: **Title** (season art backdrop, 开始 → nickname), **Lobby** (独立模拟 / 同盟模拟 → create or join; recent rooms; a 设置 button left of 玩法说明 opens the title screen's settings dialog, GitHub #238), **Room** (4 seats, invite code + copy link, difficulty picker (host), the co-op 「AI 队友最后选择」 switch (host, off by default; read-only for the others), add/remove AI, ready, start; the same 设置 button), **Briefing** (INFO_CHECK: stage, factions, disabled bonds, banned chess, boss silhouette, ready x/4), **Band draft** (grid of 40 bands with icon/name/LP/effect; turn order list; skip; 查看禁用盟约与干员 under the order list opens the briefing's bond rows, legend and banned operators again in a read-only dialog — one component, `ui/matchInfo.js`, §21.28), **Game** (Pixi field + DOM HUD: top bar with round/phase capsule (the kill counter reads 已解决 / the round's own enemies: a leak counts, a split child or summon is in neither number, §14)/LP (live during the own battle, §17.5)/timer/ready, and beside the latency a 观战席 capsule (eye + count, a roster with the host's ✕) while the room has spectators (GitHub #120); bond strip + popups; left team panel with LP/status/emote bubbles, click to watch (a boss round frames the viewer's pair in green, §15); bottom shop bar (level card, 3–5 chess cards — a slot the upgrade just opened is a bare frame until the next refresh, GitHub #332 —, item card, funds, refresh, freeze, collapse; 准备就绪 folds it and cancelling unfolds it, GitHub #138); detail panel on click/hover (header with bonds, then 特质 — the operator's own effect — right under it, 特性, stats (live: the battle's own numbers / the start-of-battle ones in prep, §18.5), skill, module, equipped items, talents: §17.5); enemy preview drawer (next round enemies with counts, factions; its 本局信息 tab lists the briefing's disabled bonds and banned operators, §21.28); 机变 overlay (the official card: framed icon + name, the effect text — data `descRaw` — under it, the taker's round avatar at the top-right; two taps, §18.2, §20.7); merge reward overlay (the promotion reward, or a special offer — 凯瑟琳's 定向投放 … — as item / chess cards under its source's label, with the count of offers queued behind it, §21.2); phase banners; the round's result box at SETTLE (the official 「作战结束」 + 「全员无伤！」 / 「生命值减少 −N」 for the viewer's own LP — after a 联防 the server's charge, `m.public.uniteResult.losses` — with the matching BATTLEOVER sound; none in the boss rounds or for a viewer not in the round; GitHub #235, PR #112); ticker (by broadcast priority, highest first; a leader-damage line plays only in its round and a player's newer one replaces the older, §21.10); toasts; emote wheel; settings (volume, quality, show damage numbers)), **Result** (victory/defeat, rounds, per-player lineup, stats, titles, back to room).

 盟约栏支持「收起／展开盟约」（GitHub #142）：收起时隐藏完整列表并关闭由列表打开的浮层，让原图标区域的指针事件到达棋盘，只保留至少 44×44 CSS px 的恢复按钮。折叠状态在本局阶段变化、观看队友时保留；展开后仍按当前观看对象显示最新盟约。按钮支持 Enter / Space，聚焦按钮时 Space 不触发准备或暂停。

 A bond the mode never activates (标准: official `inactiveBondIdList`) reads 本局禁用 on shop / reward cards (struck through), detail chips and the popup — no 在场 count, no 当前效果 (§21.7); a strategy built around one carries a 本局禁用 tag and note in the strategy draft and stays selectable (bands.json `bondIds`, §21.26).

-Interactions (research 09 §1.2 / §5, which supersede the first draft): shop card, upgrade and 机变 card are two-tap (select → confirm, §18.2) — a tap anywhere on the card, its confirm strip included, is the card's tap (shop cards have no ⓘ corner; the armed bar sits above the detail card; a shop card that cannot be bought right now opens its detail on tap, §20.7); a press / drop goes by the tile under the pointer (§18.1); drag bench↔board — a legal board drop (also onto the piece's own tile) opens the 4-direction deploy wheel (swipe UP/RIGHT/DOWN/LEFT with the rotated range striped orange; release in the centre, ✕ 点击取消 or Esc cancels); tap a unit's tile ⇒ underframe with **出售 +N** (and **撤退** on the board; items/Arts in the hand / temp: **销毁** — equipped items are locked, §6.2) plus its rotated range; no drag-to-sell; drag an item onto a unit's tile ⇒ equip that unit (both slots used ⇒ the replace dialog `ui/equipReplace.js` shows the incoming and the two equipped items; the player picks the one to replace — it is destroyed — `g.equip replaceUid`; cancel changes nothing); Arts are dropped on a tile and get the direction step; right-click or long-press ⇒ detail (or a tap on a shop card while the bar is locked); keyboard shortcuts — the defaults, which each player rebinds in 设置 → 快捷键 (`ui/gameLogic/shortcuts.js`, saved with the settings; §25.15) — `R` refresh, `F` freeze, `D` level-up, `Q` retreat / `X` sell the selected operator, `Space` ready (in a solo battle: pause / resume, §14 Solo pause); `Esc` closes (fixed); none acts while a text field has the focus or a dialog is open. All actions optimistic-free: UI waits for server state (latency is small).
+Interactions (research 09 §1.2 / §5, which supersede the first draft): shop card, upgrade and 机变 card are two-tap (select → confirm, §18.2) — a tap anywhere on the card, its confirm strip included, is the card's tap (shop cards have no ⓘ corner; the armed bar sits above the detail card; a shop card that cannot be bought right now opens its detail on tap, §20.7); a press / drop goes by the tile under the pointer (§18.1); drag bench↔board — over a legal tile the dragged unit stands on it, as in the official deploy drag, held under the pointer elsewhere (§27.62); a legal board drop (also onto the piece's own tile) opens the 4-direction deploy wheel (swipe UP/RIGHT/DOWN/LEFT with the rotated range striped orange; release in the centre, ✕ 点击取消 or Esc cancels); tap a unit's tile ⇒ underframe with **出售 +N** (and **撤退** on the board; items/Arts in the hand / temp: **销毁** — equipped items are locked, §6.2) plus its rotated range — the range follows the open detail card (`screens/game/range.js inspectRange`, GitHub PR #281): an own board piece in prep, and in battle or on a scouted board any ally operator / summon on the field with its current range (the live stats' `range` and `dir`, the snapshot's tile), never an enemy or a bench unit (UnitInfo `area` hand / temp); drag, the wheel and the pen hide it; no drag-to-sell; drag an item onto a unit's tile ⇒ equip that unit (both slots used ⇒ the replace dialog `ui/equipReplace.js` shows the incoming and the two equipped items; the player picks the one to replace — it is destroyed — `g.equip replaceUid`; cancel changes nothing); Arts are dropped on a tile and get the direction step; right-click or long-press ⇒ detail (or a tap on a shop card while the bar is locked); keyboard shortcuts — the defaults, which each player rebinds in 设置 → 快捷键 (`ui/gameLogic/shortcuts.js`, saved with the settings; §25.15) — `R` refresh, `F` freeze, `D` level-up, `Q` retreat / `X` sell the selected operator, `Space` ready (in a solo battle: pause / resume, §14 Solo pause); `Esc` closes (fixed); none acts while a text field has the focus or a dialog is open. All actions optimistic-free: UI waits for server state (latency is small).

-Audio: autochess BGM per phase, UI SFX (buy/sell/refresh/level/merge/ready/timer), per-unit attack/hit SFX (throttled; an impact plays for a hostile attack's phys / arts / true damage only, never an operator's skill-mode file for a normal attack, ≤ 2 copies of one battle sound — §18.4), boss music in R14/15.
+Audio: autochess BGM per phase, UI SFX (buy/sell/refresh/level/merge/ready/timer), per-unit attack/hit SFX (throttled; an impact plays for a hostile attack's phys / arts / true damage only, never an operator's skill-mode file for a normal attack, ≤ 2 copies of one battle sound — §18.4; a skill cue that reaches the client before its unit — the sim emits a deploy-tick cast ahead of the unit's `['spawn']` — is held and sounds once when the unit is tracked, `AudioManager.pendingSkill`, GitHub PR #292 by @LimitlessHPPK; entering a field late hands the sound the buffered `spawn` tuples only, `screens/game/early.js audioEarly`, so no old death, deploy or cast replays), boss music in R14/15.

 ---

@@ -113,16 +113,16 @@ Base stats (from data): `maxHp, atk, def, res` (0–100), `aspd` (100 base), `ba

 DamageInfo: `{ amount (pre-mitigation), type: 'phys'|'arts'|'true'|'element', element?: 'burn'|'necrosis'|'neural'|'apoptosis', atkScale, defIgnoreFlat, defIgnorePct, resIgnoreFlat, resIgnorePct, mul (multiplier), canDodge, isSkill, isSplash, ignoreSelect (no selection — a 无视无法选择 ability, a direct pick, a debuff's tick: it reaches an airborne 起飞 ally, §21.22), tags[] }`.

-Engine helpers (content must use these, never touch internals): `battle.dealDamage(source, target, dmgInfo)`, `battle.heal(source, target, amount, {overheal=false})`, `battle.applyStatus(target, key, { duration, source, value })`, `battle.addBuff(unit, buff)`, `battle.removeBuff(unit, key)`, `battle.spawnToken(ownerUnit|playerId, tokenId, row, col, opts)`, `battle.spawnEnemy(enemyKey, { routeIndex, pos, mods, tag })`, `battle.addProjectile({ from, to|target, speed, onHit, visual })`, `battle.push(e, force, { from, dir, fixed, fixedAngle, inward, effect })` / `battle.pull(e, force, { to, center, stop })` / `battle.pullToFront(e, unit, force)` / `battle.forceLevel(e, force)` / `battle.pushDistance(e, force, { effect })` (the official 力度 − 重量 tables, §20.3; a 静态刚体 — data `staticBody` — is never moved, §21.9; `battle.displace(e, dir, distance)` is the raw mover), `battle.allySelectable(a, by)` / `battle.alliesFor(by, ownerId)` (friendly selection, skipping 孤立 units, §20.3), `battle.unitsInGrid(unit, grid)`, `battle.enemiesInRadius(x, y, r)` (every enemy: enemy-side auras / heals, collisions), `battle.foesInRadius(x, y, r)` (the enemies an operator's area effect can select — no untargetable, no unblocked unrevealed 隐匿 enemy; PRTS 作战机制 §AOE伤害判定, §21.19), `battle.alliesInRadius(x, y, r, ownerId?)`, `battle.addLayers(playerId, bondId, n, reason)` (no-op when gains disabled; at most the room left under `BOND_LAYER_CAP` 999 — `shared/constants.js layerGainRoom`, §20.12; records into result), `battle.loseHp(target, amount, { source, from, sourceless })` (流失 — it skips the damage events: no 受击回复, TAKE_DAMAGE, shield or counter, `damage.js isHpLoss`; only for what the official calls 流失 — "N 真实伤害 per second" effects such as 源石溶剂 are `periodicDamage` damage, §21.12; a 物理流失 is first reduced by the target's DEF, 阿戈尔's devour, §21.19; `sourceless` = 无来源, `source` keeps the credit — leader parts pass damage on 1:1 with it, §20.10), `battle.addCoins(playerId, n)`, `battle.fx(kind, params)` (client VFX event), `battle.rng()` (never `Math.random`), `battle.getPlayer(playerId)` (bonds/band/effects view), `battle.redeploy(unit)` (on its rest tile: where a knocked-out operator lies, else its home; `battle.isReservedTile(r, c)` = a tile no automatic placement takes, `battle.downOn(r, c)` = the knocked-out operator lying there, whose tile no ally deploys or moves onto — §21.24), `battle.after(seconds, fn)` (scheduled callbacks), `battle.every(seconds, fn)`.
+Engine helpers (content must use these, never touch internals): `battle.dealDamage(source, target, dmgInfo)`, `battle.heal(source, target, amount, {overheal=false})`, `battle.applyStatus(target, key, { duration, source, value })`, `battle.addBuff(unit, buff)`, `battle.removeBuff(unit, key)`, `battle.spawnToken(ownerUnit|playerId, tokenId, row, col, opts)`, `battle.spawnEnemy(enemyKey, { routeIndex, pos, mods, tag })`, `battle.addProjectile({ from, to|target, speed, onHit, visual })`, `battle.push(e, force, { from, dir, fixed, fixedAngle, inward, effect })` / `battle.pull(e, force, { to, center, stop })` / `battle.pullToFront(e, unit, force)` / `battle.forceLevel(e, force)` / `battle.pushDistance(e, force, { effect })` (the official 力度 − 重量 tables, §20.3; a 静态刚体 — data `staticBody` — is never moved, §21.9; a push / pull then holds the enemy in its 失衡 state — no walking, normal attack or skill — for the PRTS 位移时间 / force window, 0.2.2, SIM.md §6; `battle.displace(e, dir, distance)` is the raw mover), `battle.allySelectable(a, by)` / `battle.alliesFor(by, ownerId)` (friendly selection, skipping 孤立 units, §20.3), `battle.unitsInGrid(unit, grid)`, `battle.enemiesInRadius(x, y, r)` (every enemy: enemy-side auras / heals, collisions), `battle.foesInRadius(x, y, r)` (the enemies an operator's area effect can select — no untargetable, no unblocked unrevealed 隐匿 enemy; PRTS 作战机制 §AOE伤害判定, §21.19), `battle.alliesInRadius(x, y, r, ownerId?)`, `battle.addLayers(playerId, bondId, n, reason)` (no-op when gains disabled; at most the room left under `BOND_LAYER_CAP` 999 — `shared/constants.js layerGainRoom`, §20.12; records into result), `battle.loseHp(target, amount, { source, from, sourceless })` (流失 — it skips the damage events: no 受击回复, TAKE_DAMAGE, shield or counter, `damage.js isHpLoss`; only for what the official calls 流失 — "N 真实伤害 per second" effects such as 源石溶剂 are `periodicDamage` damage, §21.12; a 物理流失 is first reduced by the target's DEF, 阿戈尔's devour, §21.19; `sourceless` = 无来源, `source` keeps the credit — leader parts pass damage on 1:1 with it, §20.10), `battle.addCoins(playerId, n)`, `battle.fx(kind, params)` (client VFX event), `battle.rng()` (never `Math.random`), `battle.getPlayer(playerId)` (bonds/band/effects view), `battle.redeploy(unit)` (on its rest tile: where a knocked-out operator lies, else its home; `battle.isReservedTile(r, c)` = a tile no automatic placement takes, `battle.downOn(r, c)` = the knocked-out operator lying there, whose tile no ally deploys or moves onto — §21.24), `battle.after(seconds, fn)` (scheduled callbacks), `battle.every(seconds, fn)`.

 ### 5.5 Combat rules (engine)

 - **Initial deployment** at t=0, free: operators by column — the left column first, top to bottom within a column (right-side boss player: from its own left, the highest field column) — then the summons (PRTS 卫戍协议/帮助 "按从上到下>从左到右的顺序部署。优先部署干员，随后为召唤物", read as a scan: down each column, the columns left to right — the order PRTS gives the 阿戈尔 devour, "更靠左和靠上"; row-major until 0.1.3, §23.23; `aggroSeq` ranks every start-of-battle summon after all operators) — a skill's placed summon (赫默 医疗探机, 巫恋 诅咒娃娃) included, once and free ("所有手动部署的召唤物，无视所属干员的持有状态…作战开始时立即部署一次"; afterwards it re-appears on its tile each time the owner's skill gives one, §20.1); on a shared field (联防, boss) the players' i-th operators deploy together [ASSUMED]. Deploy effects fire (`deploy` with `initial: true`). Then operators carried into 联防 knocked out (`carryState.down`) are forced out (`FORCED_EXIT`, HP 0, §19.3). Then `battleStart`, after which the forced-out operators' timers are re-read (redeploy-time effects that start with the battle, e.g. 机变 征召, cover them too).
 - **DP**: per player, start 10, +1/s, cap 99; used only for auto-redeploy (dead op: after `respawnTime`, if the tile it lies on is free and DP ≥ cost ⇒ redeploy there with full HP, SP = initSp). A knocked-out operator (and, since 0.1.2, one forced out by its own effects, §22.15) stays **down** on the tile it fell on — its own home when it fell on another board piece's home (that piece on the field or not) and its home is free, and 乌尔比安 knocked out while his S3 has moved him always on his deployment tile when it is free (the owner's deviation of 2026-10-07, `Unit.downAtHome`, §25.17.3) — until it redeploys there, and no ally deploys or moves onto that tile meanwhile (PRTS 卫戍协议/帮助 §作战阶段 单位部署; `Battle.downOn` / `restTile` / `_layBody`, §21.24; b.snap `down`, §18.3: counting / waiting for DP / its tile taken, a safeguard) — also one that entered 联防 knocked out (`FORCED_EXIT`, its full redeploy timer — confirmed by the user — with the battle-start redeploy effects, §19.3); summons, devices and enemies leave at once.
- **Blocking** (contact radius, PRTS 游戏数据基础 §阻挡半径 / 作战机制 "中点判定…案例：阻挡"; `constants.js BLOCK_RADIUS`, `Battle._checkBlock`, §19.2): an operator with free capacity blocks an unblocked, blockable enemy whose position lies within its block radius of its centre — ground 0.7071 (squared 0.49999037), 0.8944 × the blocker's 阻挡半径倍率 for blockFly units vs flyers (起飞; mod `blockRadiusScale` = 倍率 − 1 — PRTS 游戏数据基础 "飞行阻挡半径 = 0.8944 × 阻挡半径倍率", 凯尔希·思衡托 +0.23), devices 0.4472 (阻隔工事 / 障碍物); air units (FLY, 近地悬浮, 浮空) only by blockFly units. A ground enemy is never blocked by a unit standing on a tile ground units cannot pass (围墙 / 围栏: PRTS 地形机制 "部署在其中的单位，若当前阻挡类型为'地面阻挡'则无法阻挡敌人" — an enemy pushed or pulled against the fence walks on; air blocking there stays [ASSUMED], §21.23) nor by an airborne (起飞) unit, which blocks flyers only (§21.22). An enemy holding 不可阻挡 (PRTS 异常效果 BLOCK_FREE 「无法阻挡/被阻挡，自动解除阻挡」) — `unblockable` (恐惧, 诱导, enemy abilities), 浮空, and 沉睡 (SLEEPING = 无法行动+无敌+不可阻挡) — is never blocked and is released at once: a sleeping enemy frees its blocker's slot, stays where it is and, once awake, is blocked again only by a blocker with room (§24.9). The nearest such operator wins (ties: scan order [ASSUMED]). It is re-checked every tick, moving or not, so another operator in contact with room takes the enemy over when its blocker dies, is withdrawn or is stunned, and an operator whose capacity frees up grabs an overlapping enemy; one that finds no room walks on. A stunned or frozen enemy is checked too — 晕眩 / 冻结 hold no 不可阻挡 (PRTS 异常效果: the STUN / FROZEN state machines stop its moves and attacks) —, so an operator redeployed beside a held enemy blocks it where it stands and lifts its 隐匿; one held short of contact is blocked once it walks in after the status (GitHub #232; until 0.2.0 the check waited for the status to end). A head-on enemy stops at contact (≈ 0.71 tile), outside the blocker's tile. Enemy uses `blockCnt` capacity (default 1). Blocked enemies stop moving and attack their blocker if melee (or anyone in range if ranged, blocker first). **Every blocker** — melee units (重装要塞 / 领主 / 哨戒铁卫 included), summons and a ranged operator on a melee tile alike — whose attack hits enemies may always target the enemies it blocks, in range or not, whatever its facing, and targets them first ("可以选择且优先选择阻挡单位", PRTS 选择器; the user's rule after playtest #6, "阻挡了就一定要能打到": officially the collision pushes a blocked enemy to its blocker's front; `Battle.blockedTargets`, `targeting.js sortEnemyTargets`; §20.3 — it replaces the playtest #5 QA's melee-only restriction of §19.2). A healer (dmgType `heal`, or a skill attack turned into a heal) selects injured allies only and keeps healing while it blocks (PRTS 卫戍协议/帮助 "对于医疗干员（咒愈师分支除外），攻击目标为需要治疗的单位"; §21.17); 咒愈师 follow the blocker rule. The huge leaders are 自缚 + unblockable (§7).
+- **Blocking** (contact radius, PRTS 游戏数据基础 §阻挡半径 / 作战机制 "中点判定…案例：阻挡"; `constants.js BLOCK_RADIUS`, `Battle._checkBlock`, §19.2): an operator with free capacity blocks an unblocked, blockable enemy whose position lies within its block radius of its centre — ground 0.7071 (squared 0.49999037), 0.8944 × the blocker's 阻挡半径倍率 for blockFly units vs flyers (起飞; mod `blockRadiusScale` = 倍率 − 1 — PRTS 游戏数据基础 "飞行阻挡半径 = 0.8944 × 阻挡半径倍率", 凯尔希·思衡托 +0.23), devices 0.4472 (阻隔工事 / 障碍物); air units (FLY, 近地悬浮, 浮空) only by blockFly units. A ground enemy is never blocked by a unit standing on a tile ground units cannot pass (围墙 / 围栏: PRTS 地形机制 "部署在其中的单位，若当前阻挡类型为'地面阻挡'则无法阻挡敌人" — an enemy pushed or pulled against the fence walks on; air blocking there stays [ASSUMED], §21.23) nor by an airborne (起飞) unit, which blocks flyers only (§21.22). An enemy holding 不可阻挡 (PRTS 异常效果 BLOCK_FREE 「无法阻挡/被阻挡，自动解除阻挡」) — `unblockable` (恐惧, 诱导, enemy abilities), 浮空, and 沉睡 (SLEEPING = 无法行动+无敌+不可阻挡) — is never blocked and is released at once: a sleeping enemy frees its blocker's slot, stays where it is and, once awake, is blocked again only by a blocker with room (§24.9). The nearest such operator wins (ties: scan order [ASSUMED]). It is re-checked every tick, moving or not, so another operator in contact with room takes the enemy over when its blocker dies, is withdrawn or is stunned, and an operator whose capacity frees up grabs an overlapping enemy; one that finds no room walks on. A stunned or frozen enemy is checked too — 晕眩 / 冻结 hold no 不可阻挡 (PRTS 异常效果: the STUN / FROZEN state machines stop its moves and attacks) —, so an operator redeployed beside a held enemy blocks it where it stands and lifts its 隐匿; one held short of contact is blocked once it walks in after the status (GitHub #232; until 0.2.0 the check waited for the status to end). A head-on enemy stops at contact (≈ 0.71 tile), outside the blocker's tile. Enemy uses `blockCnt` capacity (default 1). Blocked enemies stop moving and attack their blocker if melee (or anyone in range if ranged, blocker first). **Every blocker** — melee units (重装要塞 / 领主 / 哨戒铁卫 included), summons and a ranged operator on a melee tile alike — whose attack hits enemies may always target the enemies it blocks, in range or not, whatever its facing, and targets them first — a 速射手's air priority (优先攻击空中单位) excepted, which comes before the enemy it blocks (the owner's decision of 2026-10-08, GitHub #220) — ("可以选择且优先选择阻挡单位", PRTS 选择器; the user's rule after playtest #6, "阻挡了就一定要能打到": officially the collision pushes a blocked enemy to its blocker's front; `Battle.blockedTargets`, `targeting.js sortEnemyTargets`; §20.3 — it replaces the playtest #5 QA's melee-only restriction of §19.2). A healer (dmgType `heal`, or a skill attack turned into a heal) selects injured allies only and keeps healing while it blocks (PRTS 卫戍协议/帮助 "对于医疗干员（咒愈师分支除外），攻击目标为需要治疗的单位"; §21.17); 咒愈师 follow the blocker rule. The huge leaders are 自缚 + unblockable (§7).
 - **Operator attacks**: every `interval` s if a valid target exists; targets = up to `maxTargets` (default 1) from `unitsInGrid(rangeGrid)` (a `rangeAoe` profile — 阵法术师, 轰击术师 — takes every selectable enemy in range, §21.18; 「同时攻击阻挡的所有敌人」 — 强攻手 / 重剑手 / 推击手 and the skills worded so, `hitAllBlocked` — up to the block count, never fewer than 1, since 0.2.0: PRTS 分支特性信息 "普通攻击最大目标数等于阻挡数（不会低于1）"), priority: (1) the enemies it blocks (every blocker, in range or not, §20.3), (2) profession/trait priorities (e.g. fly-first snipers, `trait` text parsed by build-data into `targetPriority`), (3) enemy with the **least remaining path distance** to its goal, ties → earliest spawned. Melee (no ranged trait) and ground-only attacks cannot hit air units (`Unit.isFlying`: FLY, or a 近地悬浮 / 浮空 enemy — PRTS "算作空中单位", §19.1). Stealthed enemies can't be targeted unless blocked or revealed — or within 3 s of a block's end (§22.8). Ranged ops fire projectiles (speed ~12 tiles/s, instant for melee/arts-beam classes as data indicates); damage applies on impact; a 锁定攻击范围 AoE (`rangeAoe`) strikes every selectable enemy on its range at once with instant hits (§21.18).
 - **Healers** (`dmgType heal`): target the ally in range with the lowest HP% below 100%, never a 禁疗 / 孤立 unit (tokens.json `abnormal`, `noHeal`; §20.3; 史尔特尔 during 余烬, §22.7), and never the enemies they block — also while blocking (§21.17); do nothing otherwise (SP still regenerates).
- **Enemy AI**: follow the route (checkpoints; WALK paths recomputed on the grid when devices change); when blocked stop & fight (`dmgType 'none'` enemies never attack unless content arms them through `e.profile` — 转译基底's armed forms, §21.4); ranged enemies (`rangeRadius > 0`) attack ops whose 0.25 collider touches their radius (centre distance ≤ `rangeRadius` + 0.25, §3, §20.4): the enemy's own target rule filters the candidates first (`profile.canTarget`: 只攻击地面单位 …), then priority blocker → special priority → highest taunt → **latest deployed** (`aggroSeq`) — PRTS 索敌 "阻挡→特殊优先级→仇恨值→最早出现"; a stealthed ally is attacked only by the enemy it blocks (§19.1) — and not even by that one when its 索敌不受阻挡影响 (自制投石机, `profile.blockFree`: it selects as if unblocked, §25.18) —, and an enemy's area effects — splash, blasts, area skills, zones, jumps — and buff auras skip it unless it blocks that enemy, while 迷彩 does not protect from splash and the other area effects (`targeting.js areaSelectable` / `auraSelectable`, §22.12); an airborne (起飞) ally is never a ground enemy's selection — picks, area damage, zones and statuses skip it, while 无视无法选择 abilities, direct picks, debuff ticks and flying units still reach it (§21.22); unblocked, they stand for each attack's clip (enemies.json `attackAnim`: its wind-up before the strike and the rest of the clip after it, shortened when the attacks come quicker than the clip; `ATTACK_PAUSE = 0.35 s` after the strike when no clip is known; a 「不停止移动」 attacker never stops; §22.14); any enemy whose block ends after its strike (its blocker stunned by it, knocked out, retreated) stands for the rest of that clip too — PRTS 状态机: the enemy's COMBAT state ends with its attack — and a displacement ends a stand (0.2.0). Every enemy attack strikes at its clip's damage frame after its swing starts — the first one of a contact too —, and a stun / freeze / sleep / 浮空, hiding, 缴械, 恐惧 or 战栗 (blocked) before that frame cuts the swing: no strike, the next swing starts from its wind-up (0.2.0, `ai.js attackWindup`; PRTS 状态机 / 异常效果 STUNNED; no clip known: no wind-up). FLY enemies ignore ground obstacles (only blockFly units block them) and follow their checkpoint route. A feared enemy runs between random 恐惧可达地块 away from the source (PRTS 诱发移动; `fear.js`); 恐惧 outranks 诱导; the 恐惧 / 诱导 walks re-plan from where the enemy stands after a push or an obstacle change, and the route re-plans when they end (§20.4).
+- **Enemy AI**: follow the route (checkpoints; WALK paths recomputed on the grid when devices change); when blocked stop & fight (`dmgType 'none'` enemies never attack unless content arms them through `e.profile` — 转译基底's armed forms, §21.4); ranged enemies (`rangeRadius > 0`) attack ops whose 0.25 collider touches their radius (centre distance ≤ `rangeRadius` + 0.25, §3, §20.4): the enemy's own target rule filters the candidates first (`profile.canTarget`: 只攻击地面单位 …), then priority blocker → special priority → highest taunt → **latest deployed** (`aggroSeq`) — PRTS 索敌 "阻挡→特殊优先级→仇恨值→最早出现"; a stealthed ally is attacked only by the enemy it blocks (§19.1) — and not even by that one when its 索敌不受阻挡影响 (自制投石机, `profile.blockFree`: it selects as if unblocked, §25.18) —, and an enemy's area effects — splash, blasts, area skills, zones, jumps — and buff auras skip it unless it blocks that enemy, while 迷彩 does not protect from splash and the other area effects (`targeting.js areaSelectable` / `auraSelectable`, §22.12); an airborne (起飞) ally is never a ground enemy's selection — picks, area damage, zones and statuses skip it, while 无视无法选择 abilities, direct picks, debuff ticks and flying units still reach it (§21.22); unblocked, they stand for each attack's clip (enemies.json `attackAnim`: its wind-up before the strike and the rest of the clip after it, shortened when the attacks come quicker than the clip; `ATTACK_PAUSE = 0.35 s` after the strike when no clip is known; a 「不停止移动」 attacker never stops; §22.14); any enemy whose block ends after its strike (its blocker stunned by it, knocked out, retreated) stands for the rest of that clip too — PRTS 状态机: the enemy's COMBAT state ends with its attack — and a displacement ends a stand (0.2.0) and holds the enemy in its 失衡 state (0.2.2: no walking, no attack started, no skill — PRTS 失衡位移机制). Every enemy attack strikes at its clip's damage frame after its swing starts — the first one of a contact too —, and a stun / freeze / sleep / 浮空, hiding, 缴械, 恐惧 or 战栗 (blocked) before that frame cuts the swing: no strike, the next swing starts from its wind-up (0.2.0, `ai.js attackWindup`; PRTS 状态机 / 异常效果 STUNNED; no clip known: no wind-up). FLY enemies ignore ground obstacles (only blockFly units block them) and follow their checkpoint route. A feared enemy runs between random 恐惧可达地块 away from the source (PRTS 诱发移动; `fear.js`); 恐惧 outranks 诱导; the 恐惧 / 诱导 walks re-plan from where the enemy stands after a push or an obstacle change, and the route re-plans when they end (§20.4).
 - **Damage**: phys `max(A − max(0, D×(1−defIgnorePct) − defIgnoreFlat), 0.05·A)`; arts `max(A × (1 − R'/100), 0.05·A)` with `R' = max(0, R×(1−resIgnorePct) − resIgnoreFlat)`; true = A; 元素伤害 ('elemental') `max(A × (1 − 元素抗性/100), 0.05·A)`. Then × `dmgDealtMul` (source) × `dmgTakenMul` & type-specific taken multipliers (target) × `dmg.mul` — 元素伤害 takes 元素脆弱 (`elementalTakenMul`) alone, not `dmgTakenMul` (脆弱 = "受到的物理、法术、真实伤害提升"). Then **限伤**: on a leader (`isBoss` — the 10 bossInfoDict enemyIds and their mirrored copies; never parts, drones or escorts) in a `'boss'` / `'hidden'` battle a hit with `ceil(final) ≥ 300000` is cancelled — no HP or pool loss, no credit, no damage number, no `damaged` / kill (`shared/constants.js BOSS_HIT_LIMIT`, `damage.js leaderHitCancelled`, after every multiplier and the hit-step blocks, before HP shields; `loseHp` too — a part's 传递, but not 胄's drone link, a share of the pool that is no hit (`noHitLimit`, §25.13.4); the client's `MAX_BATTLE_DAMAGE`, §20.12). Then shields absorb; dodge checks for phys/arts. Element damage fills a gauge (1000; enemy leaders 2000) × `elemTakenMul` × max(0.05, 1 − 损伤抵抗/100), 损伤抵抗 = data `epResistance` (PRTS 元素; the 5 % floor of DMG_e, PRTS 游戏数据基础). An element fill on a target with no HP left is refused (damage.js `hasHp`: no burst on a corpse). A full gauge bursts with the side-specific official effects, as **无来源** damage (no damage-dealt multiplier or penetration of the unit that filled the gauge, which keeps the kill and the stats; §19.5) (sim/constants.js `ELEMENT`; operators: 灼燃 1200 arts + RES −20, 神经 stun 10 s then 1000 true, 凋亡 15 s of 阻回 (`noSp`: no SP gain of any kind) + 静默 with −1 SP/s and 100 arts/s, 侵蚀 DEF −100 then 800 phys; enemies: the "·我方" terms). Every burst starts a **爆发冷却** = its duration (灼燃 / 神经 / 侵蚀 10 s — enemies' 侵蚀 8 s — 凋亡 15 s) during which NO element of the unit fills or can be recovered; when it ends every gauge of the unit resets (§18.3).
 - **Leak**: an enemy reaching its goal is removed, counted as leaked for its field owner (`sourcePlayerId` for unite). LP accounting is done by the match, not the battle.
 - **Results**: `damageDealt` / a unit's `dmg` count only HP removed from the other side — a unit's own drain or 流失 and friendly damage are the target's `taken` only (§21.12).
@@ -139,7 +139,7 @@ Duration end cleans the window and emits `skillEnd`; it leaves zero charges and
 Redeployment opens a new full window. Permanent passives retain `passive`; first-kill recasts of 缄默德克萨斯 explicitly
 restart a full window. Special timed kits and their effect rules are described in SIM §7.2.

-Auto-trigger rules (`skillTrigger.rule` in data — the official 技能策略, PRTS 卫戍协议/帮助 §作战阶段 技能操作; §20.2): `DEFAULT` (SP ready AND about to attack/heal AND an enemy/injured ally inside the **initial** range, or an enemy the unit blocks; with `trigger.allies` + `hpAtMost` an ally of the skill grid at or below that HP ratio — 塞雷娅 S1 急救, whose cast replaces that attack and is withdrawn with its charge if the ally recovers first), `SKILL_RANGE` (a MANUAL skill with a 技能范围 of its own, not an attack-range change: any living enemy, stealthed ones included, on the skill range, checked every tick — "仅在技能范围内存在敌人（无视其不可选中）时释放技能"; 13 skills / 26 records by the official rule, 14 / 28 in the data with 余 S2's deviation, §22.10), `ACTIVE_RANGE` (the owner's deliberate deviation of 2026-10-05, 0.2.0: a MANUAL skill on the basic strategy — or, by the owner's further decisions of the same day, on the `SEARCH` row, and 深巡 S2 on top of its §21.29 DEFAULT — whose attack range while it runs strictly contains the operator's own range — its rangeId grid, or the own range grown by `ability_range_forward_extend` — checks the DEFAULT condition on that running range every tick, no attack needed; `customRangeGrid` = the running range, `rawRule` the official row (DEFAULT, SEARCH, 深巡's TAKE_DAMAGE); 39 skills / 78 records, of them 薄绿 S1, 蜜蜡 S1, 卡涅利安 S3, 玛恩纳 S2 and 安洁莉娜 S3 from SEARCH; 烛煌 S2 / S3, whose ranges do not contain her 3-1, keep DEFAULT), `TAKE_DAMAGE` (on taking damage — any damage, 无来源 included, never a 流失; every MANUAL 重装 skill, §21.12 — but the six of the deliberate deviation, §21.29: 深巡 / 雷蛇 S2, 号角 S2 / S3 and 灰毫 S1 / S2 are `DEFAULT`, `rawRule` keeping the official TAKE_DAMAGE — 深巡 S2 then `ACTIVE_RANGE` on its 3-2; and 余 S2, `SKILL_RANGE` on its x-1 (§22.10); the 重装 of data/backups.json — the 补位 stand-ins 预备干员-重装 / Mechanist and every 自选 重装 — cast with an enemy in range, `DEFAULT` (or `ACTIVE_RANGE` when the skill's running range is larger), by the owner's decisions of 2026-10-05), `SP_FULL` (immediately — 执旗手 / 战术家 / 吟游者 MANUAL skills, AUTO DP skills such as 焰尾 S1, 凛御银灰 S1 (伺夜 S1 / S2 too, but only while his pack is on the field — PRTS 备注 「仅场上存在狼群时可触发技能」, the kit's own cast; the owner's decision of 2026-10-05 for S1; S1 also while the pack sits in its 战术点形态, which the cast ends — S2 too since 0.2.0: its DP at once, its gift kept for the returning pack [ASSUMED: S2's 备注 uses S1's words]; 黑键 S2 荒芜回响 likewise only while one placeable tile of his range is free, no enemy needed — PRTS 备注 「攻击范围内不存在可部署位时，技能不会被触发」, its kit's cast, 0.2.0), and, by their kits, AUTO skills whose effects act on allies only: 引星棘刺 S1, 浊心斯卡蒂 S2, 纯烬艾雅法拉 S1 — the owner's decision of 2026-10-05 for the last), `CUSTOM_RANGE` (enemy inside the custom grid), `SEARCH` (an enemy inside the initial range, checked every tick — no attack needed; 解放者 / 阵法术师 MANUAL skills and 安洁莉娜 S2, those without a larger running range), `GDGLOW_SKILL_2` (a valid target anywhere on the field: 荒芜拉普兰德, 纯烬艾雅法拉 S3), special ids. The class rows cover **every MANUAL skill** of the class and never an AUTO skill (AUTO skills keep their own rule — an inference: PRTS names whole classes; the data's `skillIndex 0` / `-1` sentinel argues the other way, §20.2). **Operation cooldown**: the engine auto-casts a MANUAL skill no sooner than `AUTO_OP_COOLDOWN` (3 s) after its previous cast and after the unit's battle-start deployment ("自动操作具有3s冷却，在完成一次操作或作战开始时部署的单位将进入冷却"; `Battle._deploy` initial, `battle.flags.startOpCooldown`); kits with an automatic cast of their own check `skill.opCooling`; a pending "next attack" never takes a second charge. Permanent passive skills are always on; finite deployment effects use the duration lifecycle above.
+Auto-trigger rules (`skillTrigger.rule` in data — the official 技能策略, PRTS 卫戍协议/帮助 §作战阶段 技能操作; §20.2): `DEFAULT` (SP ready AND about to attack/heal AND an enemy/injured ally inside the **initial** range, or an enemy the unit blocks; with `trigger.allies` + `hpAtMost` an ally of the skill grid at or below that HP ratio — 塞雷娅 S1 急救, whose cast replaces that attack and is withdrawn with its charge if the ally recovers first), `SKILL_RANGE` (a MANUAL skill with a 技能范围 of its own, not an attack-range change: any living enemy, stealthed ones included, on the skill range, checked every tick — "仅在技能范围内存在敌人（无视其不可选中）时释放技能"; 13 skills / 26 records by the official rule, 14 / 28 in the data with 余 S2's deviation, §22.10), `ACTIVE_RANGE` (the owner's deliberate deviation of 2026-10-05, 0.2.0: a MANUAL skill on the basic strategy — or, by the owner's further decisions of the same day, on the `SEARCH` row, and 深巡 S2 on top of its §21.29 DEFAULT — whose attack range while it runs strictly contains the operator's own range — its rangeId grid, or the own range grown by `ability_range_forward_extend` — checks the DEFAULT condition on that running range every tick, no attack needed; `customRangeGrid` = the running range, `rawRule` the official row (DEFAULT, SEARCH, 深巡's TAKE_DAMAGE); 39 skills / 78 records, of them 薄绿 S1, 蜜蜡 S1, 卡涅利安 S3, 玛恩纳 S2 and 安洁莉娜 S3 from SEARCH; 烛煌 S2 / S3, whose ranges do not contain her 3-1, keep DEFAULT), `TAKE_DAMAGE` (on taking damage — any damage, 无来源 included, never a 流失; every MANUAL 重装 skill, §21.12 — but the six of the deliberate deviation, §21.29: 深巡 / 雷蛇 S2, 号角 S2 / S3 and 灰毫 S1 / S2 are `DEFAULT`, `rawRule` keeping the official TAKE_DAMAGE — 深巡 S2 then `ACTIVE_RANGE` on its 3-2; and 余 S2, `SKILL_RANGE` on its x-1 (§22.10); the 重装 of data/backups.json — the 补位 stand-ins 预备干员-重装 / Mechanist and every 自选 重装 — cast with an enemy in range, `DEFAULT` (or `ACTIVE_RANGE` when the skill's running range is larger), by the owner's decisions of 2026-10-05), `SP_FULL` (immediately — 执旗手 / 战术家 / 吟游者 MANUAL skills, AUTO DP skills such as 焰尾 S1, 凛御银灰 S1 (伺夜 S1 / S2 too, but only while his pack is on the field — PRTS 备注 「仅场上存在狼群时可触发技能」, the kit's own cast; the owner's decision of 2026-10-05 for S1; S1 also while the pack sits in its 战术点形态, which the cast ends — S2 too since 0.2.0: its DP at once, its gift kept for the returning pack [ASSUMED: S2's 备注 uses S1's words]; 黑键 S2 荒芜回响 likewise only while one placeable tile of his range is free, no enemy needed — PRTS 备注 「攻击范围内不存在可部署位时，技能不会被触发」, its kit's cast, 0.2.0), and, by their kits, AUTO skills whose effects act on allies only: 引星棘刺 S1, 浊心斯卡蒂 S2, 纯烬艾雅法拉 S1 — the owner's decision of 2026-10-05 for the last), `CUSTOM_RANGE` (enemy inside the custom grid), `SEARCH` (an enemy inside the initial range, checked every tick — no attack needed; 解放者 / 阵法术师 MANUAL skills and 安洁莉娜 S2, those without a larger running range), `GDGLOW_SKILL_2` (a valid target anywhere on the field: 荒芜拉普兰德, 纯烬艾雅法拉 S3), special ids. In every enemy condition a registered ally target (白铁's 铁钳号·原型机, an enemy-camp summon our attacks select) counts like an enemy for a skill that acts on it (the owner's rule of 2026-10-08): one acting through the unit's own attacks, or whose kit flags `allyTargets` (its selectors take the device too — 蕾缪安 S3, 艾丽妮 S3, 陈 S2, 引星棘刺 S2); [ASSUMED] a skill that picks enemies only is not opened by it. The class rows cover **every MANUAL skill** of the class and never an AUTO skill (AUTO skills keep their own rule — an inference: PRTS names whole classes; the data's `skillIndex 0` / `-1` sentinel argues the other way, §20.2). **Operation cooldown**: the engine auto-casts a MANUAL skill no sooner than `AUTO_OP_COOLDOWN` (3 s) after its previous cast and after the unit's battle-start deployment ("自动操作具有3s冷却，在完成一次操作或作战开始时部署的单位将进入冷却"; `Battle._deploy` initial, `battle.flags.startOpCooldown`); kits with an automatic cast of their own check `skill.opCooling`; a pending "next attack" never takes a second charge. Permanent passive skills are always on; finite deployment effects use the duration lifecycle above.

 **SkillSpec** (authored per chess in `sim/content/kits/tierN.js`; values come from the chess's skill blackboard at its level — normal Lv4, elite Lv7 — so a spec is a *function of the blackboard*):
 ```js
@@ -177,6 +177,6 @@ Each domain file `server/sim/content/{tokens,bonds,garrisons,items,bands,enemies
 - items: 56 equipment (+golden) + 3 Arts; 变形同构体 bond grant; VI-tier set/combos.
 - bands: 40 strategies (meta + battle). 鸭爵 “神秘顾客” swaps 0–2 ground enemies (not the leader, its parts or bounties; on a pair boss field only the player's half — the boss / hidden `onBattleStart` event carries `routes` and `side`) in every teammate's normal / Final Assault / Hidden Core battle from R5 into the act2 鸭爵 / 高普尼克 / 流泪小子 / 圆仔 (lpr 1, 1 fund to the killer, 联防 included) (§21.5); the swaps carry the spawn tag `'duck'`, and the swapped 流泪小子 hides again 1 s after a block ends, not the official 0 s (the owner's decision of 2026-10-07, §25.22.6).
 - enemies: special-type behaviours (FLY, TIMES, ELEMENT, DOT, INVISIBLE, REFLECTION, SPECIAL) and notable individual abilities from `data/enemies.json`; bosses 1–10 scripted — the huge leaders (假想敌：胄 ×2, 假想敌：管 ×2, 盐风主教昆图斯, 阿利斯泰尔, “萨米的意志”) are 自缚 + 无法被阻挡 (PRTS 天赋; `content/bosses.js SELF_BOUND`: a persistent noMove + unblockable buff from spawn) with their hit area (§19.4); 圆仔 and 鸭爵 are 无法被阻挡 (PRTS 天赋, §21.5); the knock-out forms follow PRTS — 深池逐火 余烬 (5 / 10 hits, 隐匿, blockable, back after 10 s), 转译基底·α (damage cancelled until its 2 s change into 寻仇者 / 特战术师 / 幽灵), the leaders' 重生, 守墓石像's statue and flight — and every 重生 clears what operators put on the enemy and restarts its skills from their initial cooldown (§21.4, §21.19); the mode's air units are 静态刚体, never moved (§21.9); leader parts (斩胄之剑 / 破胄之锤 / 碎铳之簧) pass every damage they take to their leader 1:1 as 无来源 HP loss (`PART_TRANSFER` 1, PRTS "等量"), a 假想敌：胄 drone that dies costs it 2 % of the pool (`DRONE_LINK_BASE 'pool'`) (§20.10, §20.13); devices/terrain.
- choices: 机变 families (悬赏决策 bounties, 道具补给, 机密商店, 战术决策 team/personal buffs). 悬赏决策 = the official structure of its round (choices.json `bountyDrafts`, 86 drafted cards: R3 a fixed set of six 两场 cards, R9 6 of a boss group's list, R11 6 of a seen 7-card list; §21.2); 机密商店 (R11 the official composition) and 战术决策 can offer the same card twice (`shopDraft` / `tacticDraft`); bounty cards carry `descRaw`; no draft offers a multi-round card, and 教鞭's lasts two battles (`choices.js MULTI_ROUND_BOUNTY_BATTLES`, §20.6). 教鞭 = a random 战术特训 card (3 drawn, 1 taken).
+- choices: 机变 families (悬赏决策 bounties, 道具补给, 机密商店, 战术决策 team/personal buffs). 悬赏决策 = the official structure of its round (choices.json `bountyDrafts`, 86 drafted cards: R3 a fixed set of six 两场 cards, R9 6 of a boss group's list, R11 6 of a seen 7-card list; §21.2); 机密商店 (R11 the official composition) and 战术决策 can offer the same card twice (`shopDraft` / `tacticDraft`); bounty cards carry `descRaw`; no draft offers a multi-round card, and 教鞭's lasts two battles (`choices.js MULTI_ROUND_BOUNTY_BATTLES`, §20.6). Pointing Stick (教鞭) offers up to three private Tactical Training cards during PREP; the owner confirms one through `g.choice` with its `choiceId`, with Ready blocked until confirmation or the existing prep deadline's auto-pick. “神秘顾客” remains random (META §2.5).

 ---
@@ -22,6 +22,13 @@ Part of [DESIGN.md](../DESIGN.md) (the index; section numbers are global).
 ## 4. Time model

 - Simulation time is **game seconds**. Fixed step `TICK = 1/30 s`.
+- Timers count in ticks and take what is within 1e-9 as reached — attack cooldowns and enemy wind-ups (`ai.js
+  attackCountdown`), time SP against its cost (`skills.js gainSp`), skill durations, buff intervals, `every()` /
+  `after()`, enemy abilities, the 失衡 end —, so a whole number of ticks takes exactly that many: 1 s = 30 ticks (PRTS
+  作战机制/sandbox 「帧对齐机制对攻速的影响」: one frame = 1/30 s). Floating point alone held 1 s attacks and 10-SP charges a
+  tick longer until 0.2.2 (PR #402, §27.61). A non-integer count ends on the tick that crosses the line, ⌈30 × t⌉ (1.25 s
+  = 38); the official game rounds it (四舍五入 since 2019-12-24: 能天使's 0.678 s is 20 frames there, 21 here) — an older
+  difference, not changed.
 - Combat runs at **2× real time** (forced, like the original): 60 ticks per real second — in the browser's runner (§14) or, for server-run fields, the server's accumulator (2 ticks every real 1/30 s, never more than 8 ticks per interval to avoid spirals).
 - Combat time limits come from `data/config.json → modes[m].rounds[r].combatTimeLimit` (= the level's `maxPlayTime`; also `modes[m].combatTimeLimit[r]`) and are **real seconds of the forced-2× battle** (verified by the balance pass: spawn schedules exceed the limit when read as game seconds). The server passes `2 × value` game seconds to the Battle (`server/match/gamedata.js → combatTimeLimit`; `config.combatTimeScale`, default 2); the countdown players see equals the data value. When time runs out, living non-boss enemies count as leaked. 联防 uses the round's limit.
 - Final Assault / Hidden Core have **no hard stop** and their clocks are real seconds too: the boss level's `levelMaxPlayTime` (120 real s) is only the HUD countdown (`m.public.deadline`; the battle goes on past it), and the overtime drain (`bossOvertimeAfter` 150 / `bossOvertimeDrainPerSec` 1 = official `bossTurnHpReduceTime`) takes 1 team LP per whole **real** second from 150 real s (= 300 game s on the 2× field clock; first point at 151 s — `gamedata.js bossOvertimeDue`). `m.public.overtimeAt` = the ms epoch when the drain starts; both are placed on the field clock.
@@ -7,14 +7,15 @@ Part of [DESIGN.md](../DESIGN.md) (the index; section numbers are global).
 ### 6.1 State machine (Match.js)

 ```
-LOBBY(room) → INFO_CHECK (co-op 25 s; solo and any single-human match untimed (§18.2); all ready ⇒ skip; the operator loadout (§16) locks when it ends) → BAND_DRAFT (co-op: random order, ONE countdown — `BAND_TURN_SECONDS` 30 s per turn = m.public.deadline, AI seats pick at once, 1 skip each, a strategy a teammate already took is refused (队友已选), timeout ⇒ the strategy the player highlights (`g.bandFocus`) while free, else band_bldsk, else the first free one; solo: free pick, no timer, no skip; a single human: untimed) → BATTLE_CHECK (3 s)
+LOBBY(room) → INFO_CHECK (co-op 25 s; solo and any single-human match untimed (§18.2); all ready ⇒ skip; the operator loadout (§16) locks when it ends) → BAND_DRAFT (co-op: random order — the room option 「AI 队友最后选择」 (host, off by default; `room.setAiPicksLast`) puts every human seat before every AI seat, each group keeping the drawn order, and draws no extra random number; a skipping human goes behind the other humans still to pick and ahead of the AI seats, or to the very end when no other human remains [ASSUMED] —, ONE countdown — `BAND_TURN_SECONDS` 30 s per turn = m.public.deadline, AI seats pick at once, 1 skip each, a strategy a teammate already took is refused (队友已选), timeout ⇒ the strategy the player highlights (`g.bandFocus`) while free, else band_bldsk, else the first free one; solo: free pick, no timer, no skip; a single human: untimed) → BATTLE_CHECK (3 s)
 → loop r = 1..lastRound:
     ROUND_START   (income, upgrade price −1 (floor 0), temp NOT wiped — what overflowed after the last prep's deadline is shown in this
                    prep (§6.2 temp overflow), unfrozen shop slots rerolled, frozen kept, <进入休整期时> effects)
-     [SP_DRAFT]    (if r ∈ spRounds: 机变 draft; co-op 30 s first picker / 16 s others; solo 3 cards; solo / single human untimed;
+     [SP_DRAFT]    (if r ∈ spRounds: 机变 draft; co-op: random order (「AI 队友最后选择」, off by default: humans first, the same partition as the strategy draft), 30 s first picker / 16 s others; solo 3 cards; solo / single human untimed;
                    a card is picked with two taps, §18.2)
     PREP          (co-op: timer from config, ends when all alive humans+bots ready; solo / single human: untimed, ends on ready)
-     PREP_END      (<休整期结束时> effects; unfrozen shop cleared; temp overflow: Ready is blocked while temp is non-empty; at the deadline
+     PREP_END      (<休整期结束时> effects; unfrozen shop cleared; Ready is blocked while a personal choice or temp overflow is pending;
+                    at the deadline personal choices are auto-picked first, then
                    the temp pieces due at this prep are resolved (chess sold back, items destroyed, summon stacks removed — they come back at the next round start, §20.1) — pieces that arrived
                    after Ready or during the <休整期结束时> effects wait for the next prep (§6.2 tempDue))
     COMBAT        (one Battle per alive player, all in parallel, 2×; watchers get snapshots; solo: pausable, §14 Solo pause)
@@ -34,6 +35,18 @@ LOBBY(room) → INFO_CHECK (co-op 25 s; solo and any single-human match untimed
 ```
 Solo FUNNY has 9 rounds (boss at R9); everything is read from `data/config.json → modes[modeId]`.

+**Confirm-page setup reroll (remake convenience proposal, GitHub #77 / PR #408).** During INFO_CHECK the host may
+request a reroll; every non-departed human explicitly consents, including a human on AI control. AI and spectator
+seats do not vote, and one human rerolls immediately. All humans must be connected to start a vote; a refusal, host
+cancellation, disconnect or departure cancels it. While voting, the confirmation deadline and any queued all-ready
+transition are suspended. Cancellation restores the previous remaining time and ready states; success clears human
+ready states (bot-controlled seats keep auto-ready) and restarts the full confirmation deadline, still untimed for a
+single human. Success redraws the stage, leader, hidden leader, special factions and their full wave schedule, and
+bans, rebuilding shared and per-player DIY stock. Mode-fixed rules still apply and draws may repeat. The room,
+connections, loadout, ownership, DIY picks, potential / cultivation and AI-picks-last option remain unchanged. A
+separate RNG stream leaves existing shop / draft streams untouched. Setup revisions and vote IDs reject stale requests;
+the minimum interval between proposals is 300 ms. `match/setupVote.js` owns this flow; the lobby authorizes the host.
+
 ### 6.2 Player state (PlayerState.js) — authoritative per player

 ```js
@@ -44,6 +57,7 @@ Solo FUNNY has 9 rounds (boss at R9); everything is read from `data/config.json
  board: Map<'r,c', Piece>,                          // chess & token pieces on the normal field
  bonds: { [bondId]: { count, active, tier, layers } },  // layers persist all match
  effects: [ EffectRef ],                             // band/choice/team/garrison persistent effects & counters
+  personalChoice: null | { id, round, sourceItemId, cards: [ BountyCard ] }, // private PREP offer; raw cards on the server
  counters: { refreshes, buys, sells, … },
  stats: { … for results } }
 Slot  = { kind:'chess'|'item', id, price, sold:false, frozen }
@@ -62,6 +76,12 @@ Economy/shop/pool/merge/hand rules: exactly as research 00-INDEX §3–§4 (inco

 **Temp overflow (临时整备区, row 8; user playtest #3 item 3).** Passive gains that find the hand full (merge elites, grants — the operator a 突变细胞 transformation gains included, §21.1 —, battle-result rewards, equipment returned by a merged / sold / transformed carrier) overflow into the 5 temp slots, and Ready is refused while any is occupied (official "直到溢出情况排除才可开始进行作战"). A temp piece is resolved — a chess sold back to the pool with its summons removed, items destroyed, a summon stack removed (it comes back at the next round start, PRTS "干员所属召唤物会于下一回合返还", §20.1) — only at the deadline of the **first prep in which its player could act on it** (`PlayerState.tempDue(piece)` vs `prepsEnded`; every write into a temp slot goes through `_putTemp`, which records it): a piece that arrived during a prep before Ready is due at that prep's end; one that arrived after Ready or during the `<休整期结束时>` effects, or outside PREP (COMBAT, 联防, SETTLE — battle-result grants, a SETTLE merge's elite, returned equipment —, the next ROUND_START, 机变) is kept, visible and usable (move to a free hand slot, place, equip, use, destroy, sell) through the **next** prep. Un-readying (`setReady(false)`) makes whatever arrived while ready due at the current prep. ROUND_START no longer wipes temp (it used to destroy what SETTLE had put there before the player ever saw it). A merge's elite whose merge consumed a deployed copy takes that copy's tile and facing (§20.11); otherwise it goes to the hand, overflowing into temp (also outside PREP). `server/match/audit.js` (`tools/matchrun.mjs --check`) tracks the arrivals itself and flags any temp piece that outlives its due prep. The UI frames the row on the board with the rule (§17.5).

+Pointing Stick creates at most one pending offer per alive, non-Ready player in PREP. The owner confirms it with
+`g.choice { idx, choiceId }`; confirmation validates its ID, round and index, adds one bounty and clears the offer.
+Ready requires no pending offer as well as an empty temp area. The co-op prep deadline resolves the offer randomly
+before resolving temp and setting Ready; untimed solo / single-human prep does not schedule a separate choice timer.
+The offer is recipient-private, including after reconnect, and never enters `m.public.sp` or a teammate's scout view.
+
 ### 6.3 Bonds & layers (bondsMeta.js)

 - `count` = number of **distinct base chess** on the board carrying the bond (`BOARD`); `BOARD_AND_DECK` bonds also count the hand; 绝技 counts elites on board; 调和 adds +1 to each core bond count that has a real member on the board (the views mark such an entry `harmony: 1`, §21.26), etc. (research 02).
@@ -81,7 +101,10 @@ Per match: pick stage (weighted among the mode's `config.modes[m].stages`, weigh

 ### 6.6 AI players (bot.js)

-Deterministic-ish heuristic player used for (a) AI teammates, (b) auto-play for disconnected humans (only auto-ready + auto-picks, not shopping, unless "AI 托管" is toggled), (c) tests. Strategy (§21.6): pick a band (weighted by starting LP; never one built around a bond the mode switches off — `gd.bandBondIds`, §21.26); each prep: take offers, sell dead bench weight, fill the board, complete affordable merges, level on the curve, then buy only merge progress / lineup upgrades worth more than the refreshes their price would buy (shop odds of completing held pairs), freeze for an unaffordable third copy (坎诺特 keeps 5 funds back); build around a committed focus core bond (a teammate's main bond on its strip counts against it; bonds the mode switches off never) + a second bond; value dealers by DPS vs the round's DEF / RES (an attack on every enemy in range counts double — 阵法术师 / 轰击术师), with the range the unit is deployed with; place by the exposure model + rehearsal on the server's deploy map (no 深水区; a tactician's 援军 within `PlayerState.summonRange`; a melee chess whose trait reads 「可以放置于远程位」 (钩索师 / 推击手) on a 高台 whose range covers the enemy road, else on the ground, counted as a blocker in the lineup either way — §3; `g.move` judges legality); items by effect (META §1.5: 突变细胞 on the least valuable normal single below 6阶, a cell left in temp gets a hand slot made for it, the operator its transformation gains is deployed by the next prep's lineup step like any owned unit — AI 托管 too; the buy loop's bench shed never sells a piece gained since the bot's last prep, §21.1); 机变 bounties by expected payout − expected LP loss over the offered cards; resolve temp — on a human seat under AI 托管 never by destroying an item while a bench operator can be sold for the slot —; ready. A bot's whole prep — shop decisions and layout planning (`botPrepBeginSteps`, 25–70 ms late in a 4-bot match), the layout rehearsal and `botPrepEndSteps` — runs in wall-clock slices of `botSliceMs` (8 ms) per scheduler callback on a real host (step generators that yield between whole actions, never with a transient board; the same actions in the same order, so decisions are unchanged; at once in virtual time). Under AI 托管 a human's own 自选 pieces score `DIY_PIECE_BONUS` (12) more in the shop / reward choice and the 调度中心 is levelled one round earlier for the step that opens a slotted tier (`levelUp`, `diyOpensAt`; 0.2.0 — bots field no 自选 piece).
+Pending Pointing Stick offers use the existing bounty scorer at bot prep entry, immediately after a successful cast,
+and before Ready. The prep's deadline — and the scheduler's fallback when a step of a bot's prep fails — randomly resolves any remaining offer before temp and Ready; a seat that left is eliminated and its offer dropped.
+
+Deterministic-ish heuristic player used for (a) AI teammates, (b) auto-play for disconnected humans (only auto-ready + auto-picks, not shopping, unless "AI 托管" is toggled), (c) tests. Strategy (§21.6): pick a band (weighted by starting LP; never one built around a bond the mode switches off — `gd.bandBondIds`, §21.26); each prep: take offers, sell dead bench weight, fill the board, complete affordable merges, level on the curve, then buy only merge progress / lineup upgrades worth more than the refreshes their price would buy (shop odds of completing held pairs), freeze for an unaffordable third copy (坎诺特 keeps 5 funds back); build around a committed focus core bond (a teammate's main bond on its strip counts against it; bonds the mode switches off never) + a second bond; value dealers by DPS vs the round's DEF / RES (an attack on every enemy in range counts double — 阵法术师 / 轰击术师), with the range the unit is deployed with; place by the exposure model + rehearsal on the server's deploy map (no 深水区; a blocker whose range is its own tile only — range 0-1 — takes a free tile of the enemy road first; a tactician's 援军 within `PlayerState.summonRange`; a melee chess whose trait reads 「可以放置于远程位」 (钩索师 / 推击手) on a 高台 whose range covers the enemy road, else on the ground, counted as a blocker in the lineup either way — §3; `g.move` judges legality); items by effect (META §1.5: 突变细胞 on the least valuable normal single below 6阶, a cell left in temp gets a hand slot made for it, the operator its transformation gains is deployed by the next prep's lineup step like any owned unit — AI 托管 too; the buy loop's bench shed never sells a piece gained since the bot's last prep, §21.1); 机变 bounties by expected payout − expected LP loss over the offered cards; resolve temp — on a human seat under AI 托管 never by destroying an item while a bench operator can be sold for the slot —; ready. A bot's whole prep — shop decisions and layout planning (`botPrepBeginSteps`, 25–70 ms late in a 4-bot match), the layout rehearsal and `botPrepEndSteps` — runs in wall-clock slices of `botSliceMs` (8 ms) per scheduler callback on a real host (step generators that yield between whole actions, never with a transient board; the same actions in the same order, so decisions are unchanged; at once in virtual time). Under AI 托管 a human's own 自选 pieces score `DIY_PIECE_BONUS` (12) more in the shop / reward choice and the 调度中心 is levelled one round earlier for the step that opens a slotted tier (`levelUp`, `diyOpensAt`; 0.2.0 — bots field no 自选 piece).

 ---

@@ -95,11 +118,13 @@ Official rule (bwiki 盟约, 更新公告 5114): before a match the player canno

 **Loadout wire & state:** the client keeps a per-browser loadout `{ [baseChessId]: { skill: index, module: uniEquipId | 'none' } }` (localStorage) and sends it with C2S `room.loadout { entries }` (validated: known visible chess, legal skill index for that chess's normal AND elite status, legal module id). The lobby stores it on the seat; at match start `Match` receives `seats[].loadout`; `PlayerState` keeps it immutable for the match; board pieces are resolved to `{chessId, skillIndex, moduleId}` when building `PlayerBattleInput.units[]` (fields `skillIndex`, `moduleId`; BattleSpec carries them). Bots use defaults. m.private exposes the effective loadout so the UI (shop cards, detail panel) shows the chosen skill/module.

+**潜能 / 练度 (0.2.2, the owner's decision of 2026-10-08: 「调配干员里自己设置吧，默认满潜满加成」):** official — an operator fights at the player's own potential (调度手册 「卫戍协议中干员潜能由自身已持有干员潜能决定」; an unowned 特许 operator at potential 1) and an owned operator of the same name gets the 自持有 练度 bonus (activity_table `cultivateEffectList`: 未精英化 +0 %, 精英1 ATK / DEF +5 %, 精英2 ATK +10 % DEF / HP +5 %, 精英2 Lv.60 +10 % each — a multiplier of its own, never summed with the 直接乘算 bonuses). The 干员调配 screen sets both per operator (by charId: initial, elite and 自选 forms share them; default 潜能 6 and 精英2 Lv.60); they travel as `room.loadout.ops` (`{ [charId]: { potential, cultivate } }`, defaults dropped; shared/protocol.js `checkLoadoutOps`, strict like the entries), `seats[].ops` → `PlayerState.ops` (re-checked; bots none = the defaults), and every operator the player owns states them in its `PlayerBattleInput` entry (`potential`, `cultivate`; shared/potential.js `cultivationOf`) — never a 补位 stand-in (a not-owned NORMAL chess) or a prototype 自选 pick. The sim composes the def at the potential (data/*.json carry every potential, DATA.md §2.3; the summons' talents at the owner's) and multiplies ATK / DEF / max HP by the tier's `char_attribute_mul` (effects.json `aceffect_char_1…4`, SIM.md §12). The cards (局内数值, shop and detail cards, a teammate's unit from UnitInfo `potential` / `cultivate`) show the numbers at the settings.
+
 **Sim:** `simdata` resolves a unit def for `(chessId, skillIndex, moduleId)` (cached): `skill` = the selected SkillRecord; stats = `statsBase` + module attr (elite) ; trait/talents = base with the selected module's overrides. Kit contract (backward compatible): a kit function `(bb, chess) => { skill?, skills?: { [skillId]: SkillSpec }, talents, trait }` receives `chess.skill` = the **selected** skill and `bb` = its blackboard; the loader uses `skills[selectedSkillId]` if present, else `skill` only when the selected skill is the chess's default, else the **generic** skill spec — talents/trait from the kit always apply. Every visible chess must get hand-authored specs for all of its selectable skills (tracked by a coverage test).

-**UI:** a 干员调配 screen reachable from the lobby, room and briefing (INFO_CHECK): filter by tier/bond/class, per chess choose skill (icon, name, SP info, desc at Lv4 and Lv7) and elite module (official type icon from the local-client art when extracted (§13), else a lettered tile; name/type, stat bonus, trait/talent changes), 恢复默认, persisted; a 局内数值 section between the skills and the modules (普通 / 精锐 toggle, 精锐 first) draws the detail card's own stats block, 特性 and 天赋 for the chosen skill / module (§22.16); the chosen skill/module shown on shop cards and in the detail panel (a teammate's unit — 前往查看 in prep, 联防 / 最终攻势 fields, an observed battle — shows **its owner's** skill/module from its UnitInfo `skillIndex` / `moduleId`, `gameLogic.js unitLoadout`) — which also shows the stats, 特性, talents and attack range the unit fights with (`shared/loadoutRecord.js` `loadoutRecord` / `attackRangeGrid`, the same composition the sim's `getChess(id, loadout)` uses; the deploy wheel's range preview too) — the range it is deployed with: a selected passive range skill (引星棘刺 S3 3-9) and the 特性's permanent 攻击距离 (信仰搅拌机 SPT-Y +1) included, over all 1284 loadouts equal to the battle's range at deployment; in battle and prep the card draws the live `range` (§21.16). In battle the renderer plays the equipped skill's own Spine clip and ON_SKILL_START sound (`UnitInfo.skillIndex`; `data/assets.json` lists icons, clips and sounds for every skill index).
+**UI:** a 干员调配 screen reachable from the lobby, room and briefing (INFO_CHECK): the roster is one list, a row per operator with aligned columns under a head row — portrait / name / bonds (opens the detail), the skills and the elite's modules to tap (不装备 first; the quick choices of PR #301 by @farridge, laid out as one list instead of its three columns of cards, the owner's call of 2026-10-08), the 潜能 / 练度 selects (`screens/loadout.js RosterRow`, `screens/cultivation.js`); the detail beside it (below 1000 px wide it slides over the list, a quick choice keeps the list) opens with the 潜能与练度 section (潜能 1–6, the four official 练度 tiers with their effect texts); the detail then shows the operator's 特质 — the in-match `GarrisonBlock` / RichText of the form its 普通 / 精锐 skill toggle shows, triggers such as 「<进入休整期时>」 kept (0.2.2, PR #301 by @farridge); filter by tier/bond/class, per chess choose skill (icon, name, SP info, desc at Lv4 and Lv7) and elite module (official type icon from the local-client art when extracted (§13), else a lettered tile; name/type, stat bonus, trait/talent changes), 恢复默认, persisted; a 局内数值 section between the skills and the modules (普通 / 精锐 toggle, 精锐 first) draws the detail card's own stats block, 特性 and 天赋 for the chosen skill / module (§22.16); the chosen skill/module shown on shop cards and in the detail panel (a teammate's unit — 前往查看 in prep, 联防 / 最终攻势 fields, an observed battle — shows **its owner's** skill/module from its UnitInfo `skillIndex` / `moduleId`, `gameLogic.js unitLoadout`) — which also shows the stats, 特性, talents and attack range the unit fights with (`shared/loadoutRecord.js` `loadoutRecord` / `attackRangeGrid`, the same composition the sim's `getChess(id, loadout)` uses; the deploy wheel's range preview too) — the range it is deployed with: a selected passive range skill (引星棘刺 S3 3-9) and the 特性's permanent 攻击距离 (信仰搅拌机 SPT-Y +1) included, over all 1284 loadouts equal to the battle's range at deployment; in battle and prep the card draws the live `range` (§21.16). In battle the renderer plays the equipped skill's own Spine clip and ON_SKILL_START sound (`UnitInfo.skillIndex`; `data/assets.json` lists icons, clips and sounds for every skill index).

-**导出 / 导入 a loadout:** 导出 hands the current loadout out as a versioned envelope `{ kind: 'stronghold.loadout', v, exportedAt, count, entries }` — copied through `ui/clipboard.js copyText` or downloaded as a file. 导入 accepts that envelope, the stored `{ v, entries }` form and a bare `{ [chessId]: { skill, module } }` map (serialised text too), and refuses another tool's `kind`, junk, an empty result and a **newer** `v` (`parseImport` — a newer payload is never mis-read). Both go through `loadoutSync.applyLoadoutEntries` → `sanitizeEntries`, so a preset from another build never sends the server an entry it would refuse, an import that keeps nothing changes nothing, and the result is reported as N imported / M not (`ui/loadoutModel.js`, `screens/loadout.js`).
+**导出 / 导入 a loadout:** 导出 hands the current loadout out as a versioned envelope `{ kind: 'stronghold.loadout', v, exportedAt, count, entries, ops }` (`ops`: the 潜能 / 练度 settings, 0.2.2 — an import without them leaves the current settings alone) — copied through `ui/clipboard.js copyText` or downloaded as a file. 导入 accepts that envelope, the stored `{ v, entries }` form and a bare `{ [chessId]: { skill, module } }` map (serialised text too), and refuses another tool's `kind`, junk, an empty result and a **newer** `v` (`parseImport` — a newer payload is never mis-read). Both go through `loadoutSync.applyLoadoutEntries` → `sanitizeEntries`, so a preset from another build never sends the server an entry it would refuse, an import that keeps nothing changes nothing, and the result is reported as N imported / M not (`ui/loadoutModel.js`, `screens/loadout.js`).
 **干员持有 (0.2.0 补位):** the overlay's second tab marks operators as not owned (the 55 NORMAL chess; per-browser `sp.pref.ownership` = `{ v, notOwned }`, synced as `room.ownership { notOwned }` by the same sync engine as the loadout, `ui/loadoutSync.js installPrefSync`; 导出 / 导入 as `{ kind: 'stronghold.ownership', v, exportedAt, count, notOwned }`). A not-owned chess fights as its official stand-in with the backup skill / module — the loadout never applies to it ("对于补位干员其技能不可更改") — and is shown as the stand-in wherever the player's copy appears (shop / reward cards, hand and 临时整备区, board, detail card, result lineup, a teammate's scouting and bond popups: portrait, name, model, with a small 「替补」 mark; the owner's recall of the official mode, 2026-10-06), while its bonds, 特质, tier, price and merge stay the chess's; the 0.2.0 补位 design subsections have the rules and sources.
 **自选编队 (0.2.0):** the overlay's third tab fills the four DIY slots (5阶 ×2, 6阶 ×2) with an owned 6★ outside the chess pool (any of its three skills, any module of its elite form but a 集成战略 one) or a prototype (its locked skill / module); only operators with a kit are offered (`welcome.diyKitted`). Per-browser `sp.pref.diy` = `{ v, picks }`, synced as `room.diy { picks }` by the same sync engine; 导出 / 导入 as `{ kind: 'stronghold.diy', v, exportedAt, count, picks }` (`ui/diyModel.js`, `screens/diy.js`). In the match a slotted piece is the operator for its player (name, art, class, bonds from its factions, no 特质, the pick's skill and module; the slot's tier, price and merge), sold only in that player's shop from the slot's 调度中心 level with its own stock; the shop / reward / detail cards and the own pieces carry a 「自选」 badge (`ui/gameLogic/diy.js`); the 0.2.0 自选编队 design subsection has the rules and sources.

@@ -7,26 +7,41 @@ Part of [DESIGN.md](../DESIGN.md) (the index; section numbers are global).
 Transport: one WebSocket per tab at `/ws`, JSON text frames `{ t: '<type>', ...payload }`. Client requests may carry `rid` (request id); the server's direct reply echoes `rid` (`ok` or `error`). Server pushes have no `rid`. Max frame 64 KB inbound; rate limit 40 msgs/s per connection (excess dropped with `error RATE`).

 ### 8.1 Session & lobby
-C→S: `hello {name, token?, version}` · `room.create {mode:'solo'|'coop', difficulty}` · `room.join {code}` · `room.leave` · `room.ready {ready}` · `room.setDifficulty {difficulty}` (host) · `room.addBot` / `room.removeBot {seat}` (host) · `room.kick {seat, playerId}` (host, before the match: another human leaves the room and gets `room.closed {reason:'kicked'}`; may join again; `playerId` = the player the host confirmed, a seat that changed hands meanwhile → `BAD_TARGET`) · `room.start` (host) · `room.loadout {entries}` (§16: any time in the lobby / room and during INFO_CHECK, then `WRONG_PHASE`; heavy rate-limit bucket; the match exposes it as `m.private.loadout`) · `room.ownership {notOwned}` (0.2.0 补位, the 干员持有 tab: the base chess ids marked as not owned — ≤ 160 ids, kept leniently: anything but a droppable NORMAL chess is dropped; stored on the session and the seat, a match takes the seat's list when it starts and never changes it — during a match `ROOM_STARTED` 'stored for the next match'; heavy bucket; the match exposes it as `m.private.standIns`) · `room.diy {picks}` (0.2.0 自选编队: `{ [slotBaseId]: { charId, skillIndex?, uniEquipId? } | null }` for the four DIY slots — ≤ 8 entries, kept leniently: shared/protocol.js `checkDiyPicks` drops an illegal pick (an operator without a kit, another tier's prototype, a prototype off its locked skill, a module of another game mode (集成战略 / 生息演算), an owned operator's second slot, the same operator twice in a tier, an unknown slot), only malformed picks are `BAD_MSG`; stored on the session and the seat like the not-owned list, a match takes the seat's picks when it starts — during a match `ROOM_STARTED` 'stored for the next match'; heavy bucket; the match exposes them as `m.private.diy`) · `ping {c}`
-S→C: `welcome {playerId, token, name, serverNow}` · `room.state {code, hostId, mode, difficulty, seats:[{seat, playerId, name, isBot, ready, connected}|null ×4]}` · `room.closed {reason}` · `pong {c, s}` · `error {rid?, code, msg}` · `ok {rid}`
+
+Confirm-page reroll (GitHub #77 / PR #408): `room.rerollSetup {setupRevision}` and
+`room.cancelReroll {voteId}` are host-only, while `g.rerollVote {voteId, agree}` requires a participating human.
+`g.infoReady {setupRevision?}` confirms the setup the player saw; an omitted revision means 0, so it is accepted only
+before the first successful reroll. `m.public.setupRevision` and
+`m.public.rerollVote = null | {id, proposerId, voters:[playerId], agreed:[playerId]}` synchronize every client and resync.
+These requests only apply in INFO_CHECK. A mismatched setup revision / vote ID is `BAD_TARGET`; spectators cannot act.
+
+C→S: `hello {name, token?, version}` · `room.create {mode:'solo'|'coop', difficulty}` · `room.join {code}` · `room.leave` · `room.ready {ready}` · `room.setDifficulty {difficulty}` (host) · `room.setAiPicksLast {on}` (host, co-op, before the match — `ROOM_STARTED` during it, `BAD_TARGET` in a solo room; un-readies the other humans like a difficulty change; the match takes it as `opts.aiPicksLast`: humans draft before AI seats, off by default, no extra random draw) · `room.addBot` / `room.removeBot {seat}` (host) · `room.kick {seat, playerId}` (host, before the match: another human leaves the room and gets `room.closed {reason:'kicked'}`; may join again; `playerId` = the player the host confirmed, a seat that changed hands meanwhile → `BAD_TARGET`) · `room.start` (host) · `room.loadout {entries, ops?}` (§16: any time in the lobby / room and during INFO_CHECK, then `WRONG_PHASE`; heavy rate-limit bucket; the match exposes it as `m.private.loadout`; `ops` — 0.2.2 — the per-operator 潜能 / 练度 `{ [charId]: { potential?: 1–6, cultivate?: 0–3 } }` of the 干员调配 roster and the 自选 owned pool, ≤ 256, checked strictly with the entries, `m.private.ops`; absent = none set: every operator 潜能 6, 精英2 Lv.60) · `room.ownership {notOwned}` (0.2.0 补位, the 干员持有 tab: the base chess ids marked as not owned — ≤ 160 ids, kept leniently: anything but a droppable NORMAL chess is dropped; stored on the session and the seat, a match takes the seat's list when it starts and never changes it — during a match `ROOM_STARTED` 'stored for the next match'; heavy bucket; the match exposes it as `m.private.standIns`) · `room.diy {picks}` (0.2.0 自选编队: `{ [slotBaseId]: { charId, skillIndex?, uniEquipId? } | null }` for the four DIY slots — ≤ 8 entries, kept leniently: shared/protocol.js `checkDiyPicks` drops an illegal pick (an operator without a kit, another tier's prototype, a prototype off its locked skill, a module of another game mode (集成战略 / 生息演算), an owned operator's second slot, the same operator twice in a tier, an unknown slot), only malformed picks are `BAD_MSG`; stored on the session and the seat like the not-owned list, a match takes the seat's picks when it starts — during a match `ROOM_STARTED` 'stored for the next match'; heavy bucket; the match exposes them as `m.private.diy`) · `ping {c}`
+S→C: `welcome {playerId, token, name, serverNow}` · `room.state {code, hostId, mode, difficulty, aiPicksLast, seats:[{seat, playerId, name, isBot, ready, connected}|null ×4]}` · `room.closed {reason}` · `pong {c, s}` · `error {rid?, code, msg}` · `ok {rid}`

 Platform additions (implemented): `welcome.{resumed, version}`, `welcome.diyKitted` (0.2.0: the operators a 自选 slot may field — server/sim/content/kits/index.js `KITTED_CHARS`, the picker's list), `room.state.inMatch`, `error.detail`; `room.closed.reason ∈ 'timeout'|'kicked'|'empty'|'shutdown'` (`kicked`: room.kick / room.removeSpectator; `empty`: a spectator whose room lost its last player); WebSocket close codes 4001 (session taken over by another socket — client must not auto-reconnect), 4002 (no hello in time), 1008 (flooding), 1001 (shutdown). The host's `room.start` counts as the host being ready. After any (re)`hello` the server pushes `room.state` and, if a match runs, the match resends `m.public`/`m.private` (+`m.field`, or the field's `b.start` under client-side combat). **Reconnect window:** a dropped session stays resumable for 10 min (`net.js NET_DEFAULTS.reconnectWindowMs`; co-op seats keep playing their last lineup meanwhile, §6.6); a session that drops while its **solo** room's match runs stays resumable for the official `config.constants.singleReconnectTime` (86 400 s = 24 h; lobby option `soloReconnectWindowMs`; research 01 §1 "24小时内随时返回") — the untimed solo run simply waits. Only when the window expires does the drop become `match.onLeave` (co-op: 中途退出 = elimination; solo: the run ends `'abandoned'`). The client keeps its token in sessionStorage plus a recent-token list in localStorage, so reopening the tab resumes. Everything lives in server memory: a server restart ends every room and run. The lobby⇄match interface (constructor opts, `start/handle/onDisconnect/onReconnect/onLeave/dispose`) is documented verbatim at the top of `server/match/Match.js` and is normative. `server/data.js` provides `getData()` + index getters (`getChess`, `getBond`, `getGarrison`, `getItem`, `getBand`, `getEffect`, `getEnemy`, `getWave`, `getStage`, `getBoss`, `getToken`, `getConfig`, `getMode`). Note: the skill trigger rule lives at `chess.skill.trigger.rule` in data.

-**Spectator seats** (a remake feature — the official room has none; community report #26, owner's decision 2026-10-04; the spectator-seat subsection of §23): `room.spectate {code}` takes one of a co-op room's `MAX_SPECTATORS` (2) spectator seats, in its lobby or during its match; `room.removeSpectator {playerId}` (host) frees one (`room.closed {reason:'kicked'}` to it); `room.state.spectators: [{playerId, name, connected}]`. A spectator is never a player (not in `seats`, not counted for the start, never host, keeps no room alive — `room.closed {empty}` when the last human leaves), receives the room's broadcasts and watches like an eliminated player (`Match.addSpectator`: `b.start` watch / `m.field`, never `m.private`); it may only `g.watch`, leave, and store a `room.loadout` — anything else answers `error SPECTATOR`. Its seat is kept and given back on reconnect like a player's.
+**Spectator seats** (a remake feature — the official room has none; community report #26, owner's decision 2026-10-04; the spectator-seat subsection of §23): `room.spectate {code}` takes one of a co-op room's `MAX_SPECTATORS` (2) spectator seats, in its lobby or during its match; `room.removeSpectator {playerId}` (host, at any time — the room screen's strip and, once the match runs, the game screen's 观战席 capsule in the top bar, GitHub #120) frees one (`room.closed {reason:'kicked'}` to it, no match frame after it); `room.state.spectators: [{playerId, name, connected}]` (still pushed while the match runs). A spectator is never a player (not in `seats`, not counted for the start, never host, keeps no room alive — `room.closed {empty}` when the last human leaves), receives the room's broadcasts and watches like an eliminated player (`Match.addSpectator`: `b.start` watch / `m.field`, never `m.private`); it may only `g.watch`, leave, and store a `room.loadout` — anything else answers `error SPECTATOR`. Its seat is kept and given back on reconnect like a player's.

 ### 8.2 Match
-C→S (all carry `rid`): `g.infoReady` · `g.band {bandId}` · `g.bandSkip` · `g.bandFocus {bandId?}` (the strategy highlighted in the draft screen — what a timed-out turn takes; absent / null clears it; `WRONG_PHASE` outside BAND_DRAFT, `ALREADY` after the pick, `BAD_TARGET` for a band not allowed; §18.2) · `g.buy {slot}` · `g.refresh` · `g.freeze` · `g.levelUp` · `g.sell {uid}` · `g.move {uid, to:{area:'board', row, col}|{area:'hand', idx}, dir?}` · `g.equip {itemUid, targetUid, replaceUid?}` (`replaceUid`: which equipped item a third one replaces, §6.2) · `g.art {itemUid, row, col, dir?}` (`dir` ∈ UP|RIGHT|DOWN|LEFT from the deploy wheel, §3; absent ⇒ `to.dir`, then RIGHT; `g.move` onto the piece's own tile re-orients it; board pieces in `m.private` / `m.field {prep:true}` carry `dir`) · `g.destroy {uid}` (hand / temp items only; equipped items are locked, §6.2) · `g.reward {idx}` · `g.choice {idx}` · `g.ready {ready}` · `g.emote {id}` · `g.watch {fieldId, playerId?}` (`playerId`: the player tapped — a shared 联防 / boss pair field shows two; what an eliminated viewer follows, §25) · `g.autoplay {on}` · `g.pause {on}` (solo battles only, §14 Solo pause) · `g.unitStats {seq?}` (ROUND_START / SP_DRAFT / PREP, else `WRONG_PHASE`; eliminated: `ELIMINATED`) → push `m.unitStats` (§18.5) · `g.leave`
+C→S (all carry `rid`): `g.infoReady` · `g.band {bandId}` · `g.bandSkip` · `g.bandFocus {bandId?}` (the strategy highlighted in the draft screen — what a timed-out turn takes; absent / null clears it; `WRONG_PHASE` outside BAND_DRAFT, `ALREADY` after the pick, `BAD_TARGET` for a band not allowed; §18.2) · `g.buy {slot}` · `g.refresh` · `g.freeze` · `g.levelUp` · `g.sell {uid}` · `g.move {uid, to:{area:'board', row, col}|{area:'hand', idx}, dir?}` · `g.equip {itemUid, targetUid, replaceUid?}` (`replaceUid`: which equipped item a third one replaces, §6.2) · `g.art {itemUid, row, col, dir?}` (`dir` ∈ UP|RIGHT|DOWN|LEFT from the deploy wheel, §3; absent ⇒ `to.dir`, then RIGHT; `g.move` onto the piece's own tile re-orients it; board pieces in `m.private` / `m.field {prep:true}` carry `dir`) · `g.destroy {uid}` (hand / temp items only; equipped items are locked, §6.2) · `g.reward {idx}` · `g.choice {idx, choiceId?}` (with an ID: the recipient's PREP personal choice; without one: the public SP_DRAFT; stale or wrong IDs never fall through to the public draft) · `g.ready {ready}` · `g.emote {id}` · `g.watch {fieldId, playerId?}` (`playerId`: the player tapped — a shared 联防 / boss pair field shows two; what an eliminated viewer follows, §25) · `g.autoplay {on}` · `g.pause {on}` (solo battles only, §14 Solo pause) · `g.unitStats {seq?}` (ROUND_START / SP_DRAFT / PREP, else `WRONG_PHASE`; eliminated: `ELIMINATED`) → push `m.unitStats` (§18.5) · `g.leave`
 S→C:
- `m.public` — full public state on every change (throttled ≤ 10/s): `{ phase, round, lastRound, deadline, serverNow, modeId, difficulty, stageId, factions, disabledBonds, bannedChess, bossId, hiddenBossId?, teamLp?, bossHp?:{hp,max}, overtimeAt? /* FA / HC: ms epoch the overtime drain starts, §4 */, paused /* solo pause, §14 */, draft?:{order,turn,picks,skipsLeft,turnDeadline,turnSeconds /* one turn's length, 0 untimed; deadline = turnDeadline (§18.2) */,untimed}, sp?:{family,cards /* bounty cards carry descRaw?, §20.6 */,turn,picks}, players:[{playerId, seat, name, isBot, connected, alive, lp, bandId, shopLevel, boardCount, ready, bonds:[{bondId,count,active,tier,layers,harmony? /* 调和's +1 is in count, §21.26 */}], fieldId, status, pendingLp? /* COMBAT / 联防 of a normal round: the LP the player's own battle will cost at settlement so far, min(lpCapPerRound, counted leaks) — in 联防 min(lpCapPerRound, uniteLeft) — omitted when 0 and in every other phase (§17.5 live LP) */, uniteLeft? /* 联防: the leaker's enemies still standing — not spawned yet + alive + counted leaks, uncapped, live (§20.6) */}] , fields:[{fieldId, kind, players:[playerId...]}] }` · `m.public.uniteResult` — SETTLE after a 联防 only, its outcome as data: `{ through /* the leakers' enemies that still got through, uncapped */, helpers:[playerId...], leakers:[playerId...], losses:{playerId: the LP settlement charged this round — every alive player} }`; absent when no 联防 resolved (the client's result box then reads the round's own battle; GitHub #235, PR #112)
+- `m.public` — full public state on every change (throttled ≤ 10/s): `{ phase, round, lastRound, deadline, serverNow, modeId, difficulty, stageId, factions, disabledBonds, bannedChess, bossId, hiddenBossId?, teamLp?, bossHp?:{hp,max}, overtimeAt? /* FA / HC: ms epoch the overtime drain starts, §4 */, paused /* solo pause, §14 */, draft?:{order,turn,picks,skipsLeft,turnDeadline,turnSeconds /* one turn's length, 0 untimed; deadline = turnDeadline (§18.2) */,untimed}, sp?:{family,cards /* bounty cards carry descRaw?, §20.6 */,turn,picks}, players:[{playerId, seat, name, isBot, connected, alive, lp, bandId, shopLevel, boardCount, ready, bonds:[{bondId,count,active,tier,layers,harmony? /* 调和's +1 is in count, §21.26 */}], fieldId, status, pendingLp? /* COMBAT / 联防 of a normal round: the LP the player's own battle will cost at settlement so far, min(lpCapPerRound, counted leaks) — in 联防 min(lpCapPerRound, uniteLeft) — omitted when 0 and in every other phase (§17.5 live LP) */, uniteLeft? /* 联防: the leaker's enemies still standing — not spawned yet + alive + counted leaks, uncapped, live (§20.6) */}] , fields:[{fieldId, kind, players:[playerId...], progress?:{killed, resolved, total, done} /* §14 顶栏胶囊: resolved = the field's own scheduled enemies knocked down or leaked, null while unknown */}] }` · `m.public.uniteResult` — SETTLE after a 联防 only, its outcome as data: `{ through /* the leakers' enemies that still got through, uncapped */, helpers:[playerId...], leakers:[playerId...], losses:{playerId: the LP settlement charged this round — every alive player} }`; absent when no 联防 resolved (the client's result box then reads the round's own battle; GitHub #235, PR #112)
 - `m.private` — the recipient's full PlayerState view (§6.2 minus server-only data) on every change; in a boss round's prep also `bossMate { playerId, side, units }` — the pair partner's board on its half of the boss field (UnitInfo, mirrored on the right half), re-sent when the partner moves (0.2.0, community report item 51).
 - `m.field` — field meta when watching starts: `{ fieldId, kind, rect, stageId, units: [UnitInfo], effects? }` (the units on the field, knocked-out operators waiting to redeploy included — `Battle.fieldMeta`, §18.3). The prep-scouting field (`Match.prepFieldMeta`) puts the scouted player's **hand (整备区) and 临时整备区 into `units` as well**: held pieces as units on the hand row (row 7, col = hand slot) and temp pieces on the temp row (row 8, cols 4..8 = the temp slots; PRTS 帮助 counts the temp area with the hand; a held item is a `kind: 'item'` unit the client draws as its floating plate) — the scout renders exactly like the own prep bench, for every watcher alike, the spectator seat's copy equal to a teammate's (§23.19's "never a private view" stays the no-`m.private` rule); the scout signature `_prepScoutSig` carries the hand / temp slots, so a piece moved to another slot is a change — and carries `effects`: the scouted player's effects view (策略 / 机变 / 悬赏 …), the watched player's effects column ready-made
 - NOTE: game time in `b.snap`/`b.ev` is carried as **`gt`** (the frame's `t` is the message type).
 - Active finite zero-SP timed skills display `sp=timeLeft`, `spMax=duration` and `UF.SKILL`; after end they display
  `0/0`. Snapshot display capacity does not change runtime SP/cost.
- `b.snap` — `{ fieldId, gt, units: [[id, x, y, hp, maxHp, sp, spMax, flags, anim]] , dp, killed, total, down?, elem? }` 20 Hz (a countdown summon's `hp` is maxHp × the share of its life left, §25.16.5). Optional, only when non-empty (§18.3): `down: [[id, respawnAt (game s), respawnTime (s), state, row, col]]` — knocked-out operators waiting to redeploy, `state` 0 counting | 1 timer done, waiting for DP | 2 timer done, its tile taken — a safeguard, no ally deploys there (sim/constants.js `DOWN_STATE`); `row, col` = the tile it lies on and comes back on (§21.24); `elem: [[id, element, fill, cooldownEnd (game s), cooldown (s)]]` — the one gauge a unit shows: the fullest (ties by the official element id 神经 < 侵蚀 < 灼燃 < 凋亡), fill with 2 decimals rounded DOWN and kept at 0.01–0.99 outside a burst (§20.8); during a 爆发冷却 the bursting element with fill 1 and the cooldown numbers (else 0, 0)
+- `b.snap` — `{ fieldId, gt, units: [[id, x, y, hp, maxHp, sp, spMax, flags, anim]] , dp, killed, total, resolved, down?, elem?, ammo?, wolves?, neg?, stand?, standCut? }` 20 Hz (a countdown summon's `hp` is maxHp × the share of its life left, §25.16.5). Optional, only when non-empty (§18.3): `down: [[id, respawnAt (game s), respawnTime (s), state, row, col]]` — knocked-out operators waiting to redeploy, `state` 0 counting | 1 timer done, waiting for DP | 2 timer done, its tile taken — a safeguard, no ally deploys there (sim/constants.js `DOWN_STATE`); `row, col` = the tile it lies on and comes back on (§21.24); `elem: [[id, element, fill, cooldownEnd (game s), cooldown (s)]]` — the one gauge a unit shows: the fullest (ties by the official element id 神经 < 侵蚀 < 灼燃 < 凋亡), fill with 2 decimals rounded DOWN and kept at 0.01–0.99 outside a burst (§20.8); during a 爆发冷却 the bursting element with fill 1 and the cooldown numbers (else 0, 0); the three HP-bar readouts (0.2.2, GitHub PR #303 / #378 / #383, display only — no sim state reads them, the nine-field tuples and the SP fraction an ammo skill still puts in them are unchanged; `render/interp.js` keeps them per snapshot and `sample()` hands them out from the older snapshot like `flags`, so they step with the skill flag; a malformed entry or an id the snapshot does not list is dropped, a snapshot without the list clears the readout): `ammo: [[id, rounds left, rounds in the magazine]]` — an ally's running ammo skill, whole rounds, the magazine being the activation's real total (extra bullets and refills included; 隐现's S2 holds 14, skill_table `attack@trigger_time`) — the client draws yellow cells in place of the SP bar (the count, not a pixel copy of the official magazine widget — its art is in no game data we have), and UnitInfo `ammoSkill: true` (an ally whose skill is an ammo magazine, known from its first appearance) tells it to draw no standing skill aura; `wolves: [[id, 狼影 left, the talent's maximum]]` — 伺夜's 狼群 (`mem.shadows` / `mem.wolfCapacity`, both the placed and the automatic pack), one diamond per slot under the HP bar (plain diamonds: no official counter art is known, the portrait of PR #378 was not taken), nothing in the 战术点形态; `neg: [[id, fill]]` — the share (0.01–1) of its cap a negative-HP pool holds (斩业星熊's 我执, `unit.negFill`), drawn as the red bar over her drained HP bar. These show only for a living, deployed, visible unit and only when it has one
+  Optional, only when non-empty (PR #381; display metadata, SIM §9): `stand: [[id, until]]` — the game time an enemy's
+  attack recovery ends (`atkStandUntil`; enemies alive, deployed, visible, neither feared nor stunned): when it falls
+  between two snapshots the renderer holds the position until then and interpolates the rest (HP / SP unchanged), and
+  extrapolates nothing while the newest snapshot still holds a stand; `standCut: [[id, at]]` — the latest time each
+  listed enemy's recovery was cut or ignored (失衡, fear, stun, hiding, leaving): an interval with a cut inside it stays
+  linear. A newest snapshot showing a death, a stun or a block is not extrapolated. Old snapshots without them
+  interpolate linearly; nothing is inferred from the attack animation.
 - `b.ev` — `{ fieldId, ev: [ ... ] }` event tuples: `['spawn', UnitInfo]`, `['atk', srcId, tgtId, projKind]`, `['dmg', tgtId, amount, type]`, `['heal', tgtId, amount]`, `['skill', id, on:0|1]`, `['die', id, reason]` (reason `'killed'`, `'retreat'`, `'expired'` …; only `'killed'` plays an operator's knock-down sound), `['status', id, key, on]`, `['fx', kind, x, y, extra]` (`'hitCap' {id, n}` = a cancelled 限伤 hit, drawn as nothing, §20.12; `extra.form` = the unit's model form from then on (an enemy's mode, a 傀儡师's 替身 §22.11) — `shared/protocol.js fxForm`, state like a spawn: never dropped by the client, §21.4), `['layer', playerId, bondId, n]`, `['bounty', playerId, coins]`, `['deploy', id]`
 - `m.toast {kind, text}` · `m.ticker {text, id, type, priority, playerId}` (the strip plays the highest priority first, §21.10) · `m.emote {playerId, id}` · `m.result {...}`
- `m.unitStats {seq, round, units: [unitStatsEntry]}` — the answer to `g.unitStats` (`seq` echoed): every own board operator / summon by uid with `{id, uid, defId, hp, alive, maxHp, atk, def, res, interval, blockCnt, moveSpeed, range?, base: {…same}}` (shared/protocol.js `unitStatsEntry`; `range` = an ally's live range grid, facing RIGHT, never a target-selection grid, §21.16) as its next battle starts (§18.5)
+- `m.unitStats {seq, round, units: [unitStatsEntry]}` — the answer to `g.unitStats` (`seq` echoed): every own board operator / summon by uid with `{id, uid, defId, hp, alive, maxHp, atk, def, res, interval, blockCnt, moveSpeed, range?, base: {…same}}` and `dir?` (shared/protocol.js `unitStatsEntry`; `range` = an ally's live range grid, facing RIGHT, never a target-selection grid, §21.16; `dir` = the unit's facing now, UP|RIGHT|DOWN|LEFT — the snapshot tuples carry none and UnitInfo `dir` is the facing at send time — which the detail card's range overlay rotates by, GitHub PR #281) as its next battle starts (§18.5)

 `UnitInfo = { id, kind, side, ownerId, defId, name, tier, golden, spine, avatar, x, y, facing, dir, maxHp, skillIndex?, moduleId?, items?, form? }` (`dir` = the ally's deploy direction; the renderer picks the model (Back for UP, mirrored for LEFT) and the ground wedge from it; `skillIndex` = the ally's equipped skill (§16): the renderer plays that skill's Spine clip (`anims.skills[i]`) and the audio its own ON_SKILL_START sound; `moduleId` = an elite ally's equipped module (uniEquipId | 'none'); `items` = an ally operator's equipped item ids (absent without any; never on enemies or summons): the bond popup lists a 变形同构体 wearer from them and the detail card shows its 同构 chip and 装备 (§21.11); all three also on the prep-scouting `m.field` units (`Match.prepFieldMeta`, the held pieces included), so a teammate's unit shows its owner's loadout in the detail card and its equipped items as item pips on the unit (prep surfaces; user playtest #2 item 3); `form` = the unit's current model form (an enemy's mode §21.4, a 傀儡师's 替身 §22.11; a view built mid-battle starts in it)).

@@ -34,9 +49,10 @@ S→C:

 `m.private`:
 ```js
-{ playerId, seat, alive, lp, funds, bandId, ready, canReady /* false while temp non-empty */,
+{ playerId, seat, alive, lp, funds, bandId, ready, canReady /* false while a personal choice or temp overflow is pending */,
+  personalChoice: null | { id, round, sourceItemId, cards: [ { kind: 'bounty', id, name, desc, descRaw, ... } ] }, // recipient only, including reconnect; absent from public / scout views
  shop: { level, maxLevel: 6, upgradePrice, refreshPrice /* effective (0 when a free refresh is available) */, freeRefreshes,
-          frozen, slots: [ { kind: 'chess'|'item', id, price, basePrice, sold } | null ],
+          frozen, slots: [ { kind: 'chess'|'item', id, price, basePrice, sold, frozen } | null ],  // slot.frozen: the 冻结 toggle copies onto every unsold slot, and 梓兰's 猎头顾问 freezes ONE card of an active refresh with the toggle off — every card draws its own flag (GitHub #354)
          rewardOffer: null | { tier, source: 'merge'|'special', label, queued, slots: [ { kind: 'chess'|'item', id, price: 0, sold } ] } },  // §21.2
  hand:  [ Piece | null ],   // length 10, index = hand slot
  temp:  [ Piece | null ],   // length 5
@@ -70,20 +86,28 @@ Piece = { uid, kind: 'chess'|'item'|'token', id, golden: bool, tier, items: [ {
 **Authority per field.**
 - Normal field: the owning human's client is **authoritative** (if connected).
 - Unite field (联防) and boss/hidden pair fields: the lowest-seat **connected human** in that field is authoritative; every other human in (or watching) that field receives the same spec and simulates it locally **for display only**.
- Bots, departed/disconnected players, fields with no connected human: the **server** simulates. Normal and 联防 fields headlessly (`fields.js HeadlessJob`, with a `[gt, killed, total(, left)]` progress timeline — `left`: the 联防 leakers' enemies still standing, §20.6) in wall-clock slices of `headlessSliceMs` (8 ms) per callback on a real host — never one blocking run per field (3 bot fields ≈ 0.2–0.5 s of CPU at COMBAT start here, seconds on a mini PC); at once in virtual time; the result is released at the battle's natural end on the field clock (so round pacing and the teammates' progress look live). Boss fields in real time without snapshots (`HeadlessPacer`); when no field of a boss round has a connected human the whole round runs server-side as before (FieldRunner with streaming off).
+- Bots, departed/disconnected players, fields with no connected human: the **server** simulates. Normal and 联防 fields headlessly (`fields.js HeadlessJob`, with a `[gt, killed, total, resolved(, left)]` progress timeline — `resolved`: the capsule's numerator on the field clock (the maintainer's review of PR #157: a bot field has no authority reporting `b.progress`), `left`: the 联防 leakers' enemies still standing, §20.6) in wall-clock slices of `headlessSliceMs` (8 ms) per callback on a real host — never one blocking run per field (3 bot fields ≈ 0.2–0.5 s of CPU at COMBAT start here, seconds on a mini PC); at once in virtual time; the result is released at the battle's natural end on the field clock (so round pacing and the teammates' progress look live). Boss fields in real time without snapshots (`HeadlessPacer`); when no field of a boss round has a connected human the whole round runs server-side as before (FieldRunner with streaming off).
 - Takeover: an authority that disconnects or leaves, or misses its deadline (`timeLimit / speed` real s + 15 s grace, normal/联防) loses the field: normal/联防 → the server re-simulates the spec from t = 0 (deterministic: the same result); boss → the partner's replica is promoted (`b.start authoritative: true`), else the server runs it, fast-forwarded to the field clock through a `CreditPool` that only credits damage beyond what the client already reported (the fast-forward is spread over the pacing intervals, ≤ 240 extra ticks each). A boss authority silent for 12 s is handed over the same way. The former authority gets `b.end { reason: 'takeover' }` and keeps displaying.

 **Protocol (additions; `shared/protocol.js` is normative).**
 - `battleId` = `<prefix>.<round>.<seq>.<fieldId>` (`Match._ccField`; `<prefix>` = the match seed in base 36 plus `-<matchNo>`, the room's match number the lobby passes as `opts.matchNo`; `<seq>` counts every battle of the match; the informational `.<fieldId>` is dropped when the id would exceed the protocol's 64 chars): unique for the whole match although field ids repeat (`n:<pid>` every round, `u`, `b1`/`b2` in both boss rounds), and across the matches of one room. The client runner keys its local battles by it (the same id again = the same battle, no rebuild) and the server matches reports by it, so a late `b.progress` / `b.result` of a finished battle — or one that crossed into the room's next match on the same socket — is ignored instead of being credited to another battle.
 - S→C `b.start { battleId, fieldId, kind, spec, authoritative, startAt, serverNow, elapsed, speed, watch, done }` — sent at COMBAT/UNITE/FINAL_ASSAULT/HIDDEN start to every human in the field (联防: to every human; COMBAT and boss rounds: eliminated humans and spectator seats get the field of the player they follow as a display replica — the one they last watched with `g.watch`, else the first player still in; research 09 "keep-watching auto-observes the first available field", 0.2.0 item 56), on `g.watch`, and on reconnect/resync. `elapsed` = game seconds on the field clock (a finished field: its end time); the client fast-forwards to it. `speed` = game seconds per real second (2; tests/tools may speed up).
- C→S `b.progress { battleId, gt, killed, total, leaks?, left?, bossDmg?, by?, done? }` — authoritative clients, ~1 Hz (boss fields 4 Hz). Normal/联防: `leaks` = counted leaks so far; 联防: `left` = per leaker its enemies still standing (`spec.js uniteLeft`, clamped by the server to `fields.js uniteBillBounds`; §20.6). Boss: `bossDmg` = cumulative damage of this field to the shared pool, `by` = the same per player, `leaks` = the field's cumulative LP cost (leaks × `lpr` + leader "扣除目标生命" effects; `spec.js attachLpMeter`). The server applies deltas (never twice). Stale reports (unknown battle, not the authority, after the match ended) are ignored, never answered with an error; the platform never answers a rid-less `b.progress` with an error frame (it would surface as a toast).
+- C→S `b.progress { battleId, gt, killed, total, resolved?, leaks?, left?, bossDmg?, by?, leaksBy?, done? }` — authoritative clients, ~1 Hz (boss fields 4 Hz). `resolved` = the HUD capsule's numerator of this field (below; omitted by a client that cannot count it — the teammate UI then falls back to `killed`); `killed` / `total` are the field's own counters: `killed` counts every `counted` knock-out (a runtime split / summon included) and may exceed `total`, which counts only the enemies the round scheduled. Normal/联防: `leaks` = counted leaks so far; 联防: `left` = per leaker its enemies still standing (`spec.js uniteLeft`, clamped by the server to `fields.js uniteBillBounds`; §20.6). Boss: `bossDmg` = cumulative damage of this field to the shared pool, `by` = the same per player, `leaks` = the field's cumulative LP cost (leaks × `lpr` + leader "扣除目标生命" effects; `spec.js attachLpMeter`), `leaksBy` = its split per player (the LP the enemies that reached that player's goal cost: every leak entry the result will show × its `lpr`; `leaks` minus its sum is the leader's effects; `spec.js battleProgress`) — the server keeps the highest value per player and, before a perfect-payout bounty pays in a client-reported field, requires the result's leaked list of that player to show at least that much and the LP asked for to have been split at all (`Match._bossLeaksAgree`, the boss path does not re-simulate). The server applies deltas (never twice). Stale reports (unknown battle, not the authority, after the match ended) are ignored, never answered with an error; the platform never answers a rid-less `b.progress` with an error frame (it would surface as a toast).
 - C→S `b.result { battleId, result }` — authoritative client at battle end; `result` = `compactResult(BattleResult)` (the fields the match consumes; ≤ 4 players, ≤ 400 leaks, ≤ 64 unitsEnd, ≤ 160 unitStats, ≤ 40 layer gains; `fitResult` keeps the frame ≤ 60 KB: unit statistics, then leak mods, then — boss only — leak entries go first). Structural checks in `isBattleResult`; `server/net.js` routes `b.*` frames to the match (lobby `routeGame`).
 - S→C `b.pool { hp, max, teamLp, acked: { [fieldId]: cumulative damage counted } }` (boss rounds, ≤ 4 Hz; exact numbers, never rounded) — clients reconcile their `LocalBossPool`: displayed pool = server hp − (local cumulative damage − acked). Pools never hold less than 1 HP (`BOSS_POOL_MIN_HP`: `SharedBossPool` and `LocalBossPool`; a browser pool reads any finite rest below 1 as 0 — §20.5); the 'cleared'-result dust rule (a boss field whose client result is `'cleared'` while < 1 HP is left on the server empties the pool) only catches noise at that boundary.
 - S→C `b.end { battleId, fieldId, reason }` — `'cleared'` (pool 0) / `'forced'` (team LP 0): the client calls `battle.forceEnd` and reports; `'takeover'`: stop reporting, keep displaying. Both also reach a battle the client is still preparing (the sim loading, the silent catch-up): Client runner below.
- `m.public.fields[]` gains `progress: { killed, total, done }`; `m.public.combatMode`; `m.public.paused: boolean` (always present; solo pause below).
+- `m.public.fields[]` gains `progress: { killed, resolved, total, done }`; `m.public.combatMode`; `m.public.paused: boolean` (always present; solo pause below).
 - `b.snap` / `b.ev` are **not sent for combat** in client mode. `g.watch` rules: see Spectating.

-**Result validation (server, `fields.js validateClientResult`).** A client result is accepted only if structurally valid and plausible for its spec: every spec player reported and nobody else; finite numbers; `killed ≤ total ≤ 4·spawns + 100`; every leaked `enemyKey` is in the spec's spawns (multiset bound) or a key the spawned enemies' data mentions (content summons/splits, bounded by the total) — boss fields excepted (their leaks cost team LP through `b.progress`); `counted: false` only for enemies that never count (data `notCountInTotal`, schedule `countInTotal: false`, boss/part entries, content spawns); 联防: leaks + `unspawned` per (enemy, leaker) ≤ what that leaker sent in (nobody's survivors are billed to a teammate); per-bond `layerGains ≤ min(60 + 4·round, 999 − the bond's starting layers)` (§20.12) — the flat part raised by what the player's layer 特质 can add to that bond (their per-battle caps; an uncapped one leaves only the 999 room, §21.26) —, none when the spec disables gains, bonds must exist and be named by the player's bond snapshot, band, effects, units or items; `coins ≤ Σ bounty coins of the spawns`; `perfect` consistent with the counted leaks; unit states only for the player's own board uids with hpPct ∈ [0, 1]. The accepted result is **rebuilt** from whitelisted fields (leak mods/tags/sources from the spec — a content-spawned enemy (split, summon) keeps the mods of the schedule entry they equal, i.e. its parent's round multipliers and bounty id, so its 联防 re-entry and bounty match the server run; unit names from server data). Otherwise (or on any exception) the server re-simulates (normal/联防); a boss field is **handed over** instead (below). `SP_VERIFY=all|sample|off` (default `off`): `all` re-simulates every accepted normal/联防 result before accepting it (the server result wins on a digest mismatch); `sample` checks ~1 battle in 8 in a later callback and logs mismatches. `Match.verifyStats = { checked, mismatches, rejected, takeovers }`. Desync between display-only clients and the authority is tolerated: the authoritative outcome is what counts and is broadcast at settlement.
+**顶栏胶囊 (the top bar's 击倒 / 漏掉 counter; PR #157, official reading).** The capsule shows `resolved / total`:
+- `total` = the enemies **this field itself scheduled** (`Battle` `total`: the `spawns` the match built, `inTotal`) that count — `countInTotal: false`, `tag: 'boss'` / `'part'` and data `notCountInTotal` entries stay out, like the 领袖 of a boss field.
+- `resolved` = `min(total, killedInTotal + leakedInTotal)` (`Battle.resolved`): that same set's enemies that are 已解决 — knocked down (`killedInTotal`) **or** leaked (`leakedInTotal`). Official: 开局 0/3 → 漏一个 1/3 → 打死一个 2/3 → 打死会分裂的 3/3.
+- Everything spawned while the battle runs is in **neither** part: a split child (磨砻's 木制瑞印), a summon (枯朽萃聚使徒's 枯朽之种, a boss's 余音 / 妖怪 / 不祥幻影 / 王权号令 equipment), a 变身 copy, a part. They still count for LP, 完美作战 and `killed` exactly as master did (`counted`): a leaked split child costs its `lpr` and breaks 完美作战, it just moves no capsule number.
+- `killed` (`Battle.killed`, per player too) keeps the official `counted` reading — a runtime split / summon knocked down counts — so it may **exceed** `total` (4 磨砻 killed plus one of their 8 children reads `killed` 5 with `total` 4 while the capsule reads 4/4). Nothing clamps it: the match's stats and the bot's scoring read those knock-outs.
+- Carried by `b.snap`, `BattleResult` (battle level and per player, the per-player `resolved` clamped to the player's `total`), `b.progress` and `m.public.fields[].progress`. **A finished field publishes the field's own number, never the sum of the players':** in a two-half field (the 联防 lane, the boss pair's crossing routes) an enemy that spawns on one half and leaks on the other is billed to one player's `total` and to the other's `leakedInTotal`, so each player's own `min(total, …)` reads 0 for an enemy the field resolved (capsule 1/1, players 0 and 0). `validateClientResult` therefore keeps the result's field-level `resolved` (the client's `Battle.resolved`, clamped to the field's `total`) and `_fieldProgress` publishes it — for a server-run field too, whose `Battle.result()` carries the same number; the per-player numbers are only the fallback of a result that has none, and with none at all `progress.resolved` is `null`. A bot / server-run field has no authority reporting `b.progress`, so its capsule reads the battle's own counters through the `HeadlessJob` timeline sample (`[gt, killed, total, resolved(, left)]`), never the report-driven `progress.leaks` (which stays 0 there); `progress.resolved` starts at `null` and is only adopted from a real reported integer — a reported `0` is a value, "not reported" is not.
+- The HUD reads it as `resolved ?? killed` (`ui/hud.js` two capsules, `ui/combatHud.js` teammates' rows, `ui/gameLogic/format.js snapHud`, `battle/observe.js teammateProgress`), so a server that does not send `resolved` displays master's numbers unchanged.
+
+**Result validation (server, `fields.js validateClientResult`).** A client result is accepted only if structurally valid and plausible for its spec: every spec player reported and nobody else; finite numbers; `killed ≤ maxTotal` and `total ≤ maxTotal` (`4·spawns + 100`, the content-spawn allowance — `killed ≤ total` no longer holds, see the capsule above) and the field-level `resolved ≤ total`; every leaked `enemyKey` is in the spec's spawns (multiset bound) or a key the spawned enemies' data mentions (content summons/splits, bounded by the total) — boss fields excepted (their leaks cost team LP through `b.progress`); `counted: false` only for enemies that never count (data `notCountInTotal`, schedule `countInTotal: false`, boss/part entries, content spawns); 联防: leaks + `unspawned` per (enemy, leaker) ≤ what that leaker sent in (nobody's survivors are billed to a teammate); per-bond `layerGains ≤ min(60 + 4·round, 999 − the bond's starting layers)` (§20.12) — the flat part raised by what the player's layer 特质 can add to that bond (their per-battle caps; an uncapped one leaves only the 999 room, §21.26) —, none when the spec disables gains, bonds must exist and be named by the player's bond snapshot, band, effects, units or items; `coins ≤ Σ bounty coins of the spawns`; `perfect` consistent with the counted leaks; unit states only for the player's own board uids with hpPct ∈ [0, 1]. The accepted result is **rebuilt** from whitelisted fields (leak mods/tags/sources from the spec — a content-spawned enemy (split, summon) keeps the mods of the schedule entry they equal, i.e. its parent's round multipliers and bounty id, so its 联防 re-entry and bounty match the server run; unit names from server data). Otherwise (or on any exception) the server re-simulates (normal/联防); a boss field is **handed over** instead (below). `SP_VERIFY=all|sample|off` (default `off`): `all` re-simulates every accepted normal/联防 result before accepting it (the server result wins on a digest mismatch); `sample` checks ~1 battle in 8 in a later callback and logs mismatches. `Match.verifyStats = { checked, mismatches, rejected, takeovers }`. Desync between display-only clients and the authority is tolerated: the authoritative outcome is what counts and is broadcast at settlement.

 **Boss rounds.** The shared pool (`SharedBossPool`, size = `bloodPoint[difficulty]` × the players alive when the fight starts, solo × 1 — the owner's decision of 2026-10-06 adopting PR #209, which replaces the fixed pool of 「保持固定血量」, §25.13.4) lives on the server and is fed by the authorities' `b.progress` damage deltas (attributed per player through `by`; BOSS_HIT tickers from each player's damage to that round's pool, §21.10) and by server-run fields; a server clock (250 ms) applies the overtime drain (−`bossOvertimeDrainPerSec` team LP per whole **real** second past `bossOvertimeAfter` real s, measured on the field clock — `gamedata.js bossOvertimeDue`, §4), the silence watchdog and the end checks: pool 0 → `b.end cleared` (victory), team LP 0 → `b.end forced` (defeat) — the first one the server registers decides the verdict (`Match._finalEnding`) and after a forced end no report (an in-flight `b.progress`, the final `b.result`) credits the pool (§20.5); results are awaited ≤ 6 s (× timerScale) after a forced end. `b.pool` is the live channel for the pool and the team LP: a boss `b.progress` (4 Hz per field) and the team-LP losses never re-mark the whole public state — `m.public` (`bossHp`, `teamLp`, `fields[].progress`) and the players' merged-LP shares (`_syncTeamLp`) refresh at ~1 Hz (`Match._bossPublic`, `BOSS_PUBLIC_MS`), and at once when the round ends. A boss field's `b.result` ends it only when the shared pool is empty or after that `b.end` — except a `'cleared'` result whose cumulative report covers what the pool still holds (within its 1 HP floor) while the plausibility budget (one field credits ≤ the pool per `BOSS_MIN_CLEAR_GS` 5 game s) has not caught up: it is **held** (`f.heldResult`; 999-layer boards kill a leader in 2–4 game s, §20.14) — the boss clock keeps crediting the report at the budget's rate, the silence watchdog and handovers skip the field, a re-sent copy is ignored, and the pool's end (`_endFinal`) completes the field with it; a forged report gains nothing a `b.progress` could not already get. Any other result — `'forced'` / `'timeout'` at t = 0, `'cleared'` while the pool still holds more than the report covers — and any result failing validation would stop the pair's fight (and, with every field done, end the Final Assault as a defeat), so the field is **handed over** (`Match._bossHandover`) to the partner's display replica (promoted with `b.start authoritative: true`) or, without one, to the server (fast-forwarded to the field clock); the sender gets `b.end takeover` and is never that field's authority again (`f.demoted`). There is no synthetic zero result for a running boss field: a stats-only stand-in (`fields.js syntheticResult`, carrying the damage already credited, never charging LP) only fills a result that cannot be had otherwise — the Final Assault already ending (no result, or an invalid one, within the grace) or a server-run battle whose `result()` threw.

@@ -95,7 +119,7 @@ Piece = { uid, kind: 'chess'|'item'|'token', id, golden: bool, tier, items: [ {

 **Corrections from research 08/09 (normative):**
 - Waves: official generator (research 08 §2): 3 of 6 types, each owns exactly 3 of the 15 round slots, SPECIAL the rest (shuffled per match); per round one type + one weighted special entry (half by round); each template SPAWN **action** becomes `clamp(roundHalfEven(n·P(t)/P(new)), 1, 5)` units over the same window; placeholder actions of the other movement class are skipped. Stage per match from the mode's list (标准 = 战场#01 only, §6.5). No custom balance multipliers (`data/tuning.json` keeps only titles); co-op boss pool = `bloodPoint[difficulty]` unscaled.
- Ground pathing: official SPFA-style 4-direction search from the goal with crates as cost-1000 tiles, then straight-line smoothing (research 08 §3.4). act1 m02 crates #001–#004 inactive in 下半. Remake refinement (user playtest #2, narrowed by §21.15): route lengths and routes stay the official ones unless 0.1.0's road-over-floor preference route crosses strictly fewer non-blockable (floor / gate / 深水区) tiles — a corner touch is not a crossing; ties keep the official route. E.g. 战场#01's lower-gate enemies take the col-8 road, while 战场#04's keep the official diagonal from row 9 into row 10 (server/sim/grid.js header; test/sim/pathing-official.test.js, pathing-blockable.test.js).
+- Ground pathing: official SPFA-style 4-direction search from the goal with crates as cost-1000 tiles, then straight-line smoothing (research 08 §3.4). act1 m02 crates #001–#004 inactive in 下半. Remake refinement (user playtest #2, narrowed by §21.15): route lengths and routes stay the official ones unless 0.1.0's road-over-floor preference route crosses strictly fewer non-blockable (floor / gate / 深水区) tiles and no more 深水区 (GitHub #375) — a corner touch is not a crossing; ties keep the official route. E.g. 战场#01's lower-gate enemies take the col-8 road, while 战场#04's keep the official diagonal from row 9 into row 10 (server/sim/grid.js header; test/sim/pathing-official.test.js, pathing-blockable.test.js).
 - Observing: no watching while your own normal battle runs; after it ends (and in prep) tap a teammate → 前往查看 (local replica, fast-forwarded). ‹ › only inside 联防/FA fields; FA: own pair only; eliminated: free.
 - Prep interactions: tap a unit → underframe with 出售 +1 (and 撤退 on board); items/Arts → 销毁; shop and upgrade are two-tap (select → confirm; the remake's 机变 cards too, §18.2); strategy draft forbids a strategy already taken by a teammate (队友已选).
 - Emotes: 6 themes × 6 battle emotes (research 09 §4.2), image-only bubbles beside the sender avatar, 1 s cooldown, 3 s bubble; `g.emote {id}` with the official emoji id.
@@ -350,6 +350,8 @@ Two public GitHub issues after 0.1.0, both client-only. Normative lines rewritte

 User: 「给每个卫戍干员都加语音」, then, after the first build, 「整备阶段不需要干员语音，战斗开始后才需要语音，也就是只需要作战中语音」.

+**Changed in 0.2.2** (§27, the owner's request of 2026-10-08 「添加一下干员点击上去的语气一样的语音」): 选中干员 is no longer battle-only — the detail panel opening on an operator the player taps says it in every phase, the prep, the shop and the hand included (`DetailPanel voice` is true on the game screen), with the official FOCUS_CHAR timing: no cooldown and no global gap on an idle channel, a newer tap replaces the 选中 line on air, a higher-priority line is never interrupted, and a tap starts no gap of its own for the battle lines (`VOICE_TAP_SLOTS`); every shop / reward card opened counts as a tap, so two cards of one operator each speak (`selectVoiceKey`). Every other line stays battle-only. 0.2.2 also adds the Japanese dub (「全套的日配语音」) as a second tree beside the Chinese one, `audio.voiceJp` — the same slots and file names from the dump's `voice/` folder — which the client plays when the player picks 日本語 under 设置 → 语音语言 (default 中文, not tied to the interface language), falling back to the Chinese line it lacks. The paragraphs below describe 0.1.1.
+
 **Where the lines come from.** Every operator the mode can field already has official battle lines in the same voice dump the BGM and SFX come from — `voice_cn/{charId}/cn_nn.mp3` (JP/EN/KR: `voice/`, `voice_en/`, `voice_kr/`, the same file names, `tools/fetch-assets.mjs --voice-lang=jp|en|kr`). `charword_table.json`'s `placeType` says when each line is used, and `indexVoice()` turns it into the manifest's `audio.voice[charId]`: 行动出发 `start` (BATTLE_START), 行动开始 `faceEnemy` (BATTLE_FACE_ENEMY), 选中干员 `select`, 部署 `place`, 作战中1-4 `skill1..skill4` (BATTLE_SKILL_N), 编入队伍 `squad`, 任命队长 `squadFirst`, 结算 `resultFour/Three/Two/Lose` (完成高难行动 / 3星结束行动 / 非3星结束行动 / 行动失败) and 干员报到 `gacha`. A slot with several lines stays an array and the client draws one. Only the base word key is planned (`wordKey === charId`): the dump has no folder for a skin variant's word key (`char_x_ita`, `char_x_epoque#28`), so those lines 404. **Only the slots a battle can play are planned** (`audio.mjs VOICE_BATTLE_SLOTS`): 干员报到 / 编入队伍 / 任命队长 are lines of the official client's own 养成 / 编队 UI, nothing in this mode ever requests them (`test/docs-consistency.test.js` checks the client's `voice(...)` call sites), and downloading them cost 360 files (19.3 MB) on every run — `--voice-all` adds them back for the complete official set. Measured: 120 of the 121 in-mode charIds have a full set — the 17 预备干员 and the mode's own 盟约·辅助干员 have no battle voice at all, so they simply get no entry — 1680 lines of the 2040 upstream, in a manifest of 5694 files / 324.2 MB (files and `stats.voiceChars` in `data/assets.json`).

 **Where each line plays — battle only.** Every line comes from a running battle's own event stream or from its settlement, so no hook needs a phase check and the 休整期 stays silent: the shop does not say 干员报到 (a merge-completing buy neither says 任命队长), a retreat does not say 编入队伍 and dragging a piece onto the board in prep says nothing. In battle: 开战 — the first operator deployed says 行动出发 and the others 部署 (the sim's `deploy` events; `setFieldUnits` resets the flag per field — in this mode every piece enters at once, which is why the gate below exists, and a knocked-out operator's redeploy says 部署 again); 接敌 — the sim's new `['engage', id]` (a unit's first attack that hits an enemy, official ENCOUNTER_ENEMY); 技能 — `skillN` from the unit's `skillIndex` (作战中1-4); 选中 — the detail panel opening on an operator **while a battle runs** (`DetailPanel voice` = the game screen's `combat`); 结算 — `resultVoiceSlot()`: 完美作战 ⇒ 3星结束行动 (绝境 / 终极 ⇒ 完成高难行动), 有漏怪 ⇒ 非3星结束行动, 一个没杀 ⇒ 行动失败, spoken by a drawn operator **of that battle's own field** (`resultSpeaker`, from the result's `perPlayer.unitsEnd`: the operators of that field, survivors first, a summon piece never), driven by the runner's new `emit('result', …)` — never the field on screen, so watching a teammate cannot put their operator's voice on the viewer's settlement line (review on #73). `unitsEnd` names each unit by its chess id (`chess_char_*`), so `onResult` passes the chess record's `charId` as `resultSpeaker`'s `charOf`; as merged, the speaker looked for `char_*` ids only and no battle ever said its settlement line (found while porting #73 to 0.2.0; `test/ui/audio.test.js` runs a real battle result through it).
@@ -103,6 +103,8 @@ Eight workstreams were merged into `playtest6` (WH → WG → WD → WC → WB
 - **#19**: PRTS 帮助 "临时刷新3名高于调度中心等级1阶（最多6阶）的干员"; the official 玩法介绍 image shows three different operators at price 0. v2.4.1 rolled three times independently (14–21 % of offers had a duplicate). **Now** three different chess, copy-weighted; a tier short of different chess tops up from the tier below [ASSUMED]; the offer reserves no copies. The integration QA found the same repeat in the other free pick-one chess offers — 信标 (same tier), 寻呼模块 (same bond ≤ shop level, content/items/meta.js) and the engine default `builtinMeta use_equip_reward_special_goods_char_chess` — they now draw different operators too (信标 tops up from the tier below, 寻呼模块 shows fewer cards when the pool has no other one), and `pushRewardOffer` drops a repeated id from any list it is given (`test/content/items.test.js`).
 - Tests: `test/match/playtest6-matchflow.test.js` (the draft, two battles for a multi-round card end to end, 联防 counts, rewards), `test/match/followups.test.js`, `test/ui/playtest6-unite.test.js`, `test/ui/playtest6-bounty.test.js`, `test/content/items.test.js` (教鞭).

+Superseded in 0.2.2: 教鞭 no longer takes one of its three 战术特训 cards at random — its owner chooses (docs/META.md §2.5).
+
 ### 20.7 机变 card text; the shop's confirming tap on phones (#6, #10) — `ui/choiceOverlay.js`, `css/screens/game-panels.css`, `ui/shopBar.js`, `css/screens/game-shop.css`, `css/devices.css`

 - **#6 seen** (screenshot 2, Android 756×366 CSS px, 终极 co-op): the 道具补给 cards showed only icon, name and tier. **Cause**: the card was a centred column sized for 16:9; with the root font clamped at 40 px a 3×2 card was 105 px tall and showed 0.2 lines of text (217 of 223 texts cut), in all four families; at 1920×1080 nine long texts were cut at 4 lines. **Official** (research 06 §11.5; screenshots Mb2mtd9 / Wrkhtxv / SEbySVl): a framed icon at the top-left with the bold name beside it, the description below, numbers in blue, the tier chip on the icon's corner, the taker's round avatar alone at the top-right. **Now**: that layout; the text from `cardText` (the card's own `descRaw`, else the data `descRaw` while its plain form equals the server's `desc`, else the server text), clamped to 7 lines (8 at line height 1.3 on phones under 700 px wide: with Noto Sans SC — the page's web font, metrics of Android's CJK font — 海沟实验体 wraps to 8 lines at 640×360, residual QA); short screens get a tighter overlay header (no CONTINGENCY micro line) and .19rem text; the taker badge is the avatar alone (name in its tooltip and the card's aria-label) and the taken card's head reserves its width. All 223 texts (51 items, 129 bounty, 43 tactic) fit in full, taken or not, from 640×360 to 2560×1080, measured with Noto Sans SC.
@@ -19,6 +19,8 @@ Tags: **[DATA]** client game data · **[VERIFIED]** official or wiki rule text,
 | `07-assets.md` / `07-assets.json` | Verified URL patterns and resolved URLs for avatars, portraits, skills, bond/item/band icons, UI, Spine (ops + enemies), BGM/SFX, fonts; build plan. Addendum: spot check 40/40 OK. | `operators[*]`, `enemies[*]`, `bonds`, `items`, `bands`, `autochessUi`, `audio` |
 | `11-limits-official.md` | Official limits read from the client (2.7.71 il2cpp) and the community: bond layers cap at **999** per bond (`MAX_GARRISON_STACK`, `AddBondCount` = min(L + n, 999)); in boss battles outside training a single hit of **≥ 300000** on a leader (`IsBossEnemy` = an `autoChessData.bossInfoDict` enemyId) is **cancelled** ("限伤", `MAX_BATTLE_DAMAGE`); implemented in DESIGN §20.12. | `shared/constants.js BOND_LAYER_CAP / BOSS_HIT_LIMIT` |
 | `12-flying-visuals-official.md` | Official flying visuals and model placement, read from the client (PR #211 by @xcdoge; ported into 0.2.0 by the owner's decision of 2026-10-06): the fly offset is the single model-independent constant **0.35** (`CharacterAnimator..ctor` stores `Vector3(0,0.35,0)` at field +0x114; `_SetFlyMountPointOffset` / `_SetFlyHitOffset` add it while flying, negate it on landing) — in the client's *character* space, whose unit is the standard prefab scale 0.27, so **0.35 / 0.27 ≈ 1.3 tiles**, confirmed by pixel-measuring an official screenshot (1.2–1.4). `FLY_HOVER` 0.32 → **1.3**, flat (no per-model term). The same sweep found two model quirks: `modelScaleY` 1.263 (帝国炮火先兆者 pair, vertically stretched) and `mirrorX` (木制瑞印, mirrored prefab). | `public/js/render/units.js FLY_HOVER / enemyModelScaleY`, `data/enemies.json`, `tools/local-extract/enemy_model_offsets.py` |
+| `13-knockback-official.md` | Push / pull (推拉) in the official client (PR #380 by @xcdoge; written in Chinese): the knockback abilities (`Knockback`, `KnockBackWithDirection`, `KnockBackWithCharacterDirection`, `DragTowardSource`) carry an impulse (`m_force`) and a velocity decaying under friction (`m_friction*`, the tile's `additionalFriction` — `GrasslandType` NORMAL_LAND / ICE_LAND / SLIME_LAND); the data has no `move_duration` / `move_distance` key (distances are emergent; 力度 comes from `skill_table` / `enemy_database`); every tile of this mode's 11 maps is normal land. Also noted, not modelled: wall bounces (`_bounceTimesAsDamageTimes`) and `unbalanceProtectDueTime`. The client's slide (`v0 = 2D/T`, `a = v0/T`, T = 0.14·√D in 0.12–0.45 s) is ours [ASSUMED: no source gives the official timing]. | `render/units.js DISPLACE_SLIDE / slideTo`, `render/app.js` `displace` fx |
+| `14-death-animations.md` | Enemy death animations (PR #380 by @xcdoge): the art never drops the body (the PR author's 234-clip sweep — no Die/Idle/Move/Attack clip translates the root), the 1.6 s death cap cut 25 clips short (盐风主教昆图斯 7.97 s) so it is 8 s, and a flying unit keeps its lift for the whole Die clip, dropping only inside the fade tail (the client removes its fly offset in `CharacterAnimator.OnFinish`). | `render/units.js DIE_CLIP_MAX / DIE_FADE_TAIL` |

 ## 2. Match structure

@@ -922,7 +922,7 @@ Members (5 chess, 4 in current shop pool; by tier in shop: {'2': 1, '3': 1, '5':

 **Implementable spec**

- **[2 distinct]** If a 突袭 member has not attacked for 10 s, or its skill is ready, and no enemy is in its range: it is immediately redeployed (keeping SP) onto a tile next to a ground enemy; while in that redeployed state its ATK and maxHP x(1 + 0.25 + 0.01*L). A passive skill that is on counts as ready (GitHub #49: the reporter's footage shows 缄默德克萨斯 jumping within her passive's 10 s; the engine keeps a passive on for the whole deployment [ASSUMED]), and so does a deploy-timed skill while its window runs (#109 made 宴 S2, 斯卡蒂 S2, 伊内丝 S3, 缄默德克萨斯 S1–S3 and 耀骑士临光 S2 duration skills that start at each deployment; once the window has ended only the idle trigger is left); the landing tile must have the target in range (GitHub #51 [ASSUMED]).
+- **[2 distinct]** If a 突袭 member has not attacked for 10 s, or its skill is ready, and no enemy is in its range: it is immediately redeployed (keeping SP) onto a tile next to a ground enemy; while in that redeployed state its ATK and maxHP x(1 + 0.25 + 0.01*L). A passive skill that is on counts as ready (GitHub #49: the reporter's footage shows 缄默德克萨斯 jumping within her passive's 10 s; the engine keeps a passive on for the whole deployment [ASSUMED]), and so does a deploy-timed skill while its window runs (#109 made 宴 S2, 斯卡蒂 S2, 伊内丝 S3, 缄默德克萨斯 S1–S3 and 耀骑士临光 S2 duration skills that start at each deployment; once the window has ended only the idle trigger is left); the landing tile must have the target in range (GitHub #51 [ASSUMED]). A 隐匿 (unrevealed) ground enemy is a target too, and one inside the member's range counts as an enemy in range (no jump) — the owner's decision of 2026-10-08 from the official game (GitHub #316); asleep, untargetable and flying enemies are not.
 - **[L >= 50]** ALL operators ASPD +50.
 - Formulas: `atkHpMultiplier = 1.25 + 0.01*L`; `idleSec = 10`; `aspdAll(L>=50) = 50`
 - How layers are gained: 史尔特尔 部署时 +8 (<=50/battle); 休谟斯 each 2 kills +1; 瑕光 +4/deploy; 斯卡蒂; 伊内丝 +5; 机变 "斯卡蒂的盟誓" +8, "德克萨斯的盟誓" +10.
@@ -18,7 +18,7 @@ Rich-text tags (`<@ba.vup>…</>`, `<$ba.inspire>…</>`) are stripped in this d
 | 6 | E2 Lv1 / S-Lv4 / mod 0 | E2 Lv60 / S-Lv7 / mod 3 | 4 (golden entry also 4) | 1 (golden 1) | char_elite_base_2 / char_elite_gold_2 |

 - Every chess has `upgradeNum = 3`: 3 copies of the same normal chess -> 1 golden (精锐) chess (`upgradeChessId` = `goldenChessId`). Golden chess have `upgradeChessId = null` (no further upgrade). Exception via 特质: 风丸 `chess_char_2_11` (garrison_93: "只需2个相同的本干员即可合成精锐") needs only 2 copies.
- favorPoint = 0 in every status -> **trust bonus 0**. The status has no potential field (the tables here are at potential 0); the remake fights at **full potential** (潜能 6 — the owner's decision of 2026-10-07, GitHub #252: a tournament video shows 刺玫 at ATK 435 / cost 15 = her E1 Lv55 413 / 17 + 攻击力+22 and two 部署费用-1; data/chess.json, build-data `OPERATOR_POTENTIAL`). Module (模组) is **inactive on normal chess** (equipLevel 0) and **active on golden chess** at stage 1 (tiers 1–5) or stage 3 (tier 6). The module used is always `defaultUniEquipId` (always the operator's `uniequip_002_*` = first advanced module, X or Y type depending on operator; 凛御银灰 `char_1045_svash2` has none).
+- favorPoint = 0 in every status -> **trust bonus 0**. The status has no potential field (the tables here are at potential 0); the remake fights at the **player's potential** (潜能 1–6 per operator in 干员调配, default full potential 潜能 6 — the owner's decisions of 2026-10-07, GitHub #252: a tournament video shows 刺玫 at ATK 435 / cost 15 = her E1 Lv55 413 / 17 + 攻击力+22 and two 部署费用-1, and of 2026-10-08: the 调度手册 「卫戍协议中干员潜能由自身已持有干员潜能决定」; data/chess.json is built at full potential and carries the lower ranks, DATA.md §2.3). Module (模组) is **inactive on normal chess** (equipLevel 0) and **active on golden chess** at stage 1 (tiers 1–5) or stage 3 (tier 6). The module used is always `defaultUniEquipId` (always the operator's `uniequip_002_*` = first advanced module, X or Y type depending on operator; 凛御银灰 `char_1045_svash2` has none).
 - Golden stat jump is large: tier 1 goes E1 Lv55 -> E2 Lv50 (so skill 3 exists but chess keep their `defaultSkillIndex`; the skill index does NOT change on upgrade, only the level 4 -> 7).
 - Prices: see 1.1 table; purchasePrice is per tier (T1=2, T2=3, T3=3, T4=3, T5=4, T6=4 coins); selling any chess returns 1 (data `chessSoldPrice`=1 for all; the shop/economy doc should confirm whether golden sells for more via other rules). Note: garrison `SERVER_PRICE` effects override the purchase price for specific chess (e.g. "购买价格为1").
 - Board/bench limits (constData): maxBattleChessCnt = **8** on field, maxDeckChessCnt = **10** on bench (整备区), storeCntMax = 6, shopRefreshPrice = 1, fallbackBondId = `emptyShip` (协防干员), borrowCount = 20 (max friend-support borrows).
@@ -35,7 +35,7 @@ Rich-text tags (`<@ba.vup>…</>`, `<$ba.inspire>…</>`) are stripped in this d

 - `stat(level) = kf0 + (kf1 - kf0) * (level - kf0.level) / (kf1.level - kf0.level)` using `character_table[charId].phases[evolvePhase].attributesKeyFrames` (two keyframes: Lv1 and max level). maxHp/atk/def/cost/blockCnt/respawnTime rounded half-up to int; others kept as float.
 - Golden: + module `battle_equip_table[defaultUniEquipId].phases[equipLevel-1].attributeBlackboard` (max_hp, atk, def, attack_speed, magic_resistance, …). Module trait additions are listed as `module.traitEffect` / `module.hiddenTalentEffects` in JSON.
- No trust. No potential in these tables — data/chess.json adds every `potentialRanks` attribute step and takes the talent candidates up to `requiredPotentialRank` 5 (full potential, the owner's decision of 2026-10-07). Talent-driven permanent stat increases (e.g. "攻击力+8%") are NOT folded into the numbers; apply them in the sim from `talents[]`.
+- No trust. No potential in these tables — data/chess.json adds every `potentialRanks` attribute step and takes the talent candidates up to `requiredPotentialRank` 5 (full potential, the default) and carries what each lower potential changes (`potDown`, chained talents — DATA.md §2.3: the player's 干员调配 setting since 0.2.2). Talent-driven permanent stat increases (e.g. "攻击力+8%") are NOT folded into the numbers; apply them in the sim from `talents[]`.
 - `cost` (DP) and `respawnTime` are used by automatic redeployment in battle (see 1.4).
 - Attack interval = `baseAttackTime * 100 / attackSpeed` seconds.

@@ -352,7 +352,7 @@ Combo check: `equip_chess_id` in blackboard is the partner's id WITHOUT the `_a/

 - Buff keys: 画卷 `trap_copy_front_char{}`; 教鞭 `trap_create_self_choice{choice_event=hunter_band_1}`; 神秘顾客 additionally `trap_disney_special{count=1}`.
 - `hunter_band_1` (choiceType PERSONAL_CHOOSE, "悬赏决策") pool is server-side. [ASSUMED] offer 3 random `enemyeffect_b_*` bounties (e.g. 碎骨·悬赏 +1, 萨卡兹百夫长·悬赏 +2, 泥岩·悬赏 +4, 锏·悬赏 +5, “复仇者”·悬赏 +6 funds; buff `add_enemy_kill_gain_coin{count,enemy_id,coin,round=1}`): adds 1 enemy to your next battle, killer gets `coin` funds.
- 画卷 copy: duplicates the operator (same chess id incl. elite `_b`) AND clones its equipped items into the copy [DATA desc "包括精锐状态和装备"]; copy goes to hand [ASSUMED]; if hand full -> temporary hand.
+- 画卷 copy: duplicates the operator (same chess id incl. elite `_b`) AND its equipped items [DATA desc "包括精锐状态和装备"]; the copied items arrive unequipped (PRTS 画卷 备注 "使用后销毁，获得的装备为未装备状态"), gained like any item — the hand, overflow temp (destroyed with the usual toast when both are full); copy goes to hand [ASSUMED]; if hand full -> temporary hand.

 ## 8. Item sources other than the shop [DATA]

@@ -352,8 +352,9 @@ reported the 战场#01 lower-gate enemies walking up the col-9 floor lane, where
 `server/sim/grid.js` keeps every official route LENGTH and the official route itself unless the 2026-09-29
 road-over-floor preference route (equal-length ties to the chain with the fewest non-blockable floor / gate tiles,
 smoothing that never cuts across floor its grid route does not walk, the corner tiles of a diagonal step included)
-crosses strictly fewer non-blockable tiles; in that comparison a segment that only touches a floor tile's corner does
-not cross it. Used alone (0.1.0) the preference also bent official diagonals into L shapes — the D5 report after
+crosses strictly fewer non-blockable tiles and no more 深水区 (GitHub #375, 0.2.2: the 深水区 is non-blockable too, and
+on 战场#08(下半) the patrolling leaders had waded through two water tiles to dodge one floor tile); in that comparison a
+segment that only touches a floor tile's corner does not cross it. Used alone (0.1.0) the preference also bent official diagonals into L shapes — the D5 report after
 0.1.0: on 战场#04 the lower-gate enemies walked (9,10) → (9,8) → (10,8) instead of the official diagonal (9,10) →
 (10,7), which only brushes the corner of the floor (10,9). Lanes that still differ from the table above (21 of the 154
 stage × gate × field routes, test/sim/pathing-official.test.js):
@@ -0,0 +1,138 @@
+# 13 · 推拉（击退 / 拖拽）的官方实现：冲量 + 摩擦
+
+这份文档回答两个问题：**官方推拉到底怎么动**，以及**摩擦是谁的参数**。结论全部来自官方客户端本体的
+`global-metadata.dat`（il2cpp 元数据）、官方 gamedata（社区镜像 Kengxxiao/ArknightsGameData）与官方 AB 资源包，
+并附上每一步的验证方式，便于复核。
+
+## 0. 结论速览
+
+| 问题 | 答案 |
+|---|---|
+| 官方怎么动 | **冲量 + 摩擦**：给被推者一个力（`m_force` / `m_forceVectorX/Y`），速度**逐帧被摩擦衰减**直到停下 |
+| 有固定时长/距离吗 | **没有**：数据里不存在 `move_duration` / `move_distance` / `move_tiles` 键（`moveDuration`、`moveTiles` 属于 UI 类）⇒ 位移量是**涌现**的，这也是社区只能给出"**推力-位移近似对应表**"的原因 |
+| "力度"从哪来 | **官方数据**：`skill_table.json` 的 blackboard `force`（41 个技能带它），敌人侧在 `enemy_database.json`（8 条） |
+| 摩擦是谁的参数 | **地板（地形）+ 单位质量**，**不是**技能参数：`GrasslandType { NORMAL_LAND, ICE_LAND, SLIME_LAND }`，每格是一条 `GrasslandData { modeIdx, grasslandType, additionalFriction }`，运行时还会被 `UpdateFrictionFactor` / `RestoreFrictionFactor` 临时改写 |
+| 卫戍协议里有几种地板 | **实际只有普通地面**：11 张地图共 18 种格子字符，无冰面/黏液 ⇒ 本模式下摩擦恒定，滑行时长**不必按地板分档** |
+| 飞行目标怎么算 | **没有独立的空中摩擦**（`boatAirFactor` / `m_airBase` 属于"船"玩法）：飞行用的是**同一套**公式（单位摩擦 × 所在格子的 `additionalFriction`）；差别只在**运动模式**（`ENEMY_MOTION_TYPE_FLY`、`HEIGHT_CHANGED_TO_FLY_BLOCK_MOTION_MODE` ⇒ 忽略地面阻挡），以及**失衡免疫 / 静态刚体**的空中单位根本不可位移 |
+| 曲线形状 | 恒定减速 ⇒ `d/D = 2k − k²`（ease-out quad） |
+| 推多久 / 拉多久 | **失衡（UNBALANCE）状态机**的时长（0.2.2，PR #392 的思路、按来源重做）：推 = PRTS 游戏数据基础「推力-位移近似对应表」的**位移时间**（受力等级 −2/−1/0/1/2/≥3 → 0.2 / 0.4 / 0.8 / 0.9 / 32⁄30 / 35⁄30 游戏秒，正好是 30 fps 的整帧）；拉 = 推与拉的**拉力作用时间**（1 s，受力等级 < −1 时 0.5 s，急停后仍保持到结束）；静态刚体受力 > 0 时 0.1 s 保底。客户端滑行用同一段游戏时间（默认 2 倍速 ⇒ 现实时间减半）；`T = 0.14·√D` 只留作旧录像（fx 不带 `dur`）的回退 [ASSUMED] |
+
+## 1. 官方类与字段（`global-metadata.dat`）
+
+推拉能力的类簇（与 PRTS 的"推与拉"术语逐字对应）：
+
+```
+Knockback · KnockBackWithDirection · KnockBackWithCharacterDirection · KnockBackTargetsByRootTile · DragTowardSource
+_decreaseForceLevelWhenNotInDirection（PRTS"特殊修正"−2 力度）· _dontChangeFaceByDirection · _GetPushDir
+TOO_CLOSE_TOLERANCE_SQR（PRTS"< 0.25 格转径向"）
+```
+
+运动学字段（同一批类里）：
+
+```
+m_force · m_forceVectorX/Y · forceScale · m_mass · m_velocity · get_mass · get_rigidbody2D
+m_friction · m_frictionBase · m_frictionFactor · get_frictionFactor · m_frictionFactorAdditional · get_additionalFriction
+UpdateFrictionFactor · RestoreFrictionFactor · _UpdateFrictionFactorAdditional · RestoreCached…
+_ApplyForce · _UpdateVelocityAndPos · _UpdateAllUnitDeltaPos · _GetForceByDirectionAndDamage
+m_bounceFrictionFactor · bounce · bounceease · _bounceTimesAsDamageTimes · _bounceBBKey · _bounceConditionValidator
+unbalanceProtectDueTime
+```
+
+⇒ `m_force*`（冲量）+ `m_velocity` / `_UpdateVelocityAndPos`（逐帧速度积分）+ `m_friction*`（衰减）就是"冲量 + 摩擦"；
+`UpdateFrictionFactor` / `RestoreFrictionFactor`（改完还要还原）说明摩擦**由外部临时提供**，不是能力自带。
+
+## 2. 地板（地形）摩擦
+
+地板系统就在**卫戍协议自己的格子类**里（同簇可见 `TileState { SHARED, MY_SIDE, MATE_SIDE }`、
+`CooperateStartTile` / `CooperateEndTile` / `CoopFootballTile`、`get_modeIndex` / `get_tilePlayerSide` /
+`_OnPlayerDying` / `_OnPlayerRevive`）：
+
+```
+_grasslandDatas · GrasslandType { NORMAL_LAND, ICE_LAND, SLIME_LAND } · GrasslandData { modeIdx, grasslandType, additionalFriction }
+DynamicBuffQuickSandTile（流沙）· DynamicBuffTile · DirectionTile（带方向的格子）
+```
+
+⇒ 每格可带 `additionalFriction`，即"**地板提供摩擦**"。另有冰面相关因子（`iceFactor`、`iceTile`、`OnIce`、`Frozen`）
+与"船/水/空"那套（`m_boatFrictionFactor`、`m_boatAirFactor`、`m_airBase`、`m_waterFlowForce*` —— 属于多索雷斯船玩法，
+**不是**飞行单位）。
+
+### 卫戍协议实际用到哪些地板
+
+`docs/research/05-maps.json`（11 张地图）里出现过的全部格子字符：
+
+```
+# ×1776（墙/虚空）  r ×580（道路）  X ×477（分区隔断）  f ×448（地面）  a ×330  b ×250  A ×165
+p ×77（预览栏）  d ×76（装置）  S ×66（起点）  I / E ×33（路线起/终）  h ×24（手牌栏）  O ×22  m ×12  i ×8  g ×8  R ×4
+```
+
+**没有冰面或黏液** ⇒ 本模式下所有可站地板摩擦一致。
+
+## 3. 力度来自官方数据（不是我们拟定的）
+
+`excel/skill_table.json` 的 blackboard 里 **`force` 出现 357 次、41 个技能**，例如：
+
+| 技能 | 名称 | 力度（按等级） |
+|---|---|---|
+| `skchr_weedy_1` / `_3` | 温蒂 炮管敲击 / 液氮大炮 | 0,1,2 / 1,2,3 |
+| `skchr_forcer_1` / `_2` | 见行者 护身射击 / 惊爆射击 | 0,1,2 / 1,2,3 |
+| `skchr_panda_1` / `_2` | 食铁兽 铁意六合 / 崩拳式 | 0,1,2 / 1,2,3 |
+| `skchr_rope_1` / `_2` | 崖心 勾爪发射 / 复式勾爪 | 0,1,2 / 0,1,2 |
+| `skchr_glady_1` / `_3` | 歌蕾蒂娅 缺水的大洋裂断 / 缺水的碎漩狂舞 | 0,1,2 / 0,1,2 |
+| `skchr_sqrrel_1` / `_2` | 阿消 水蒸气泵 / 高压水炮 | 0,1,2 / 0,1,2 |
+| `skchr_moeshd_2` | 可颂 磁爆锤 | 1,2,3 |
+| `sktok_archook` | 钩索 token 发射 | **11** |
+
+同一张表里 **没有** `friction` / `mass` / `knockback` 键 ⇒ **力度是数据（可查表），摩擦与质量在引擎里**。
+
+### 靠"职业"就能推拉的干员
+
+`character_table.json` 的 `subProfessionId` 有两个专门职业，**普通攻击**即推/拉，力不在技能 blackboard 里：
+
+- **`pusher`（推击手）**：阿消 · 见行者 · 食铁兽 · 温蒂
+- **`hookmaster`（钩索师）**：暗索 · 歌蕾蒂娅 · 杏仁 · 雪雉 · 崖心
+
+其余靠技能/天赋推拉的还有：Misery、灰烬、黑键、燧石、溯光星源、可颂、焰狐龙梓兰、傀影、莱伊、罗宾、圣聆初雪、
+琳琅诗怀雅、乌尔比安，以及 `勾爪` / `暴风雪` / `喷拒器` / `猎潮的骑士` / `流形` / `weedy_token` 等 token 与装置。
+
+## 4. 顺带查清的两条官方机制（尚未建模）
+
+- **撞墙弹跳**：`bounce` / `bounceease` / `m_bounceFrictionFactor` / `_bounceBBKey`，并且 **`_bounceTimesAsDamageTimes`
+  —— 弹跳次数计入伤害次数**（温蒂 / 推击手那类撞墙伤害的官方口径）。
+- **失衡保护**：`unbalanceProtectDueTime` —— 连续推拉之间有保护窗口。
+
+## 5. 客户端的做法（本次实现）
+
+`render/units.js`：模拟位移是**瞬时**的（`sim/battle/displacement.js` 在一次调用里走完 0.1 格步进），随后敌人处于
+**失衡状态**（0.2.2：不走路、不发动普通攻击，时长见 §0「推多久 / 拉多久」），快照只带终点，
+所以客户端补上中间帧——`UnitView.slideTo(x, y, { dur, friction })` 建立一次滑行，`update()` **逐帧做速度积分**
+（`v ← max(0, v − a·dt)`、`d ← d + v·dt`），终点由模拟给出（**权威**），`v0 = 2D/T`、`a = v0/T` 保证
+`T` 秒内正好走完 `D` 且速度归零 —— 即官方"冲量 + 摩擦"的等效形式；`T` = `displace` fx 带来的失衡时长 `dur`（游戏秒）÷
+播放倍率（`ctx.animRate`，默认 2 ⇒ 受力等级 0 的推 0.8 游戏秒 = 0.4 现实秒），滑行结束时模拟的失衡也正好结束；fx 不带
+`dur`（0.2.2 以前的录像）时才回退到 `T = 0.14·√D`（夹 0.12–0.45 s）[ASSUMED]。`friction` 是**地板因子**（本模式恒为 1；将来有冰面就乘
+< 1 ⇒ 滑得更久）。PR #392 的 `0.6387·√格` / `0.3194` 是用 PRTS 页面的「μ = 0.5 单位假设」推出来的，页面没有写出、也对不上
+位移时间表（每档高约一帧），未采用。`render/app.js` 在插值器开始显示「以带终点的快照结尾」的那一段时就用 `displace` fx 启动滑行：
+到达那张快照之前画面停在原处，之后从原处减速滑入模拟的实时位置（终点，以及敌人随后继续走的路），落地不跳。
+（PR #380 合入时模拟层、协议、golden 未改；0.2.2 的失衡状态改了模拟层，`displace` fx 多带一个 `dur`。）
+
+滑行途中死亡：滑行照常走完，尸体停在被推到的位置（模拟里它已经在那里；0.2.2 合入时改掉了原 PR 的「回到出发格」）；
+倒地的单位交给 `setDown` 放到它的格子上。
+
+## 6. 复核方式
+
+1. 元数据：`node` 读 `global-metadata.dat`，搜上列标识符（本文件所有类名/字段名都是这样得到的）。
+2. 力度表：`zh_CN/gamedata/excel/skill_table.json` → 各技能的 `levels[].blackboard` 里 `key === 'force'`；
+   敌人侧 `levels/enemydata/enemy_database.json`。
+3. 职业：`excel/character_table.json` 的 `subProfessionId ∈ {pusher, hookmaster}`。
+4. 地板：`docs/research/05-maps.json` 的格子字符统计（本文件 §2）。
+
+## 7. 附：官方 AB 资源包的读取要点（踩过的坑）
+
+供以后要读官方战斗配置时参考（本次为定位"摩擦是不是数据"而打通）：
+
+- `Flags` 里的 `0x200`（`UsesAssetBundleEncryption`）在本客户端上是**虚设**：blocksInfo 既没有 70 字节加密前导，
+  也不需要任何 16 字节 key（暴力扫 `global-metadata.dat` 144 万候选、`GameAssembly.dll` 5 万候选均无命中即因此）。
+- blocksInfo 位置 = `align16(headerEnd)`；**块数据是连续存储**的（不要逐块对齐）。
+- 压缩：`comp 3` = 普通 LZ4 ✓；`comp 4 / 5` = Arknights 自定义 LZ4AK（nibble 交换后按 LZ4 解，
+  终止条件是 `op === uncompressedSize`，**严格相等**）。
+- 包内官方配置是**带 `$type` 的 JSON**（如 `Torappu.Battle.Action.Nodes+SetBodyDirection`），
+  内嵌类型树 ⇒ 字段名可直接读；但**摩擦不在其中**（本文档 §2 已说明原因）。
@@ -0,0 +1,109 @@
+# 14 · Enemy death animations: the art never drops the body, and 25 clips were being cut short
+
+A player report — "a dying 帝国炮火先兆者 seems to drop at the very end", then "what about the other enemies'
+animations?" — answered by reading the client's own data and its code.
+
+## 0. TL;DR
+
+- **No enemy animation moves the whole body.** Playing all 251 enemies' clips and sampling every bone's world position,
+  the **root bone's vertical displacement is 0.000 tiles in every Die (234), Idle (235), Move (234) and Attack (215)
+  clip**. Limbs move (up to 2.16 tiles for “帝国的甲胄”'s arms); the body never does. Every vertical placement is the
+  renderer's: the fly lift, the model's own art hang (`.skel bounds.y`) and nothing else.
+- **A dying flyer dropped at the wrong moment.** The client removes its fly offset in `CharacterAnimator.OnFinish` —
+  when the finish state ends, not when the unit dies — so a drone stays aloft through the death animation and drops as
+  it goes. The remake dropped it the instant `alive` went false, so the death clip played on the ground. Fixed: the lift
+  is held for the Die clip and released inside the fade tail (§4).
+- **25 death clips were cut short by a 1.6 s cap** — 盐风主教昆图斯's 7.97 s clip played 20 % of itself, 鼠王 37 %,
+  杰斯顿·威廉姆斯 40 %, and 22 more (§3). The cap is now 8 s, above every clip in the data.
+- 17 manifest entries have no Die clip at all; they are the ground items/collectibles (§5).
+
+## 1. How the clips were read [DATA]
+
+A sweep script of the PR author (PR #380, not in this repository) reads `data/assets.json` + the extracted `.skel` files
+and **plays** each clip: `Skeleton` + `AnimationState`, stepped at 30 Hz, `state.apply(skeleton)` and
+`skeleton.updateWorldTransform()` after every step, recording the range of `skeleton.y` (the root) and of every
+`bone.worldY`. Ranges are converted to tiles with the model's `modelScale` (320 skeleton units = one tile at scale 1 —
+the conversion the flying-visuals write-up uses for the same models).
+
+Do **not** try to read the timelines directly: in this build of `@pixi-spine/runtime-3.8` they hold packed numeric
+arrays (`timeline.frames[0]` is a number, not `{time, x, y}`), so a keyframe walk silently reports zero motion — the
+first attempt here did, and 270 Attack clips "had no rotation". Sampling the played skeleton is layout-agnostic and
+matches what the client draws.
+
+## 2. Nothing translates the body [DATA]
+
+| clip | read | root displacement > 0.01 tiles |
+|---|---|---|
+| Die | 234 | **0** |
+| Idle / Default | 235 | **0** |
+| Move / Walk | 234 | **0** |
+| Attack | 215 | **0** |
+
+The largest single-bone travel inside a Die clip is 2.16 tiles (“帝国的甲胄”, a flailing limb), then 1.40 (节日气球),
+1.29 (“独轮车玩具”), 1.18 (业余竞演者) — normal animation, not a fall. A "the body sinks" effect can therefore only come
+from the code paths that place the model (`FLY_HOVER`, the `zTarget` ease, `lift`, and the artist's `bounds.y`).
+
+## 3. Death clip durations, and the 1.6 s cap that cut them [DATA]
+
+234 Death clips: shortest **0.50 s**, median **1.00 s**, longest **7.97 s**. `render/units.js` used
+`clamp(d / rate, 0.35, 1.6)`, so **25 clips were trimmed** (11 %), every one of them a boss or elite:
+
+| enemy | Die clip | played by the old cap |
+|---|---|---|
+| 盐风主教昆图斯 | 7.97 s | 1.60 s (20 %) |
+| 鼠王 | 4.33 s | 37 % |
+| 杰斯顿·威廉姆斯 | 4.00 s | 40 % |
+| W / 泥岩 | 3.33 s | 48 % |
+| 假想敌：胄 ×2 | 3.10 s | 52 % |
+| “巨大的丑东西” | 2.83 s | 56 % |
+| 腐败骑士 | 2.67 s | 60 % |
+| 高普尼克 ×2 / 卢西恩，“猩红血钻” / 假想敌：管 ×2 | 2.33 s | 69 % |
+| 假想敌：弦 | 2.30 s | 70 % |
+| 雪孩子 / 伊利昂的木驮兽 / 鸭爵 ×2 / 纠缠藤蔓 / 凋零骑士 | 2.00 s | 80 % |
+| 陷落雪祀 | 1.80 s | 89 % |
+| “帝国的甲胄” / 底海滑动者 / 富营养的滑动者 | 1.67 s | 96 % |
+
+The lower bound (0.35 s) trims nothing: the shortest real clip is 0.50 s. The client plays its clip out — its `OnFinish`
+fires when the finish state ends — so the cap only ever made the remake's deaths shorter than the official's.
+
+## 4. A dying flyer drops at the end of the clip, like the client [DATA]
+
+The client's fly offset is one constant (`Vector3(0, 0.35f, 0)` in `Torappu.Battle.CharacterAnimator`, applied by
+`_SetFlyMountPointOffset` / `_SetFlyHitOffset` while a unit flies and negated when it lands; the remake draws 0.35 in
+character space ≈ **1.3 tiles** as `FLY_HOVER` — the flying-visuals write-up has the full derivation). Its only direct
+caller is `OnFinish` (#62924, call site
+0x1805feb87), disassembled as `xor edx, edx; mov rcx, rbx; call _SetFlyMountPointOffset` — **arg 0**, the negation
+branch — followed by clearing `[this+0x128]` and a tail call to another cleanup. `OnFinish` has no direct `call` site of
+its own: it is a virtual lifecycle method (there is a `<>xLuaBaseProxy_OnFinish`), so the battle framework calls it
+through the vtable when the character's **finish** state ends.
+
+Since the Die clips carry no vertical motion (§2):
+
+- **Client**: the drone stays aloft for the whole death animation and drops as the finish state ends.
+- **Remake before**: `hoverTo = this.flying && this.alive ? FLY_HOVER : 0` — `die()` clears `alive` *before* the clip
+  plays, so every flyer sank to the ground inside ≈0.42 s (lerp τ = 1/6 s) and played its death clip **on the ground**.
+- **Remake now**: the hover target is `this.alive || (this.dying > DIE_FADE_TAIL && !this.down)` — the flyer holds
+  `FLY_HOVER` for the whole clip (the view fades over the last `DIE_FADE_TAIL` = 0.55 s of `dying`) and drops inside
+  that fade, which is the client's `OnFinish` moment. A knocked-down (`down`) flyer still lies on the ground.
+
+All 27 flying enemies of the data have a Die clip of 0.67–1.00 s, well inside the cap, so the cap change never touches
+them; 帝国炮火先兆者 and 帝国炮火中枢先兆者 are unexceptional (Die 1.00 s, root displacement 0.000).
+
+## 5. The 17 entries with no Die clip [DATA]
+
+未装配刀片 · 防护背心 · 冲击式施术单元 (deployable equipment) and 木制瑞印 · 红木瑞印 · 小说卷轴 · 诗画卷轴 · 青铜镜 ·
+黄铜镜 · 木制镇纸 · 红木镇纸 · 青瓷茶器 · 彩瓷茶器 · 铜矛头 · 铁矛头 · 铜灯盘 · 铁灯盘 (the collectible/relic props). They take
+the documented frozen-first-idle-frame path (see the `die()` comment in `render/spine.js`), which is right for an item
+that is picked up rather than killed.
+
+## 6. Implementation (2026-10-06, client-only)
+
+- `public/js/render/units.js`: new `DIE_FADE_TAIL = 0.55` (the fade tail `die()` already appended as a literal) and
+  `DIE_CLIP_MAX = 8`; `dying = clamp(d / rate, 0.35, DIE_CLIP_MAX) + DIE_FADE_TAIL`; the hover target as in §4; the three
+  `0.55` literals in `die()` / the knockdown branch / the fade now use `DIE_FADE_TAIL`.
+- `test/render/unitview.test.js`: a dying flyer keeps `FLY_HOVER` through the Die clip and is under a quarter of it
+  inside the fade (a walker stays at 0); a data guard — no enemy Die clip of `data/assets.json` exceeds `DIE_CLIP_MAX`
+  (234 clips, the longest 7.97 s: the §3 durations re-checked from the manifest when this was merged in 0.2.2) — and
+  the 7.97 s clip is scheduled whole, the view still on screen 5 s in.
+- The root-displacement sweep of §2 is the PR author's measurement (their script is not in the repository).
+- Client-only: the server serves `units.js` per request, so a hard refresh (Ctrl+F5) applies it — no restart.