@4268` `` in the location card, styled by `` `#corpse-chip {@1759` ``; rendered by `` `function _renderCorpseChip()@36158` ``, called from `storyUpdateStatus` at `` `_renderCorpseChip(); // §DEATH-01 Inc B@36151` ``. Text escapes through `` `+ ' at ' + _mpEsc(where);@36166` ``; the `` `el.title = 'Your remains lie out in the world@36167` `` tooltip itemises each grave. Local-grid marker at `` `const _graveHere = (S_story.corpsesQuests || []).filter@36724` `` → `` `cell.classList.add('mc-grave');@36727` `` inside `` `function _renderMapGrid()@36672` ``. **NG+ guard (Inc C).** `` `function storyNewGamePlus()@24001` `` opens with `` `const bodies = S_story.corpsesQuests;@24006` `` and `` `const ok = window.confirm(@24010` `` — still the only `window.confirm` in 38,712 lines — placed before the NG-title overlay so a decline is a clean no-op. **The four "where is my body" surfaces at HEAD**, two pre-existing and two added here: journal `` `hd.textContent = '☠ Fallen Hero — Corpse Retrieval';@30819` `` · node card `` `_mkSection(null, '🦴', 'Remains')@35732` `` · the chip · the grid marker. Recovery runs through `` `function _storyRetrieveCorpse(questId)@26035` ``, which calls `storyUpdateStatus()` and therefore re-renders the chip — **the signal cannot outlive the body it points at.** --- ## V. Spec → shipped delta | Spec item | Shipped | Delta | |---|---|---| | **Inc A** honest adaptive message | ✅ verbatim to the specified shape | Adds an unspecified zero-loot branch — *"Your body lies there, though it carried nothing of value"* — which is what *"never prints 0 item(s)"* actually requires | | **Inc A** finding #5, no code change | ✅ comment only | Comment repeats the report's wrong mechanism (§VI) | | **Inc B** chip in `#story-location-hd-card` after `#objective-chip` | ✅ | — | | **Inc B** chip hidden at 0, `"N bodies at "`, `"N places"` for >1 | ✅ | — | | **Inc B** `title` itemises each grave | ✅ | — | | **Inc B** `onclick = storyMapToggle` | ✅ inline attribute | Escapes visible text through `_mpEsc` — a hardening the spec did not ask for | | **Inc B** `.mc-grave` + 🦴 overlay + hover text | ✅ hover string verbatim | **Two blind spots** (§VI) | | **Inc B** CSS *"light + dark theme rules"* | ⚠ **one palette only** | `#corpse-chip` ships a single dark palette — but so does its `#objective-chip` sibling, so the chip matches its neighbour. Cosmetic; no row | | **Inc B** *"wires the long-dead `.corpse-chip` CSS"* (ship commit) | ❌ **NOT SHIPPED** | Inc B built an **ID**-based chip beside the dead **class** rules (§VI) | | **Inc C** `window.confirm` guard, decline aborts | ✅ byte-for-byte as drafted | — | | **Inc D** gold-% / equipped-drop | ✅ correctly dropped | — | | Docs synced to `mechanics.md` + `docs/mechanics/mechanics-combat.md` | ✅ both carry §Corpse-Run | `mechanics.md:595` repeats the *"ends in `storyAutoSave()`"* phrasing | | Test `src/tests/integration/death-loot-grave.test.js` | ✅ 4/4 green | **No `mc-grave` assertion**, though the plan names one | | Regression `effort-xp` 3/3 · `enemy-ai` 4/4 · `courier-map` 1/1 | ✅ **3 / 4 / 1 exact** at ship *and* at HEAD | — | --- ## VI. Findings ### Correction #2 — finding #5 was right for the wrong reason, and is now proved The report argues the death is persisted atomically because *"`storyRender` … ends in `storyAutoSave()`."* **It does not end in it.** At HEAD `storyRender` spans 34567–36056 and `` `storyCheckJournal(node);@36047` `` is followed by the save, then by **seven more calls** — `` `storyCheckVictory(node);@36049` ``, `_renderMiniMap`, `_renderWorldMiniMap`, `_renderGlobeMap`, `_updateExitLinks`, `_updateWaypointBtn`, `_updateHuntBtn`. The same loose phrasing was copied into the shipped code comment and into `mechanics.md`. The conclusion holds, and this pass replaces the assertion with a measurement: a brace-walk of all seven post-save callees finds **zero writes to `S_story` in any of them**. The strip and the corpse push precede `storyRender`, the save is synchronous with no `await` on the path, and nothing after it mutates persisted state. **Finding #5 is STALE, and the reason is stronger than the one given** — not *"the save is last"* (it isn't) but *"everything after the save is read-only"* (it is). Unenforced: no gate pins it, so a future render layer writing `S_story` after line 36048 re-opens it silently. ### Defect A — the dead affordance the ship notice claims to have revived (→ §DX-02dc 🟢) The report correctly identified `.info-chip.corpse-chip` as *"a dead affordance"*: four CSS rules — `` `.info-chip.corpse-chip { border-color: #a07840;@1317` `` and its `:hover`, plus `` `.info-chip.corpse-chip { border-left: 3px solid var(--border);@2906` `` and its `:hover` in the light-theme block — with no JS to apply them. The ship commit then announced that Inc B *"wires the long-dead `.corpse-chip` CSS."* It did not. `_renderCorpseChip` resolves `document.getElementById('corpse-chip')` — an **ID**. The class form `info-chip corpse-chip` has **0 commits in the entire repository history, all branches, all paths.** The rules were born dead in `213d14b` (2026-05-26) and are dead today, now with an ID-based look-alike in the same stylesheet. Building a new chip was right — the location card is not an info-chip row — but the tombstone was left standing. This is §AUDIT-03aa's *"comment asserting the removal of a symbol that still resolves"* run in reverse: **a commit asserting the revival of a selector nothing applies.** ### Defect B — the grave marker's two blind spots, and the assertion that was planned and never written (→ §DX-02dd 🟢) Proved in a real browser, not inferred. `_renderMapGrid()` was driven three times against planted corpse geometries: | Grave position | `.mc-grave` cells painted | |---|---| | Visited node, in the ±7×±10 window, not the player's cell | **1** (with its `.mc-grave-mark` 🦴) ✅ | | **The cell the player is standing on** | **0** | | In-window node that is **unvisited and not on the recent trail** | **0** | Both zeroes are structural. The marker lives inside `_renderMapGrid`'s `else if (code)` → `if (isVis || isTrail)` branch, so the `isCurrent` arm — which paints `◉` and returns — wins on the player's own cell, and an unvisited cell falls to a third arm the marker never reaches. Neither is a loss: standing on the grave, the node card's 🦴 **Remains** section carries the signal *and* the retrieve action; unvisited, the chip still names the place. But the map is **silently inconsistent with the chip beside it**, and the otherwise-exhaustive honesty ledger mentions neither case. The verification plan states the Inc-B test will assert *"Local-grid marker present on the grave cell when in view."* **The shipped test contains no `mc-grave` assertion, and no other spec in the repo references the class.** The one Inc-B surface touching the map renderer is the one surface with no coverage — which is how both blind spots survived 37 days. --- ## VII. Invariants scored @ HEAD | Invariant | Verdict | |---|---| | **§CELL-13 — no jump travel, ever** | ✅ the chip's only action is `storyMapToggle()`; `checkpointNode` respawn remains the sole warp. `` `function storyMapToggle()@36659` `` opens a tab and moves nothing | | **DUEL:CORE untouched** | ✅ every edit is single-player `S_story` | | **Free-Movement — no gate or mover reads corpse state** | ✅ all 14 `corpsesQuests` sites are defaults, death, retrieval, or render | | **Shards + gate keys protected** | ✅ `critTypes` unchanged — but see §DX-02ax: the set filters on `type`, so two items carrying the ornamental `drop:false` and no `type` are **not** protected | | **`S_story.xp` monotonic** | ✅ untouched | | **Corpse chip cannot outlive its body** | ✅ *(not stated by the report; verified here)* — `_storyRetrieveCorpse` calls `storyUpdateStatus()` | --- ## VIII. Honesty ledger — re-measured at HEAD, 37 days on All three deferrals the author declared are **still honestly deferred**; none was quietly closed and none was quietly built. - **Map region-centering on the grave — still NOT built.** `mapViewR` / `mapViewC` have **0 occurrences in any product file**; `mapViewR` appears in exactly two places in the repo, both prose (`plan-archive.md`'s §NAV-01-FU item 4 and this report). The proposed name never became code, which is the correct outcome for a deferral. - **Grave markers on the World / Full / GLOBE canvases — still NOT built.** `_paintFullWorld`, `_renderWorldMiniMap` and `_renderGlobeMap` contain **0** `corpsesQuests` references; only `_renderMapGrid` has one. Local grid only, exactly as declared. - **The corpse array is still unbounded.** The only `splice` on `corpsesQuests` is the *recovery* path at 26038 — there is no cap anywhere. Fifty deaths still means fifty records persisted to `localStorage`, fifty journal rows, and fifty `nodeName`s in one `title` attribute. Still minor, still not a loss — but *"noted for a future soft-cap"* has been noted for 37 days and filed nowhere, which is the shape instrument 112 exists to catch. Now filed as §DX-02dd (b). **Added to the ledger by this pass:** the Local grave marker does not paint on the player's own cell or on an unvisited cell (§VI B), and `#corpse-chip` ships one palette rather than the two the spec named (§V). --- ## IX. Acceptance `npx playwright test src/tests/integration/death-loot-grave.test.js` — **4 passed (3.5 s)** at HEAD: - Inc A — equipped slots unchanged post-death; message contains *"stayed with you"* / *"Your body lies there"* / *"You wake at"* and never the old lie; corpse holds 2 items + 100 gp; the shard stays in inventory; gold zeroed; `coc_autosave` written. - Inc A — zero-loot death reads *"nothing of value"* and never *"0 item(s)"*. - Inc B — chip hidden at 0, names one place at 1, *"2 places"* at 2, click calls `storyMapToggle`. - Inc C — decline: one `confirm` call, corpse kept, sentinel field intact, nothing wiped; accept: corpses wiped; no corpse: **0** `confirm` calls. Regression baselines re-counted at both the ship commit and HEAD: `effort-xp` **3** · `enemy-ai` **4** · `courier-map` **1** — the report's figures are exact at both trees (instrument 106 clean). --- ## X. Defects filed - **§DX-02dc 🟢** — `.info-chip.corpse-chip`: four CSS rules, zero class applications in the repository's entire history; the ship commit claims to have wired them. Retirement-sweep member; argues for extending the dead-symbol census to CSS class selectors. - **§DX-02dd 🟢** — (a) the Local grave marker does not paint on the player's own cell or on an unvisited in-window cell, and has **no test**, though the verification plan named one; (b) the corpse array is unbounded — the author's own deferral, unfiled for 37 days. Pre-existing rows that touch this surface and are **not** re-filed: **§DX-02ax** (`drop:false` is read by nobody; `_storyDeathSaveFall` filters on `type`, so Glut's Gift goes to the corpse) and **§DX-02bt** (the journal's `🦴 Your body at` row interpolates a save-file `nodeName` through `innerHTML`). The chip added here escapes through `_mpEsc` and is not a member. --- ## XI. Verdict **A design lock that shipped whole, with the most honest honesty-ledger in the corpus.** Every increment works 37 days on, the acceptance suite is green, both regression figures are exact, and all three declared deferrals are still genuinely deferred — a combination this program has not previously found in one report. The author also drew the right conclusion about finding #5 from the wrong evidence, which is rarer and more interesting than being wrong. The failures are all of one kind: **claims about what the pass did to its own surroundings.** The ground truth was re-measured mid-edit and half of it silently moved to a different build; the ship notice announced a revival that never occurred; the verification plan promised an assertion never written. Only the last cost anything — and it cost exactly what such omissions always cost, since the untested surface is the one carrying both defects found here. The corpse-run itself is intact, and it is the good kind of mechanic: it turns a death into a place you have to walk back to, and the game now tells you where that place is from anywhere on the map — without ever offering to take you there.