fix: correct community-reported combat and strategy draft behavior

Guard lethal stuns against a replacement deployment; keep bard auras running
through stun; require an active Aegir bond for Clementia's strategy; allow
Compass SP to bypass recovery locks; and refresh Hidden Core gun mitigation
against current gun/spring companions. Set multiplayer strategy picks to 50s
and describe renderer fallback without diagnosing unsupported hardware.

Golden moves (full 285-scenario regeneration):
- roster-008: inactive Aegir no longer gains 18 strategy layers.
- boss-boss_4-solo: bard regeneration continues through stun.
- hidden-boss_9-pair: mitigation covers first hits/current companions.
- standin-07: hammer stun cannot follow an immediate revival deployment.
No ending-reason changes. Sources and report-by-report findings: DESIGN 28.25.

docs/history/0.2.3.md

@@ -348,3 +348,50 @@ and the queue clears. Both tests fail against the previous implementation. Toget
 刺玫 healing suites, 273 tests pass. The benchmark now writes actual diagnostic messages beside nonzero error counts,
 so failures cannot be mistaken for ordinary performance results. Recomputed golden and final benchmark results are
 recorded with the candidate; do not reuse the earlier, erroneous benchmark as a clean result.
+
+### 28.25 Community feedback of 2026-10-10
+
+The current development candidate was checked against thirteen reports, including feedback explicitly describing
+0.2.1. Sources: [PRTS mode strategies](https://prts.wiki/w/卫戍协议：盟约),
+[alliance/item notes](https://prts.wiki/w/卫戍协议：盟约_下半/PRTS盟约记录),
+[吟游者 branch notes](https://prts.wiki/w/分支特性信息/data),
+[假想敌：铳](https://prts.wiki/w/假想敌：铳), the cached official enemy and activity tables, and the owner's supplied
+克莱门莎 official showcase. The multiplayer 50-second setting is the owner's explicit official-play requirement.
+
+| Report | Verified result / change |
+|---|---|
+| 1. Lethal hammer/sailor stun after 不屈 | Fixed. Enemy normal attacks snapshot target deployments before damage. The subsequent periodic stun cannot affect a replacement deployment created during that strike. A surviving target is still stunned normally. Covers both hammer variants (third strike) and both dock sailors (fourth strike). |
+| 2. Opening strategy time | Multiplayer picks now have 50 seconds each, including skips; the header and per-turn deadline remain one clock. Solo and single-human matches remain untimed. Regenerated `config.json` from the builder. |
+| 3. Fartooth + dollkeepers | Already fixed in §28.1. Existing actual Kazemaru and Specter alter swap/return tests pass, including active-bond gating, caps and repeated swaps. |
+| 4. Gavial S2 pulls final boss | Not reproduced. Both tier-4 and tier-5 S2 implementations use the shared pull API, which rejects boss-pool leaders. New actual-S2 tests hit a final boss without moving it and additionally reject exaggerated pull force. |
+| 5. Clementia strategy gains inactive Aegir | Fixed. PRTS explicitly requires activation; the former `requireActive: false` and its old test contradicted that note. An inactive bond now gains nothing; an active one gains the knocked-out member's tier, still disabled in unite/boss battles. This is the strategy, separate from the new playable operator. |
+| 6. Pith/Aegir ATK in 0.2.1 | No confirmed current defect. Devoured ATK is `atkFinal`, added after an operator's own ATK percentages, then scaled by independent multipliers; existing numeric tests check this order. The report lacks the full lineup, skill/equipment state and exact unrounded stats, so 5340 versus “about 6040” alone cannot establish a discrepancy. |
+| 7. Stunned bards stop regeneration | Fixed for the default trait and Skadi alter's custom continuous trait. Aura refresh requires being on-field, not being able to act. Stun still blocks attacks/skill use, and leaving stops the aura. |
+| 8. Stationary redeployment loses Raid | Current code grants the ATK/HP bonus to the deployment made by Raid's relocation. Ordinary redeployment clears it; the existing passive-skill revive tests do jump promptly when their conditions hold. PRTS's “期间” does not settle whether an ordinary stationary redeploy should retain it. Preserve this rule pending an official comparison of that exact case. |
+| 9. Gun boss “90% reduction absent” | The official value is damage ×0.2 (80% reduction), already implemented. Fixed an actual coverage gap: PRTS includes another gun as well as springs in the condition. Also refresh on hit so the first hit and immediate companion removal cannot see an outdated aura. |
+| 10. Compass/tea +3 bypasses SP lock | Fixed. PRTS says all 黄沙罗盘 SP bypasses 阻回. A documented `ignoreLock` option permits the first-end refund and combo gift during `noSp` or a running skill, including the caster. Ordinary SP gifts still obey both locks. |
+| 11. Dobermann whip random bounty | Already implemented as a private three-choice selection (§27). Existing choice and reconnect tests pass. A timer/bot can still resolve an unanswered choice automatically. |
+| 12. Lemuen ammo does not gain layers | Not reproduced with real S3 consumption, normal and elite. Each ten bullets grants +2/+4 to each active named bond when at least three own operators share the row. Two actual casts cross that threshold and gain the expected layers. The existing garrison tests also cover fewer than three in the row. |
+| 13. Sudden simplified view | The message is emitted for any renderer startup failure, including import timeout, not exclusively unsupported hardware. Reworded it to describe the failure and offer reload. The prior mount-cancellation/late-disposal fix remains; the report alone does not identify this user's browser failure. No forced persistent 2D preference is written. |
+
+The new same-strike revival, bard and SP-lock regressions fail seven checks against the pre-change simulation and
+pass after these fixes. The additional band and boss tests exercise inactive activation and same-tick reduction.
+The validation record accompanying the candidate lists full CI, browser recovery and golden results.
+
+The official showcase was reviewed in talent/S1, S2 and S3 segments. S2 shows a moving cabin then a stationary vortex,
+including the second battlefield demonstration. S3's repeated bombardments agree with the documented three-hit,
+one-second pulse interval. Crownslayer's supporting skill and smoke are visible in the later S3 segment and are not
+attributed to Clementia. Edited footage does not settle the absolute first-bomb offset, S2 splash radius, or all exile
+status rules; the existing marked assumptions remain. No visual resemblance is used as proof of extra immunity.
+
+Full golden regeneration covers all 285 scenarios. Four digests intentionally move relative to the preceding candidate:
+
+- `roster-008`: 崇高牺牲 no longer grants 18 阿戈尔 layers while inactive; combat damage and kills are unchanged.
+- `boss-boss_4-solo`: the bard's regeneration continues during stun, changing healing, subsequent attacks and snapshots.
+- `hidden-boss_9-pair`: 铳's reduction is checked for the first hit/current companions; boss HP left changes
+  2,992,115 → 2,992,511.
+- `standin-07`: hammer stuns no longer cross the immediate revival deployment provided by M3茧甲; kills 9 → 10,
+  counted leaks 71 → 70, with corresponding combat/RNG/snapshot differences.
+
+No scenario's ending reason changes. The bonds, matches and DIY families are byte-identical to the preceding golden
+files. The 50-second setting and fallback-message change do not alter battle rules in those digests.