This Run` — with **nine** rows and **no career figures at all**, then and now (`function _populateGameoverChronicle() {@23852`). The two-column ledger lives on the **character sheet** (`function storyRenderCharSheet() {@37609`) and its headers are **"This Life" / "All Lives"**, the second suppressed until the first death. | | *"presented without blame or judgment — just numbers"* | ✅ **exactly right** — nine labelled rows, no commentary anywhere | --- ## V. Findings ### V-A. The headline: a certification contradicted by its own line range Register 4 states, in a table cell whose line numbers are **both exact**, that `_D100_TABLE` (22,456–22,477) is *"Active — consumables only, no magic weapons."* At `89fa13b` those 22 lines contain: ```js { weight:5, _type:'dagger', _magic:1 }, … +2, +3, +4 { weight:8, _type:'mainweapon', _magic:0 }, { weight:6, _type:'mainweapon', _magic:1 }, … +2, +3, +4 ``` Nine weapon rows, **eight of them positive-magic**, and live: `_rollD100Loot()` branches on both types, gated only by `_magicTierAllowed()`. Sixteen rows summing to 100, so a level-20 character had a **2 % chance of a +4 mainweapon from any ordinary kill** and a level-5 character a 5 % chance of a +1 dagger — while the same page declares fishing the exclusive source of positive magic and instructs future authors *"Do not add weapons to `_D100_TABLE`."* Why: §DROP-01 shipped the purge at `440eb5d` (2026-06-05 10:41) and **`88d41d1` reverted it at 13:33 the same day** — a commit whose subject is *"POST /api/import/book: documentation + smoke test cleanup"*, the silent stale-working-copy revert §DOC-02v caught red-handed. The purge did not return until `0fffce7` on 2026-07-07, **68 days later**. So this document was written into the gap, and its author read the *source report* rather than the range they cited. ***The lesson is not that the author was careless — every number on the page is right. It is that a synthesis inherits its sources' intentions and re-publishes them as observations. When a document says "reads the report against the live HTML," the one row where it did not is invisible, because the citation next to it is perfect.*** ### V-B. The mechanism restated in the summariser's own words is the one that changed The Necklace of Knowledge is awarded on **first rest at a location**. The source report says so five times (*"the first time the player rests at each unique location"*), and the engine agrees: two call sites, both rest paths. This synthesis renders it as **"first arrival at each node"** — and then builds a rule on the error: *"It should be the only reward for 'first visit' — do not create a parallel 'first visit' mechanic."* Both halves fail. There is no first-visit reward in the game at all (`visitedCells` feeds the minimap and grants nothing), and the parallel mechanic the rule forbids **was already in the file**: the Boyscout Token for a first short rest at a node (`// Boyscout Token — first short rest at this location@25849`) and Boyscout Night for a first sleep — *"🏕 Boyscout Night! Double rolls"*. Three first-time-here rewards keyed to the same act, and the rule written to prevent exactly that. ***A summary is the most dangerous sentence in a document, because it is the only one nobody checks against a file.*** ### V-C. The advice that is a trap: `_checkLevelUp()` Register 1 closes with *"`_checkLevelUp()` is the single call site for all XP gate processing … Any new XP source should call it."* Following that advice today produces a silent bug. `_checkLevelUp()` is a **mutator that only queues**: it raises `S_story.level` and pushes the level onto `_levelUpQueue`, while every benefit — the HP roll, the ASI, the Fighter feature, the gold and shield gifts — is delivered by `_showLevelUpModal`. On the skill-check **pass** path the queue is never drained, and `function _grantExplorationXp@30077` opens by clearing it, so the next step onto new ground discards the pending entry: **the player keeps the level and loses everything in it.** The *failing* branch, added later by §XP-01, does it correctly three lines away (`The attempt was not wasted.@7014`) — so failing a Ceremonia Roll can open the level-up modal and passing one cannot. Filed as **§DX-02p** (§DOC-02e), open, blast radius all 2,453 `skill_check` quests. The advice is right about the call and silent about the drain, which is the whole defect. ### V-D. Three loot channels, two random streams The report's unified-loot thesis is architecturally clean and **half-seeded**. `_rollD100Loot()` and `_rollMonsterWeaponDrop()` both draw the seeded stream (§VM-01-B), so a save determines them. The trophy channel does not: **13 of the 398 `MONSTER_DROPS` entries are arrays** — multi-drop tables picked by `function _pickDrop(table) {@7041`, which rolls `Math.random()` and writes the result to persisted inventory through `const _rawDrop = MONSTER_DROPS[S.enemy.key];@7049`. The death save is the same shape: `let d20 = Math.ceil(Math.random() * 20) + _luckMod();@25896` decides whether the character lives, off the unseeded stream, while the loot roll eleven hundred lines away was converted. Both are instances of **§DX-02m** (open, 🟠), whose stated first job is to measure how many of the file's **51** `Math.random()` sites reach persisted state; these are four of them, in the most consequential paths in the game. `check:rng` cannot see it — gate #10 proves the three mulberry32 streams agree with each other, never *which* stream a state write draws from. ### V-E. The copied half and the recalled half, in one document Two of the report's function attributions are wrong (`_onStoryVictory`, and the game-over ledger attributed to *"`storyAutoSave()` vicinity"*), and **both sit beside line numbers that are right** — the second lands three lines inside `_populateGameoverChronicle()`. Line numbers get pasted; function names get remembered. Instrument 9 in its most compact form: **the numbers a document copies are its evidence; the words it supplies around them are its claims.** ### V-F. Corpus correction — `q.xpAward` is read §DX-02n carries a sub-item stating that `q.xpAward` (53 occurrences) has **0 readers**. It has two: `// §D02 xpAward on side-type quest completion@30201` in the quest-completion loop, shipped 2026-05-26 in a commit named *"ui stuff"*, and the Warrant's Board honest-reward preview, whose own comment calls it *"the live payout"* and names `q.reward` as the dead field instead. Measured: **51 quests carry `xpAward`, 46 of them `type:'side'` — the exact shape the reader requires.** The field is live; the row's evidence line is not. → **§DX-02bw**. --- ## VI. Reachability closure (instrument 19) A symbol can resolve, execute correctly, and still be unreachable. Measured over `CELL_GRID` primacy at HEAD (416 nodes, `npm run stats`-reconciled): | Surface | Reachable? | |---|---| | Levels · Flashbang · conditions · reward formula · beads · tattoos · chronicle | ✅ global, no node gate | | Kenickie's Black Market (`CDG`) | ✅ **`CDG` is primary** in cell `21,182` of four occupants | | Luck → loot roll, death save, sheet | ✅ live | | Luck → fishing DC, catch, bait | ❌ stranded | | **`LAKE_MAGIC_DB` — the entire positive-magic economy** | ❌ **no live grant path.** The only grant site is inside `storyFishing()`, reachable only at `BOO`, which is **second in cell `2,194` behind `LYR`** — §FISH-01 / §FISH-02 | | Luck → corridor encounter rate | ⛔ deleted before the report was written | So Register 4's exclusivity design is intact as data and **dead as an economy**: monster drops are capped at base tier exactly as specified, and the fishing channel that was supposed to be the compensation cannot be entered. *The invariant shipped perfectly; the exception it defers to is behind a node nobody can stand on.* --- ## VII. The durable half — design rules that still govern These survive verification and should constrain the next combat feature: 1. **The 1.5-action round is inviolable.** Main action + one bonus. Every addition since — offhand dagger, Action Surge, spell scroll, Flashbang, conditions — is an *option inside* the bonus phase. Nothing may add a third decision per round. 2. **Automatic, announced progression.** No build menus in the flow of play. *(Noted honestly: the level-up modal does ask for an HP roll and, at seven levels, an ASI — the design broke this constraint deliberately and, per §DOC-02t, was right to.)* 3. **`XP = AC × maxHP` and `reward = ⌊0.1 × AC × HPLoss⌋`.** Any new enemy must pay on this curve or the economy tilts. Harder is self-funding; that is the promise. 4. **Trophy → d100 → weapon-quality, in that order, and no positive magic from a kill.** Every monster needs a `MONSTER_DROPS` entry — **398 of 398 at HEAD, 84 days and six new monsters later. The best-kept rule in the document.** 5. **`_magicTierAllowed(tier)` before any tiered award.** 6. **Bifurcated inventory.** Achievements are tattoos (`sell:0`, never dropped); resources are inventory. Death writes a tattoo with day, hour and place. *(§DX-02ax: the tattoos are safe by container, not by the `drop:false` flag, which has zero readers.)* 7. **Career stats never reset; run stats reset on respawn.** New telemetry is a ledger field, not a new flag. 8. **The sheet-swapper.** Post-quest NPC changes swap content on the existing node; they do not mint a new node or quest entry. 9. **The silent unlock.** A payoff surface is announced by the completion message and nothing else. 10. **No inescapable states.** The Cooperative DM rule, and the one this report's own subject matter now violates in three places: §DX-02p (a level with no rewards), §FISH-01 (an economy behind a stranded node), §AUDIT-03an (a ledger that never resets after the first death). 11. **New, from this verification:** *for every "N Xs" in a design document, re-derive N against the file — and for every "the current code does X", open the range you are about to cite.* --- ## VIII. Defects filed | Row | Grade | Substance | |---|---|---| | **§DX-02bw** (new) | 🟢 | §DX-02n's `q.xpAward` sub-item says **0 readers**; there are **2**, and 46 of the 51 carrier quests are the `type:'side'` shape the primary reader requires. Correct the row's evidence; the field is live, and the *engine's own comment* says which of the pair is dead (`q.reward`). | | **§DX-02m** | 🟠 (extended) | Four named state-writing `Math.random()` sites in the combat core: the death save (`@25896`), the initiative pair in `_storyRollInit`, the spell-scroll DC, and `_pickDrop` (13 array-valued trophy tables → persisted inventory). Total sites reconciled at **51**, matching the row. | | **§DROP-01-FU (a)** | open | Confirmed at HEAD: `LOOT_TABLE` declared, 20 entries, **0 readers**, comment still naming `_rollD100Loot()` as its caller. | | **§DX-02p** | open | Re-confirmed at HEAD: no `_showLevelUpModal` on the skill-check pass path; `_grantExplorationXp` still clears the queue. | | **§FISH-01 / §FISH-02** | open | Re-confirmed by cell primacy: `BOO` second in `2,194` behind `LYR`; no live positive-magic grant path exists. | | **§DX-02ab · §CHRON-01 · §AUDIT-03an · §DX-02ax** | open | Re-confirmed, unchanged; recorded here for cross-reference. | No new engine defect was found that an existing row does not already cover — which, for a document this old, is the most flattering thing measurement can say about the subject. --- ## IX. Conclusion Fifty-eight days after it was written, this synthesis is **94 % correct about what exists, 89 % exact about where it lives, and wrong in exactly the places where it stopped copying and started summarising.** Its tables are evidence; its prose is testimony. The single inverted claim — `_D100_TABLE` certified as weapon-free while the cited range held nine positive-magic rows — is not a transcription failure but a *category* failure: it reports an intention as an observation, because the intention was published in a report and the observation was never made. Its design half has aged extremely well. The 1.5-action round, the self-funding reward curve, the three-channel drop model, the trophy bijection, the bifurcated inventory and the silent unlock all still govern the file, and two of its stated rules — the `MONSTER_DROPS` entry per monster and the `// Layer 48:` provenance markers — have held **without a single exception** across 5,000 added lines. The tattoos still record every level-up decision, and they still outlive you. *Verified 2026-08-13 · §DOC-02bc · Synthesis Part 2 of 7 · Next: Part 3 — World & Navigation*