The §22 intro maps GitHub #25 … #35 / #16 to §22.1–§22.13 and records the owner's decisions of 2026-10-04.
§22.7–§22.12 are the wave-2 subsections verbatim; §22.13 is PR #48 (要塞 cannot hit air units, 莫斯提马's slow shows its icon).
One blank line before every §22 subsection, as in §21.
余 S2 (SKILL_RANGE on its x-1, §22.10) is appended wherever the six-skill deviation is listed: DESIGN §5.6, §20.2, §21.20, §21.29,
DATA §2 item 6, research 00 and 03 (SKILL_RANGE counts: 13 / 26 official, 14 / 28 in the data). DATA §2.1 names 要塞 in canHitFly.
§22.2: seen / official / cause / now / cost / players notice / [ASSUMED] / tests for GitHub issue #51. Research 02
§3.18's [ASSUMED] landing line (md and json) now states the rule; `node tools/build-data.mjs --offline` changed only
data/bonds.json raidShip.spec.assumed[0].
Up to 0.1.1 only the 技能就绪 trigger required a landing tile that covers the target; the 10 s idle trigger took
raidTile's fallback (a tile that covers nothing), so a member with nothing reachable hopped every 10 s between useless
tiles and never went for a second ground enemy it could have hit. Now both triggers walk the candidates in the existing
order (own field first, most advanced first, at most 8) and jump to the first one with a covering tile; none -> the
member stays, its idle time keeps counting and the 0.25 s poll looks again. [ASSUMED]: the official text only says
"再部署至一名地面敌人周围".
raidTile now enumerates only the tiles a range offset leads back to from the enemy's body (same choice as the old
25-tile scan whenever a covering tile existed: 200 000 random cases), and a player's members share one sorted candidate
list per poll (rebuilt after a jump), so a search that finds nothing stays cheap: +5.8 us per tick on the heavy boss
field with 18 members that never reach anything, against +5.5 us for 0.1.1 (the plain requirement on the old scan:
+29.6 us). Tests: two "突袭 #51" cases in test/content/bonds_addon.test.js.
131 of the 135 Back skeletons have no Die clip, so SpineActor.die() changed nothing and the
attack / skill / idle loop went on under the redeploy ring. UnitView._wantsBack() now wants the
Back model only while the unit stands (or when its Back skeleton has a Die clip of its own);
die() / revive() mark the view and its update swaps the model in place after the frame's events
(a death and a redeploy in one frame load nothing). The Front model replays the fall (held end
for a unit already down), gets the knocked-down grey, and the redeploy brings the Back model back
with the deploy clip carried over. A skeleton without a Die clip holds its idle's first frame.
- DESIGN §21 intro: §21.25 now credits PR #2, #7, #10 and #14; the closing additions are mapped (issues #1 / #8 gaps
§21.26, issue #5 / #8 item 5 client fixes §21.27, the draft's match info §21.28, the six 重装 skills' deliberate
trigger deviation §21.29), the merge order names the four closing branches (their three "§21.26" subsections
renumbered §21.26-§21.28 in merge order) and GitHub's master, and the normative lines they rewrote are listed.
- DESIGN §10: a strategy built around a switched-off bond carries the 本局禁用 mark in the strategy draft (§21.26).
- CHANGELOG 0.1.1 intro names the GitHub issues and pull requests handled.
- docs-consistency: §21.1-§21.29 present and numbered consecutively; the intro's new mappings.
test/sim/robustness.test.js "performance: heavy 2-player boss field" failed on GitHub's windows runners (0.51-0.58 ms
per tick against the 0.5 ms bar) on PR #10, PR #12, PR #14 and master's own push. The bar stays 0.5 ms on a development
machine and is 1.0 ms when process.env.CI is set (GitHub sets it); the measured time is the best of three runs of the
same battle after two warm-up battles, each run still checking the peak, the 18 operators and zero errors. Documented
in the test header, DESIGN §11 and the CHANGELOG's 工具 section.
css/screens/loadout.css: `.lo > * { position: relative }` (one class, a later rule) overrode `.lo__bg { position:
absolute }`, so the mint glow / grid layer was a 0-px flex item and never rendered. `.lo > :where(:not(.lo__bg))`
keeps the one-class specificity for every other layer and never matches the background.
Checked in headless Chrome at 1920x1080 and 844x390 (touch): the layer now covers the overlay (0 px tall before),
every other box of the screen is unchanged, no console errors. Tests: test/ui/loadout.test.js (the cascade; fails on
the old rule), opt-in test/ui/loadout.e2e.test.js (the layer's box and stacking at both sizes). DESIGN §21.25 credits
PR #14 and PR #10 (crossOrigin on the two new Image() loads); CHANGELOG 0.1.1 lines for both.
The owner's decision (2026-10-03, after community feedback): 深巡 S2 行动能力剥夺, 雷蛇 S2 反击电弧, 号角 S2 暴风号令
and S3 终极防线, 灰毫 S1 攻击力强化·γ型 and S2 专注轰击 take the basic strategy (DEFAULT) instead of the 下半 TANK class
row's TAKE_DAMAGE. Every other MANUAL 重装 skill, 深巡 S1 included, still waits for a hit.
- tools/build-data.mjs TRIGGER_DEVIATIONS: a small table keyed by the normal chess id and the skill id (the elite
follows through baseId), never by the skill id alone - 灰毫 S1 is the generic skcom_atk_up[3]. resolveTrigger applies
it last; rawRule keeps the official TAKE_DAMAGE; validateAll fails the build when an entry no longer meets a
TAKE_DAMAGE row. data/chess.json rebuilt with --offline: exactly the 12 records' rule changes.
- PR #12's kit lines (trigger 'DEFAULT' on 深巡 / 雷蛇 S2) stay and agree with the data; their comments and tests now
give this reason instead of the community summary.
- Tests: test/sim/feedback1-tank-triggers.test.js (the data, kit = data, a real battle on 战场#01(下半) in which each of
the six casts with an enemy in range before any hit and each of the other 46 MANUAL 重装 records casts on the first
hit); data.test.js trigger expectations; the 灰毫 / 号角 kit tests follow; docs-consistency.
- Docs: DESIGN §21.29 (new), §5.6, §19.2, §20.2, §21.20 (settled row), §21.23; DATA, SIM, PLAYING, CHANGELOG 0.1.1;
research 03 §1.4 drops the arknights.wiki.gg "offensive / defensive" summary and records the activity_table history
(上半 had no TANK row, 下半 added it) and this deviation; research 00-INDEX.
Conflicts resolved keeping both sides:
- bandDraft.js header: the 本局禁用 mark of a strategy tied to a switched-off bond, then the 本局信息 dialog; the code
merged cleanly (the card's 本局禁用 tag and the detail note sit beside the dialog's state and button).
- draft.css: the 本局禁用 tag / note rules, then the 本局信息 entry and dialog block (its header cites DESIGN §21.28).
- CHANGELOG 0.1.1 界面: the gaps and issue #5 / #8 lines, then the dialog line.
- PLAYING §6: the disabled-bond line says where to look again (draft dialog, in-game 🔍); the 变形同构体 line keeps its
pairing explanation.
- DESIGN: the branch's "§21.26 Match info in the strategy draft" becomes §21.28; its two §10 references follow.
Conflicts resolved keeping both sides:
- CHANGELOG 0.1.1 界面: the three gaps lines, then the issue #5 and #8 lines.
- DESIGN: the branch's "§21.26 Client fixes for GitHub issues #5 and #8" follows the gaps' §21.26 as §21.27; its
references in §9 (setCamera `shop`, the model fallback), §17.1 (the quiet budget) and §19.8 (the collapsed shop)
now point at §21.27.
Conflicts resolved keeping both sides:
- effectsMeta.js header: the QA residual's "every [holder, item] pair, taken before the first item runs" and "an item
moved during it runs at most once", with the cell branch's wording of what 突变细胞 does mid-walk.
- DESIGN §6.6: the band pick that skips a strategy tied to a switched-off bond and the deployment of the operator a
transformation gains / the bench shed that spares it.
- DESIGN §21.25: PR #2's bench rule credited (the cell branch) plus the QA residual's snapshot-timing sentence.
- PLAYING: the new 突变细胞 sentence with the residual's hammer sentence (revives, borrowed hammers, the lend window).
- docs-consistency: both new tests (batch 6 QA residuals; 突变细胞 after the WA merge).
The hammer lock per deployment for borrowers and in-place revives (both [ASSUMED] where no source decides), 卢西恩 / 锏
casting for an airborne 蒂比, rule 3 for a summon that is gone, the dispatcher snapshot taken before the first item;
PLAYING and CHANGELOG no longer say every debuff misses an airborne 蒂比 (ground-enemy auras and counters still reach
her); research 04's hammer line says once per deployment. docs-consistency pins it.
卢西恩's 【aoe】 and 锏's CircleAttack counted every ally in their radius, so an airborne 蒂比 alone in range set them off
and the blast skipped her. They count only allies a ground enemy does not evade (evadesGround).
A summon leaves its home free only once it has expired or been killed, and _layBody skipped removed pieces, so a body
on the home of 浊心斯卡蒂's expired 海嗣 stayed there and kept the 海嗣 from coming back. Every board piece's home counts
now. _checkRedeploys tests the DP before the tile scans, so an operator waiting for DP no longer scans every ally.
Step 5 read each holder's items when the walk reached that holder, so an item equipped meanwhile onto a holder not yet
walked ran in the same dispatch and an item moved to a later holder ran twice (latent: no shipped item does either).
Every [holder, item] pair is now taken before the first item runs, as the header and META already said.
The lock and its re-arm hook lived in the carrier's hammer grants, so a hammer lent by 萨尔贡 × 娜仁图亚 came back spent
in the borrower's next deployment once the lend had run out while it was down, and a window started through the lend
ended with it. deploymentOf marks the deployment whose lock is spent; one battle-level hook holds every running window
(holdsUndying, also what 信仰搅拌机 S2 steps aside for), so a window outlasts the lend. M3茧甲 and 埃芒加德 revive in place
for PRTS's 0-time / 0-cost redeploy and now open a new deployment (revivedInPlace), like 阿戈尔 5's real redeploy.
At 640×360 the mock's first shop card completes a merge; its detail opens in
the left slot and moves to the right one 20-40 ms later because the elite's
lit board tile lies under the left slot (screens/game.js dSide, DESIGN
§20.11). The check caught the left card only sometimes and then measured the
corner buttons against the right one. It now taps a card without a merge and
asserts the card stays left.
A reload (a phone browser or Chrome's Memory Saver discarding the background tab) made the asset store download
/data/assets.json a second time; createFieldView waited at most 4 s for it, every view looked its model and avatar up
once, and a failed download kept an empty manifest for the page. Now the field host seeds the store with the game
data's copies (seedAssets); without one, ready() retries like data.js, never keeps a failure and tries again with a
bounded backoff, and onChange makes every view re-resolve (UnitView.retryAssets, ItemView.setIcon). A failed or
timed-out Spine load is retried by its view (SPINE_RETRY_MS, acquire { retry }) and at once when the tab is shown.
Battle mode holds the Spine cache (spine.hold()), so a battle that began in a hidden tab keeps its models; the idle
budget still applies and the quiet budget still frees everything in the lobby, room and result.
sganggs