- The settings (sp.pref.settings) hold a key map, action → KeyboardEvent.code, for refresh / freeze / level-up /
retreat / sell / ready (the ready key also pauses a solo battle); the defaults are the 0.1.4 keys R F D Q X Space,
so nothing changes for a player who never opens the section. sanitizeSettings keeps each saved key that may be a
shortcut and gives any other its default; not a map, or two actions on one key, gives the defaults.
- One lookup, actionForKey (the character typed first, then the physical key: the 0.1.4 code-or-key match, identical
for the defaults), serves shortcutFor(e, keys) in game.js and the facing wheel; the modifier, text-field, repeat and
overlay guards are unchanged. The shop bar, ready / pause and underframe key hints read the map.
- 设置 → 快捷键 (the one SettingsModal, in a match and on the title screen): each action with its key; click or Enter
waits, the next key binds; Esc cancels (the dialog stays open), Tab leaves; Ctrl / Alt / ⌘ combinations and the
interface keys (Esc, Tab, Enter, arrows, F1–F12) are refused; a key in use is swapped and the line under the list
names both; 恢复默认. Esc stays fixed. Touch-only devices: a keyboard note and a folded list; one column under 768 px.
- Tests: test/ui/feedback5-hotkeys.test.js (map, sanitising, conflicts, lookup, capture, handlers and hints with a
rebound key), test/ui/gameLogic.test.js (sanitizeSettings now returns keys), test/ui/hotkeys.e2e.test.js (SP_E2E=1:
撤退 rebound to W with the keyboard only in the mock match; phone width).
Golden unchanged (client UI only).
.gm__watching shows at the tap, but the dev mock answers the row's g.watch 40-100 ms later (game-mock.js mockRequest). The test started the battle about 50 ms after the tap. When the answer came after that, the mock took the scouting request as a battle watch, as a server-run server does, and moved m.field and the battle frames to the teammate's field. The strip follows the field on screen, so it rightly showed that teammate, who has no bonds in the mock, and `.bslot .bond` never came (3 of 10 runs at 1920x1080). Each 640x360 touch tap takes about 100 ms, so the answer always came first there. A player never loses their own strip this way: client-side combat (the default) refuses a teammate's field while the own battle runs. Under server-run combat the late tap shows that teammate's battle with their real bonds and the eye tag.
Now the test waits until m.field is the teammate's read-only board before the phase change. It also asserts that the own battle's expanded strip is the player's own (no data-owner). Every other check is unchanged. 10 of 10 runs pass. Golden unchanged.
- docs/history/0.2.0.md §25.14 (also scratchpad draft 0.2.0.aa.md) in the §25 style: the prisoners' models and
timing (§25.14.1), 忍冬 — her default S3 has cast on its 2-2 running range since §25.5's ACTIVE_RANGE (0.1.4 waited
for her 1-tile range; §25.14.2), a bounty burned to death (§25.14.3), 枯朽萃聚使徒's seeds — fixed by 0.1.3 §23.25,
every leaked seed paid the card again in 联防 on v0.1.2 (§25.14.4); the §25 intro list, the owner's decisions, the
§25.10 table, what stays [ASSUMED] (the 联防 payee) and the open polish (enemy prefabs whose default clips differ
from the manifest: 猎狗pro, 大鲍勃, 碎骨, “斩胄之剑” / “破胄之锤”).
- test/sim/feedback5-community-pins.test.js pins the two reports this tree already answers (they pass on the parent):
忍冬 S3 casts with an enemy two tiles ahead and not three, S2 on its 3-12, S1 on her attack; a bounty 使徒's seeds
carry no bounty and pay nothing, the body pays once, and in 联防 only its own leak carries the card.
- Golden unchanged.
- Reproduced (scratchpad scripts): on its own field a bounty enemy the 活性源石 alone burns to death pays its
owner — on this tree and on v0.1.4 (all 96 walking kill-bounty enemies, 战场#04, no operator: 0 unpaid). The
loss is in 联防: a leaked bounty enemy re-enters with `bounty.ownerPlayerId` = the leaker (unite.js
planUnite), who has no per-player entry in that battle. A death without an operator killer fell back to that
owner, and Battle.addCoins dropped the coins. Until 0.1.3 the 联防 field was the round's map, so 战场#04's
活性源石 burned leaked bounty enemies for nothing — the report; since then the 联防 maps have no terrain, but a
无来源 tick, the enemy's own drain or another enemy's hit still paid nobody.
- Official: PRTS 卫戍协议：盟约 决策 "该敌人于对应玩家所属区域倒下时，使相应玩家获得额外资金"; the bounty pays whenever the
body dies (the owner's decision, relayed 2026-10-06; 0.1.1 "paid once, at the final death"). No PRTS exception
for 环境伤害.
- Now: Battle._bountyPayee (battle/economy.js): an operator or summon of a player in the battle that dealt the
blow pays its player (as before, 联防 helpers included); any other death pays the card's owner when that player
fights there (own field, Final Assault / Hidden Core — unchanged), else the player whose half of the field the
enemy fell on (Battle._ownerForTile) [ASSUMED for 联防: no source names the payee of a death nobody caused there].
- Docs: SIM.md §1.2, META.md SETTLE, PLAYING.md (联防 rules), unite.js header.
- Tests: test/sim/feedback5-bounty-payee.test.js — the 联防 case (one helper; two helpers by half; killed by
nobody, 无来源 terrain damage, another enemy, its own HP loss) fails on the parent; pinned: the own-field
活性源石 death, a helper's operator on the other half, a pair boss field. Golden unchanged.
The 0.2.0 candidate's full pass failed at the second drop ('Waiting for selector `.eqr .eqr__opt` failed'). The view
keeps a dropped piece on the drop tile for 1.3 s (render/app/tune.js DROP_PENDING_MS, unchanged since the first import)
and then flies it back in a frame-timed tween. The test picked the item up again 0.5 s after 取消, where it had been
drawn a moment before. Even alone the first try often grabbed nothing (the helper's retry hid it). With few frames
(software-rendered headless Chrome on a loaded machine; reproduced with frames ~600 ms apart) the press grabbed the
operator under the item, the operator dropped on its own tile opened the direction wheel, and the wheel swallowed the
retry. The dialog itself opens and renders as before (补位 / i18n / 自选 not involved).
Now 取消 is also checked to draw the item back on its bench slot, and the next drop waits for that; under the same slow
frames the drop succeeds at the first try. Alone: 5/5 twice; mock.e2e 32/32, devices.e2e 26/26. A test timing flake,
no product change; golden unchanged.
- Cause, client: the prisoner skeletons carry three clip sets whose collar light is grey, blinking orange
and red. The official battle prefabs (local client battle/enm_pfb_6.ab: the Graphic component's anim key
-> clip table, the modes' replace pairs, the clip the Spine starts on) draw mode Default (【禁锢】) on the
grey set, mode R (the warning) on the orange set and mode L (【解放】) on the red one. The name-based role
resolver gave 普通囚犯 / 老练囚犯 their unnumbered set — the red one — so they came out of the gate freed;
no FORMS entry existed, so no prisoner ever changed (强壮囚犯 / 拳师囚犯 / 重犯 / 传奇重犯 stayed grey).
- Cause, sim: buff_template_data enemy_confinement[atk_cnt] spends a charge ON_BEFORE_ATTACK (not while
麻痹) and switches to mode R when one is left; enemy_liberty_listener[warning] switches to mode L before
the next attack (PRTS "进行第4次攻击前，切换至解放状态"). archetypes.js prisoner() counted on the 'attack'
hook, after the damage, so the 4th attack still hit confined, and it sent no form.
- Now: prisoner() counts in `before` (an attack 麻痹 interrupts never gets there, PRTS 拳师囚犯), setForm
'warning' (fx 'phase', drawn silently: kinds.js SILENT_PHASES) before attack times-1 and 'liberty' (fx
'liberate') before attack times; 重犯's free-all and 杰斯顿's KO switch the freed ones too.
render/units.js FORMS maps 'warning' / 'liberty' for the six skeletons (普通 / 老练囚犯 *2 / unnumbered,
强壮囚犯 *2 / *3, 拳师囚犯 / 重犯 / 传奇重犯 *_orange / *_red). tools/assets/spine.mjs PREFAB_SPINE_ROLES gives
普通囚犯 / 老练囚犯 their grey *3 set; data/assets.json regenerated (fetch-assets --offline --add-only, JSON
compare: only those two anims and the hash), data/enemies.json too (build-data --offline: only their
attackAnim.clip Attack -> Attack3, same length and strike frame). The 3D board draws the same Pixi views.
- Docs: ASSETS.md (PREFAB_SPINE_ROLES, the prisoner forms), SIM.md (the form fx), design/client.md (FORMS).
- Tests: test/content/feedback5-prisoners.test.js (the six: warning before attack 3, attack 4 hits with
ATK +50 %, the forms reach the client and UnitInfo; free-all; 麻痹), test/render/feedback5-prisoners.test.js
(manifest = the prefab's grey set, FORMS sets, view clips, PREFAB_SPINE_ROLES = the committed manifest) —
both fail on the parent but the 麻痹 pin; test/render/feedback5-prisoners.browser.test.js (RENDER_E2E=1:
three prisoners on the 2D and the 3D board, each model's set follows its sim form).
- Golden: 28 scenarios. Only the warning's new 'phase' fx (the prisoner fell before its 4th attack):
roster-001, roster-003, roster-007, roster-024, roster-033, bond-miraShip-low. The 4th attack now hits
freed, one attack earlier than before, so the fight after it moves: roster-004, roster-022, roster-026,
bond-kazimierzShip-low, bond-raidShip-low, bond-soloShip-low, bond-victoriaShip-low, bond-visiShip-low,
standin-01, diy-024, diy-061, diy-065, diy-066, diy-094, diy-097, diy-104, diy-105, diy-127; only the
per-second snapshots besides the fx: roster-021, diy-017; and the bot matches whose battles field
prisoners: solo-ABYSS-2, solo-HARD-9-boosted.
Ported by hand from PR #211 by @xcdoge:
- render/units.js: FLY_HOVER 0.32 → 1.3, flat (the official fly offset is one model-independent Vector3(0, 0.35, 0) in
the client's character space, whose unit is the standard battle-prefab scale 0.27; an official screenshot measures
1.2–1.4 tiles); enemyModelScaleY multiplies the skeleton's Y scale and the head (bar) height, also in the impostor
path; a mirrorX model flips on top of the direction flip. The body, its HP bar, damage numbers, projectile hits and
skill rings ride `hover`, the shadow stays on the ground; one camera drives the units over the 2D and the 3D board.
- tools/build-data.mjs MODEL_STRETCH_Y / MIRRORED_PREFABS → enemies.json modelScaleY / mirrorX; data/enemies.json
regenerated offline here (the PR's one-line blob was not taken) — JSON compare: only enemy_1112_emppnt /
enemy_1112_emppnt_2 modelScaleY 1.263 and enemy_1196_msfyin mirrorX true.
- tools/local-extract/enemy_model_offsets.py (optional, host-side prefab reader) and
docs/research/12-flying-visuals-official.md (without the claim that the owner said 「都要」 on 2026-10-05 — the owner's
decision of 2026-10-06 is cited; where the quirk enemies appear corrected against 0.2.0's data: 帝国炮火先兆者 is a FLY
special enemy of the mode, 帝国炮火中枢先兆者 comes with the bounty 战术特训·飞行III, 木制瑞印 is in no wave). No
screenshot or extracted art committed.
Tests (from PR #211): test/render/unitview.test.js (FLY_HOVER ≈ 0.35 / 0.27; the same lift for every flyer; a flying
view lifts by FLY_HOVER — added here: its body and HP bar ride the lift, its shadow stays on the ground),
test/render/playtest6-enemies.test.js (the data keys; the 1.263 aspect and bar height; the mirrored model's x sign);
both fail without the renderer change; the 425 render unit tests pass. Docs: DATA §9, ASSETS, design client.md,
research 00 / 12, DESIGN §25.13.5 and the §25.10 table.
Golden unchanged (drawing only; the sim reads neither field).
Ported by hand from PR #209 by @qingjingshenghuo (its one-line data/config.json was not taken):
- server/match/gamedata.js bossPoolShareOf (used by GameData.bossPoolShare and by finalAssault.bossPoolHp for a plain
game-data object): solo × `solo` (1); co-op with `perPlayer` true × `coop` (1) × the players alive at the boss
round's start (bots and AI 托管 seats count, eliminated / departed seats do not; at most aliveFull 4; a count left
out = a full team). The fixed pool stays restorable through config: bossHpScale perPlayer false + solo 0.25 (and its
optional aliveScaling).
- tools/build-data.mjs: the global bossHpScale carries the rule (perPlayer true, solo 1); the modes' entries keep only
bloodPointKey / unaffectedByEnemyScale. data/config.json regenerated offline — JSON compare: only bossHpScale (global
and the 9 mode entries) changed.
- The problem #209 documented: a 胄 drone death is 2 % of the pool and 限伤 cancels a hit of ceil ≥ 300000, so the
hidden 胄 终极 pool at 3 / 4 players (21.6M / 28.8M) would cancel every drone (432000 / 576000). The drone link is
a share of the pool, no hit: Battle.loseHp takes noHitLimit and the link passes the limit [ASSUMED: the official
client checks every damage modifier and PRTS calls it 真实伤害, but which max HP it reads is undocumented]; a part's
传递 and every hit stay under the limit.
Golden (regenerated; why: the pool size):
- fields: boss-boss_1..7-pair/-solo and hidden-boss_8..10-pair/-solo (all 20 boss / hidden scenarios) — a pair
field's pool × 2 (two players), a solo template's × 4 (solo × 1, was × 0.25); boss-boss_1-pair, boss-boss_1-solo,
boss-boss_3-pair, boss-boss_3-solo, boss-boss_5-solo, boss-boss_7-pair and hidden-boss_9-solo now end at the
200 s cap instead of clearing; the two 联防 scenarios are unchanged.
- matches: solo-FUNNY-2, solo-NORMAL-1, solo-HARD-1, solo-HARD-2 now lose the Final Assault (solo leaders have 4× the
HP: team LP 0 with 93088–176395 of the pool left); solo-FUNNY-1, solo-NORMAL-2, solo-HARD-9-boosted (its Hidden Core
too), coop2-NORMAL-3, coop4-FUNNY-6, coop2-ABYSS-10-boosted, coop2-NORMAL-70-diy still win with more leader damage
(pool × players alive; solo-FUNNY-1 / solo-HARD-9-boosted end with 6 / 1 team LP less). Unchanged:
coop2-NORMAL-8-serverrun and coop2-NORMAL-14-standins (one player left at R14: × 1 as before) and the matches that
end before R14; roster / bonds / standins / diy unchanged.
Tests: test/match/boss-hp-scaling.test.js (from PR #209, adapted — every leader and difficulty × 1–4 players and solo;
R14 / R15 leave out two departed players; AI 托管 and bots count, an eliminated player does not; config restores the
fixed pool, the opposite of the PR's test; the plain-object fallback agrees; only the leader's HP scales);
test/content/enemies_bosses.test.js (hidden 胄 终极 at 3 / 4 players: the drone share lands, a hit of the same size is
cancelled — fails without the change). Updated on purpose: finalAssault, balance, bosshp, connection, docs-consistency
(the §20.9 record points to §25.13.4; the data carries perPlayer true), sim/playtest6_limits (the drone share lands
past the line), match/playtest6-bosspool (the 5 % knob halved to keep the 2-player pool's size: that test is about the
pool's arithmetic). Docs: PLAYING §7, DATA, META, BALANCE, SIM, design engine.md / network.md, the §20.9 / §20.10
records, research 00 / 01 / 08, the code comments, DESIGN §25.13.4 and the §25.10 table.
garrison_123 「<刷新时>若为本回合首次主动刷新…此干员在整备区时也有效」 counts per piece (§21.1). No source says whether a
merge resets the count; since 0.1.1 _mergeChess carried the copies' highest per-piece round count onto the elite
([ASSUMED], conservative), so two white copies that had fired this round made an elite that did not fire again. The
owner decided on 2026-10-06 that a newly merged elite counts as a new 拉普兰德.
server/match/player/acquire.js _mergeChess no longer carries the per-piece round counters (拉普兰德's refresh count,
garrison:SERVER_GAIN_BOND_LAYER_BY_REFRESH_CNT:refreshes, is the only one): the elite starts at 0 like any new piece;
an in-place promotion keeps its count. test/match/feedback1-meta.test.js: the test that locked the old inherit is
updated on purpose (the elite adds +8 on its next refresh, once), and the case "two white copies refreshed this round,
merge, refresh → +8" is new; both fail without the change. Docs: META (counter API, edge cases), PLAYING, the comments
(PlayerState, pieces, effectsMeta, garrisons/meta), DESIGN §25.13.3 and the §25.10 table.
Golden unchanged (no corpus match merges 拉普兰德 after a refresh in the same round).
0.2.0's third tab 自选编队 (WK 0def7ba, merged in d6fd800e — the toolbar check passes on its parent) widened the centre
column of the symmetric grid, and the co-op briefing's countdown + 导出 / 导入 / 全部恢复默认 no longer fit in their half:
they ran 4 px into the tabs at 667×375 (loadout.e2e's toolbar check), 63 px at 844×390 (the countdown on 自选编队) and,
in English, at every size up to 1920×1080. Not the briefing's 本局禁用干员 lines: the overlap is the overlay's .lo-top.
- the grid centres the tabs while both sides fit in equal halves (minmax(max-content, 1fr) minmax(0, auto)
minmax(max-content, 1fr)) and shifts them otherwise; the tab strip scrolls only as a last resort, never overlaps;
- the countdown caption breaks at its spaces (Loadout / Deadline; 调配截止 stays one line);
- tighter tabs at ≤ 1200 px; at ≤ 1100 px the briefing's countdown takes the place of the 已调整 / 同步 chips;
- phones in landscape (≤ 1000 px, up to the 932 px ones): no chips, and 返回 / 全部恢复默认 (全部持有, 全部清空 — each
confirms first) show their icon, the text kept as their accessible name. Desktop zh looks as before.
Checked: zh and en × briefing and lobby × the three tabs × 18 sizes from 667×375 to 2560×1440 — no overflow, overlap
or offscreen button, the tabs never scroll. The e2e toolbar check now runs zh and en at 667×375 and 844×390 (and en at
1920×1080) and asserts the three tabs show in full; it fails on the old CSS. Golden unchanged.
Its fake predicate turned true 95 ms after the page was made and the test wanted at least 4 slices of 30 ms. On a loaded
machine the 30 ms timers fire late and merge the slices: the 0.2.0 candidate's full pass got 'sliced (3 calls)'. A
preload that blocks the event loop 9 ms of every 12 reproduces it 5 of 5 times. Now the predicate holds on the 4th call:
exactly 4 calls, each with the same predicate, args and polling, each slice 0–30 ms, and the wait lasted the three
slices (≥ 85 ms). It passes 5 of 5 under that lag and under 25 ms of every 27. A helper that stops at the first
timed-out slice still fails it. A test timing flake, no product change; golden unchanged.
Seed 22's only 麻痹 source was one 5 % 催泪瓦斯 on 洛洛: a single hit at 0.1.4, and since WE1's enemy damage frame
(89f730d0, server/sim/ai.js enemyAttack; it re-timed every fight) none of its rolls hits — 0 attacks in the 17 s boss
fight, 12 in the hidden one. The item, the status and the fx work: forced to 100 % the hidden field shows 26 麻痹 and
8 瘫痪. Every other check (digest / pool / fx / 限伤 / drones / 奥术) still runs on the unmodified specs. With the copies,
瘫痪 occurs on every field of all 19 seeds (of 1–120) that pass the other checks (seed 22: 4 / 7 / 29 / 13); with the
palsy fx removed from the sim the test fails (0,0,0,0). A test fixture fix, no product change; golden unchanged.
PRTS 卫戍协议：盟约 下半, 3月27日更新#2调整 (revision 439189): 「[Ⅳ阶]华法琳：赋予的特质的叠层上限从 初始12/精锐24
降低至 初始7/精锐14」 — a nerf; the current 参战干员 record says 至多7层 / 14层 and the client blackboard is 7 / 14.
Research 02 Addendum 1 had read the line backwards, so the remake overrode garrison_95's cap to 12 / 24.
Ported from PR #192 by @kukiC: GRANTED_CAP_OVERRIDE is removed from server/sim/content/garrisons/battle.js and from
the client-result layer bound (server/match/fields.js layerAllowanceOf); a granted trait keeps its data cap like a
native one. Tests from the PR: garrisons_battle (updated on purpose: the 7th cast reaches 7 / 14, the 8th adds
nothing), feedback1-gaps (the bound with 华法琳 + 古米), playtest3 (both cards show 7 / 14); the first two fail without
the fix. Docs: research 00 / 01 / 02 corrected (02-bonds.json's note; data/bonds.json regenerated offline — only
miraShip.spec.layerGain changed), history 0.1.1's pointer, DESIGN §25.13.2 and the §25.10 table.
Golden unchanged (no corpus battle casts 8 times in front of 华法琳).
The item text 「若已拥有至少2名该初始干员，则再获得1名该初始干员；否则随机获得1名同盟约初始干员」: the 否则 is the
case of fewer than 2 owned. builtinMeta's use_equip_reward_char_chess returned only when grantChess(base) gave a piece,
so an elite 溯光星源 plus 2 normal copies (all 5 Ⅵ阶 copies) fell through to the same-bond roll (白面鸮 in the report's
seed). Now with 2 owned it asks for the 3rd and stops; the item is still consumed. The bots' carrier choice
(itemTarget) skips 2 copies whose pool is out.
Tests: test/content/items.test.js (the report with the normal and golden item; 2 copies with the pool drained; 2 copies
with copies left still merge), test/match/bot-smarter.test.js (no carrier on a pair whose pool is out); both fail
without the fix. Docs: META built-ins and bot items, research 04 / 06, DESIGN §25.13.1.
Golden unchanged.
- tools/local-extract/extract.py TOKEN_SPINES (`--only spine/token`): each token's battle prefab in the
battle token packs (pkgrps/btl_pfb_tokens_*.ab; the Windows build has all 39, the iOS build lacks
btl_pfb_tokens_0), the skeleton of its Front (or only) Spine renderer, written to
public/assets/local/spine/token/<id>/ like the enemy models (sized pma atlas, premultiplied pages;
the newer RGBA pages keep their own alpha). The enemy writer is shared (write_spine; the two enemy
models re-extract byte for byte). The skinpack job of 乌尔比安's block is gone: it extracted
nothing, and that token draws no model in the game (nor do 圣聆初雪's, 酒神's, 贝洛内's and
予愿安洁莉娜's hidden tokens: no Spine renderer, no avatar in either install).
- tools/assets/local-token-spines.json (committed, `fetch-assets --local-spines`) → data/assets.json
tokens[id].spineLocal for 35 自选 summons, 凯瑟琳's 爬行号·防护单元 and 凛御银灰's three 风雪之眼; the
manifest is the same with or without the extraction (JSON diff: only the 39 spineLocal entries and
the hash). spine.mjs LOCAL_SPINE_ROLES maps clips the resolver cannot read (电弧's 戴乌 C_Skill1_*,
酒神's 本能的召唤 Loop / End, 白铁's platforms End, 战术锚点 on its Default pose) [ASSUMED where named].
- assets.js spineEntry: a token resolves its spineLocal like an enemy (no web model behind it: the
diamond stays when it is not listed or fails to load). Scale as the web tokens (UNIT.modelScale).
- setup reports an extraction without the summon models; docs: ASSETS.md "Token models from the local
client", DEPLOY.md §6, DESIGN §13 (design/client.md), README.
- Tests: test/local-token-models.test.js (new), test/local-extract.test.js (job table, prefab walk,
alpha merge), test/assets.test.js (the local token models load like the client does),
test/package.test.js (the full zip ships them), test/static-local-art.test.js. Golden unchanged.
README「方式一」lists the two zips: 完整包 Stronghold-Protocol-v<版本>.zip (about 400 MB, 590 MB unpacked, all the art)
and 精简包 …-lite.zip (about 21 MB: setup downloads the art, Spine models, audio, fonts, emotes and 玩法说明 pages on the
first start, about 440 MB, resumable; the local-client art is not in it), and what both leave out; the disclaimer says
only the full zip carries the art; setup's download is about 440 MB (the 0.2.0 manifest, was 270 MB); the disk row and
the English summary follow.
docs/DEPLOY.md: §0 disk (440 MB of art, 65 MB of runtime dependencies in a zip), §1.1 three ways to get the game (full,
lite, git clone) and what the first start does with each, §1.5 updating with the lite zip, §3 Docker builds from a git
clone (the zips have no Dockerfile; 440 MB), §6 a lite install has no local-client art, §7 the packaging tool for
maintainers: what ships, what is left out and why, the checks, what it needs.
tools/setup.mjs LOCAL_ART_COPY_HINT (shown by doctor): copy the local art from the same version's 整合包（完整包） — the
lite zip has none.
npm run package / npm run package:lite (node tools/package.mjs [--lite] [--dry-run [--list]] [--out <dir>] [--force]
[--keep-stage] [--no-install] [--allow-dirty] [--root <dir>]) replaces the maintainer's copy-the-tree script:
Stronghold-Protocol-v<version>.zip and Stronghold-Protocol-v<version>-lite.zip, one Stronghold-Protocol/ folder each.
Ships (an allowlist over git ls-files): server/, shared/, data/, public/ but public/dev/, the start scripts, the tools a
player runs (setup, vendor, fetch-assets + tools/assets, doctor, and what setup starts: tools/local-extract,
crop-board-atlas), the four research tables the Node sim fallback and fetch-assets read (docs/research 03, 05-enemies,
05-maps, 07 JSON — the first cut left them out, so a lite zip's setup could not plan the art), package.json / lock,
LICENSE / NOTICE / THIRD-PARTY-NOTICES, README, CHANGELOG, docs/PLAYING.md, docs/DEPLOY.md; npm ci --omit=dev in the
stage adds node_modules and public/vendor. The full zip adds the files data/assets.json and data/local-assets.json list,
public/fonts and public/assets/local/ — not the rest of public/assets, so art the data dropped (焰狐龙梓兰's 15 files)
never ships. Left out: test/, maintainer tools, scripts/make-windows-bundle.mjs, the other docs and docs/img, research
notes, handoff/, .github/, types/, lint / editor / Docker files.
Checks (also in --dry-run, which writes nothing): 0.1.x's refusal list plus test, .github, AGENTS.md, public/dev,
node_modules/.cache, .env / .venv anywhere; relative and server-mount imports of shipped modules and the player npm
scripts resolve to shipped files; a full zip has every listed art file; no two paths differ only in case; no shipped
file (text or binary, art included) has a /Users/, C:\Users\ or /home/ path or this machine's account name (read from
the OS at run time, SP_PACKAGE_SCAN_USER=0 / SP_PACKAGE_SCAN_NAMES; node_modules for the name only); no shipped tracked
file has an uncommitted change. A build also checks that the stage holds exactly the plan before zipping, and refuses an
--out inside the repository.
Measured on this tree (version 0.1.3 until the bump): full zip 421,421,613 bytes (14,910 files, 589 MB unpacked), lite
zip 22,088,912 bytes (6,430 files, 85 MB); the 0.1.x layout of the same tree (every tracked file, all of public/assets)
zips to 429,644,767 bytes: 569 files and 25 MB unpacked less. Both stages started (npm start, /healthz, assets), setup
--check is green on the full one, and the lite one plans the art offline.
test/package.test.js (9 tests, about 1 s, no network, no npm install): the selection on sample paths and on the real
git ls-files (imports, npm entry points, research tables and spawned tools ship), the refusal list, a dry-run of a
temporary git checkout with fake art (full and lite lists), the scans and guards, and one --no-install build there.
The official data spells one DEC X module 'dec-X' (uniequip_003_aglina) and the others 'dec-x' (WAH-Y is the other
mixed-case type id), and the asset plan stored each under its own case: module/dec-X.png and module/dec-x.png. On
Windows / macOS those are one file, so a release zip built there could hold only one of them and a case-sensitive
server (a Linux VPS) would miss the other — the release packager found it ("planned but not staged"). The local file
is now module/{typeIcon in lower case}.png; the upstream names keep their order (the id as spelled, then lower case).
data/assets.json regenerated offline (node tools/fetch-assets.mjs --offline --add-only). JSON diff: modules.dec-X
/assets/module/dec-X.png -> /assets/module/dec-x.png, modules.WAH-Y /assets/module/WAH-Y.png -> /assets/module/wah-y.png,
hash, stats.files 6979 -> 6978, stats.bytes 458334505 -> 458333444; nothing else.
fetch-assets compares its orphans (what --prune deletes) without case: on Windows / macOS module/WAH-Y.png on disk is
the listed module/wah-y.png and must not be deleted (orphanFiles, exported).
Tests: test/assets-diy.test.js (one lower-case file per type, no two manifest URLs differing only in case, orphans).
docs/ASSETS.md: the module icon file name.
sganggs