feedback7 R22-LOADOUT: the data carries every potential (潜能 1–6), not only full potential; golden: unchanged

Potential becomes a per-player setting (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: 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」). 0.2.1 baked full potential into the data (§26.14).

- tools/build-data.mjs runs the chess / backups / tokens builders once per potential rank (ctx.potRank, 0–5) and keeps
  the full-potential build annotated with what each lower rank changes: `potDown` { rank: { dotted.path: value } } on
  chess records, backups.json forms and token variants (stats, statsBase; 望's 棋子 deployLimit / deckStack, 多萝西's
  共振装置 count), and chained talent entries (`potMin` / `potBelow`: the candidates by requiredPotentialRank, the
  fields a lower rank changes) in talents / talentsBase / modules[].talentChanges / variant talents. A self-check proves
  shared/potential.js atRank rebuilds every rank exactly; the unit-form parity runs at every rank.
- shared/potential.js (new): atPotential / atRank, talentAtRank, stripPotential, and the 练度 helpers used next
  (cultivateMul from the effects.json CHAR_MAP records aceffect_char_1…4, cultivationOf, cultivatedStats).
- composeTalents / composeUnitRecord return potential-resolved lists (no chains): a composed record is built at one
  potential (its form resolved first).
- JSON compare: the new chess / backups / tokens minus the annotations equal the old files byte for byte; every other
  data file identical; data/i18n/* rebuilt (the chained talent texts translated at the old rates: en 95.4 %, ja 95.2 %,
  ko 95.2 %, zh-TW 90 %). Data 5.21 → 5.51 MB.
- Docs: DATA.md §2.3 (new), §2, §7, §14, §16, §18; research/03-operators.md; kits/README.md.
- Tests: test/potential.test.js (new, 7): every chess and 自选 form at every potential = the raw keyframes plus the
  potential steps up to it, every talent = the last candidate with requiredPotentialRank ≤ the rank, 刺玫 at each
  potential, summons at the owner's potential (Mon3tr 3 s / 1200, 望's 棋子 6 / 7, “小自在” 15), the 练度 table =
  cultivateEffectList; data.test.js compares the composed talents without annotations.

test/data.test.js

@@ -16,6 +16,7 @@ import { join, dirname } from 'node:path';
 import { tmpdir } from 'node:os';
 import { spawnSync } from 'node:child_process';
 import { fileURLToPath } from 'node:url';
+import { stripPotential } from '../shared/potential.js';
 
 const ROOT = join(dirname(fileURLToPath(import.meta.url)), '..');
 // DATA_DIR lets the suite validate an alternative build output (e.g. `--out /tmp/x`).
@@ -558,7 +559,8 @@ test('chess: golden modules[] (+ statsBase/traitBase/talentsBase) compose back t
     // default loadout = the record's own stats / trait / talents
     assert.deepEqual(composeStats(c.statsBase, dm?.attr), c.stats, `${c.chessId}: statsBase + default attr = stats`);
     assert.deepEqual(dm?.traitOverride ?? c.traitBase, c.trait, `${c.chessId}: trait`);
-    assert.deepEqual(composeTalents(c.talentsBase, dm?.talentChanges), c.talents, `${c.chessId}: talents`);
+    // (the record's own lists chain their lower-potential entries — shared/potential.js; a composed list carries none)
+    assert.deepEqual(composeTalents(c.talentsBase, dm?.talentChanges), stripPotential(c.talents), `${c.chessId}: talents`);
     assert.equal(new Set(c.modules.map((m) => m.uniEquipId)).size, c.modules.length);
     for (const m of c.modules) {
       nMods++;