docs(ledger): reconcile the task ledger and the docs that drifted from it

[stellarfeline]

All 34 pending / in_progress ledger items audited against origin/main at
0a40db8. Every verdict below is a measurement, not a grep hit: the
workspace was built first (cargo build --workspace → Finished) and
cargo test --workspace exits 0, so no test count here was read off a tree
that did not compile.

Verdicts

#	Subject	Verdict	Evidence (measured)
6	Owner-confirm batch (9 items)	OPEN — 6 of 9 answered	(3) mark merged #286; (6) island released v1.1.0; (7) required checks merged #304 — check-required-contexts.py reports 12 jobs / 12 contexts; (8) #298+#299 merged; (9) worktree rule is CLAUDE.md:279; (1) withdrawn by #36's re-scope. Still unanswered: (2) spec-0027 AC6, (4) layout staleness — fixed in this PR, (5) souls slate: bossbar/boss_phase/fog_gate/cooldown appear 0 times in crates/dsl/src/stages.rs and in no spec; only currency+vendor were approved (spec-0032)
7	Bell remake → artifact gate	OPEN (blockers real, one premise STALE)	bell/mod.rs zone table: Z3 and Z7 both not programmed; causeway still a terminus, no tower_stair. STALE: "REMAKE.md does not exist on disk anywhere (verified)" — it does; see below
8	Cherry valley #176/#177	OPEN + STALE PREMISE	PR #260 is CONFLICTING / DIRTY, 19 commits behind origin/main, not "MERGEABLE/CLEAN"; "10/10 green" is now 12 contexts. #259's base is worker/spec-0026-foundation (CLEAN there), not main. Chain unmoved: engine #260/#259 open, content #21/#22/#26 open
10	spec-0027 phase 2/3	OPEN (phase 2/2b landed)	Obligation A: crates/grammar/tests/staging.rs:1316 still carries the §4 palette-role test-local mirror; no §4 diagnostic exists. Obligation B: reachable_with_fall still test-local in grammar/tests/support; crates/grammar/Cargo.toml still has no dependency on the compiler — and PR #316 added compiler::traversal, a third model of "can this body get there"
12	DW0461 branch-aware replay	STALE PREMISE	continuity::NpcWhere::Indeterminate excludes any NPC whose lifecycle is branch-driven (flag-gated effect, dialogue option, env trigger, reaction bundle); check_placement_position only matches At/Offstage, so a branch NPC never reaches DW0461 — DW0462/DW0483 own it. The "#186 open gap" half is closed too: emit::branch_paths emits branch-path-<slug>.json through the same critical_path_json as the exported path (PR #276)
18	DW0310 trigger-fired wave — bind or forbid	OPEN	plan.rs:1029 wave_area step 2 is unchanged: still campaign.world.content.areas.len() == 1. No anchor→area resolution, no strike-npc refusal code. Half landed: DW0497 makes a residual dangle a hard build error
23	cast::station concurrency edge	OPEN	Read cast.rs:240-269: still iterates quest_dag_order, breaks at upto_quest, and consumes the whole-path begun set — both divergences (a) and (b) stand. No diagnostic forbids two concurrently-activatable quests casting one NPC; no red fixture exists
26	Cast-arrival timing proof	OPEN (blocked on planner design, as filed)	SPRINT_TICKS_PER_BLOCK = 4 exists in nav.rs, but no proof compares a walk driver's tick cost against a player route; grep for an arrival-race diagnostic returns nothing
28	move-actor ma_tick per-leg latch	DONE	PR #303 merged as 319b4a9. cargo test -p delvec --test move_supersede → 7 passed, 0 failed, including later_move_actor_supersedes_the_running_leg. (The ledger entry's tail is a corrupted task-update payload — worth cleaning when you close it)
33	min_y into critical-path.json	OPEN	harness/src/teardown.ts:35 still reads "The harness has no wired source for a delve's min_y today", and scriptedTeardownThreshold still falls back to -100. No min_y in the compiler's critical-path export
34	hollow-vigil does not build	OPEN + STALE PREMISE	Reproduced: delvec validate → exit 1, exactly 2× DW0331 (no DW0205 — it is masked, as filed). STALE: the "residual gap (own task)" — a pipeline-repo job compiling the pinned content's campaigns — is closed: PR #322's campaign builds job is a required context, and .github/campaign-build-exclusions.toml INVERTS the assertion for hollow-vigil. tools/build-every-campaign.py run here: discovered 2, built green 1, known-red 1 as recorded. Still open: the content-repo campaign-build.yml is not on content main (only prefab-audit, release, workflow-lint)
36	#192 re-scope: /trigger out-of-order	OPEN — both findings live	Finding 1 confirmed by reading emit.rs:3160: the completion line is execute if score #party dw.qa_<q> matches 1 unless score #party dw.o_<obj> matches 1 … — quest-active + not-already-complete only; the objective's own after[] is never consulted. (Option-level requires_flags/forbids_flags/requires_state do get return fail guards, so the hole is an ungated option completing a prerequisite-bearing objective.) Finding 2 confirmed: emit.rs:3148 emits weather {token} with no duration; WorldWeather::token() is clear/rain/thunder
42	Local playtest ≠ shipped pack auto-install	OPEN	tools/playtest-server.sh:137-145 prints only the pack path / copy note; nothing says the shipped path is unexercised. resource-pack + resource-pack-sha1 in server.properties are written only by the content repo's release.yml. tools.md's row does not say so either. Its blocking dependency (the build-performance work) landed at #313
46	traversal override — "spider sheep"	STALE PREMISE	The task is filed as an OWNER CALL awaiting a ruling. CLAUDE.md:236-239 records that it was already ruled in an earlier session, and names re-asking it as a rebuke cause. What is open is the spec + implementation, not the decision — and the ruling's content is written down nowhere in either repo (only its existence is), which is itself a finding against "docs are the only persistent memory"
47	resample step-up levitation	OPEN	nav.rs:1727 documents the L geometrically ("rise in place, then cross") and why. The player-visible cost framing is absent, and the skill's shot-review step (SKILL.md:969) says nothing about step-ups. Its entire remaining deliverable is a doc change
48	crates.io + release binaries	DONE (part 2 half-executed)	PR #318 merged 922cfb6: both crates carry description/license/repository, delvec is the package name, no publish = false on either, engine-release.yml is the only publisher behind an approval environment. gh release view v1.1.0 → 5 target archives + SHA256SUMS. But gh release view v1.0.0 → ZERO assets, although the tag was force-moved onto 922cfb6 exactly as ADR-0017 §5 says — so the ADR's stated reason for the retag ("an empty shelf forever") is still unfulfilled. workflow_dispatch with the tag input re-fills it
50	shortcuts[] wrong-side hint	DONE — in a better shape than filed	PR #324 (95091d3). No bespoke hint field was added: compiler::wrongside derives the sealed side from the gate slab's thin axis, ws_arm_fns summons one 1.02f interaction per doorway cell on the sealed side only, and an author's ordinary strike/use trigger on the gate rides them — the general EnvTrigger mechanism reaching an object it previously could not. DW0425 withholds rather than guesses; tests/shortcut_wrong_side.rs:332 asserts the code
51	Widened inventory bypasses the fence	OPEN — doc half landed	PR #339 recorded the measurement on l10n::inventory and explicitly left the tier decision to the owner. Neither deliverable is done: l10n.rs:511 still says the walk "consults the campaign document and never its dsl_version", and grep -n widen CLAUDE.md returns nothing
52	delvec fmt	OPEN	delvec --help lists validate analyze build l10n-inventory schema snapshot blocking-chart edit calibrate — no fmt
53	Byte baseline for released campaigns	OPEN	The recorded baseline hash cf882e2e… appears nowhere in the tree; tools/build-every-campaign.py contains no hashlib/sha256/baseline. The campaign builds gate proves "still builds", exactly as the task says
54	Unbound-fact sweep (3 audits)	OPEN — final obligation not done	Instance + general fixes landed (content #39; engine #342, #345, #344, #320). But the closing obligation — "record the fourth vacuity shape in CLAUDE.md alongside unbound/unfenced/unemitted" — is not done: CLAUDE.md:207-215 still lists exactly three. audit-content.md / audit-engine.md / audit-gates.md are not in the scratchpad
61	Lift authored entirely in campaign JSON	DONE	PR #356 merged 0a40db8. Not one file under any crate's src/ is touched; tests/v10_lift.rs reads the owner's timing table back off the emission, and a schema-derived test asserts no DSL surface is named after a lift. spec-0031 AC9 closes
64	compose drops four harness env vars	OPEN (name detail corrected)	Set difference measured: harness reads 15 names, compose.yaml forwards 11. Missing: DELVEWRIGHT_ACTOR_FLOOR, DELVEWRIGHT_CUTSCENE_GRACE_MS, DELVEWRIGHT_ENTITY_SETTLE_TIMEOUT_MS, DELVEWRIGHT_RETRY_ON_DEATH — all four carry the DELVEWRIGHT_ prefix the ledger entry omits. No env-forwarding gate exists
65	Untiered set-piece fight compiles green	OPEN — doc half landed	stages.rs:2101 and :3163 are both tier: Option<EncounterTier>; nothing requires one. playtest-methodology.md:104 now states this correctly ("no diagnostic demands a tier"), so the doc is fixed and the engine gap is not
66	DW codes outside compiler.md bound by nothing	OPEN (count stale) + half STALE	Re-measured: 37 distinct codes across the other six reference docs (grammar 1, i18n 9, playtest 16, skill-workflow 1, tools 15), not 42. check-dw-codes.py still binds DOC_PATH = docs/reference/compiler.md alone. Exactly one code needs the PENDING concept: DW0364, absent from source and from compiler.md. STALE half: the motivating instance is fixed — mecha_crosscheck.py is in tools.md (line 274). No check-tool-inventory.py exists
68	Bot-tier debt: on_death + lethal-volume death	OPEN	harness/src/ lethal is staged.some(g => g.crush) — timed gates only. No lethal-volume and no on_death assertion anywhere in the harness. The crush-gate precedent it should copy is in executor.ts:232-609
69	Self-certifying byte-identity baseline	OPEN	worktree-bootstrap.md:112 names the invariant ("a baseline hash manifest records the git SHA and delvec --version… and the comparison asserts them") under a heading that literally reads "The stronger fix, which is not this convention". Nothing in tools/ or validation/ builds it
70	1.21.11 rejects legacy camelCase gamerules	OPEN — both sites still live	crates/admit/src/gallery.rs:171-174 still emits doDaylightCycle/doWeatherCycle/doMobSpawning/doImmediateRespawn; tools/spike-jump-arc/measure.mjs:186 still emits gamerule fallDamage false. The compiler is fine — tests/emit.rs:774 asserts the new names
71	Respawn invulnerability window	DONE	Recorded in spec-0031:432, spec-0032:166, compiler.md:3543, docs/notes/death-and-teleport-spike.md:210, raw at observations.json:331. And it reached emission: emit.rs:7377 explains that execute store success … run damage is inert and the lethal-volume guard reads the outcome (#leth_hp); tests/v10_lethal_volume.rs:198-210 asserts both the outcome guard and the absence of store success
72	Share threshold declassifies open findings	OPEN	check-capability-ownership.py:240 is still MODIFIER_MIN_SHARE = 1/3 with no absolute-count term and no "once ledgered, stays ledgered". Ran it: check D examines 112 pairs, 6 matched — while capability-ownership-audit.md:243 says 7. The 7th (TimedGateDisarm.TrapDisarm) is a MODIFIER_HOLES key the checker did not count as matched, so the audit table and the tool disagree today
73	Two affordance authorities	OPEN	eclipse.rs:311 states the defect in its own doc comment ("there are TWO affordance authorities in the compiler"). No import crosses between affordance.rs and eclipse.rs; no test compares their membership. v10_economy.rs:454 pins the shop instance, not the general form
74	Teleport moves a stake out from under its ledger	OPEN	teleport.rs has no stake handling; check_bound_affordances reads eclipse::affordances, and v10_economy.rs:450 asserts a stake is deliberately not in it. Neither candidate fix is taken
75	A local PackTest run can be a strict subset	OPEN	validation/packtest-run.sh asserts no test count at all and never compares against what the compiler emitted — grep for count/template in it returns only prose
76	A spec's Status is bound to nothing	OPEN — instance in flight, general form open	PR #357 owns spec-0029. Swept all 30 specs: spec-0027 Proposed, spec-0031 Draft, spec-0032 Draft, spec-0028 Proposed all read as unbuilt over shipped work. Fixed in this PR by adding a measured Implemented line beside (never replacing) Status. No check binds either

The seven REMAKE §7 prerequisites, measured

campaigns/the-drowned-bell/REMAKE.md §7 names them by pre-renumbering task ids.
Measured by capability, not by number:

REMAKE §7	Capability	State
1 · #180	Undefeated rest/death re-seat at full HP	EXISTS. waves[].respawns_on_rest (stages.rs:2069); and per the 2026-08-05 owner ruling a billed elite/boss still standing is deleted and re-seated at full count/health by a rest (stages.rs:2094-2099). DW0499 forbids boss + respawns_on_rest
2 · #179	Declared-subset drops	EXISTS. MobDrop = Slot{slot} | Item{item,name} (DSL v0.9, stages.rs:2333); undeclared slots keep drop chance 0.0; DW0490 refuses a slot the entity's own equipment does not fill. Note the shape is a declared subset, not a raw per-slot drop_chance knob as §5 words it — same guarantee
3 · #116	Kit potion contents	EXISTS. KitItem.contents: Option<PotionContents> (DSL v0.8, stages.rs:892); DW0486 restricts it to potion-bearing items, DW0487 catches a potion with no contents
4 · G1	Observability: standable cell outside the hazard span with line of sight	EXISTS, already at the tier §7 asks for. DW0388 (nav.rs:97), HAZARD_STANDOFF derived as one second of sprint. §7 asks it be "error tier for the remake" — nav.rs:3647 hazard_tier(souls, …) already converts every finding to a build error (exit 3) for a souls campaign, and nav.rs:3637 defines that as "a campaign that places a bonfire". The remake declares six. Nothing to promote
5 · G2	Timed-gate disarm	EXISTS. TimedGate.disarm: Option<TimedGateDisarm> (stages.rs:1793), PR #275
6 · #163	Grammar back end	EXISTS, with two named gaps. crates/grammar with 6 of 8 zone programs. Z3 and Z7 are the open engine blockers: causeway is a terminus with no exit lane past its guard post (wants a berm_gate parameter), and nothing in the vocabulary ascends — drop_shaft/dumbwaiter are one-way down by gate, boulder_stair is flat, so a bell tower cannot be climbed (wants tower_stair). Plus counterweight_lift, which today's IR cannot express
7 · #181	Narrate/skin/prompt/wave-guidance batch	CONTENT, unstarted. campaigns/the-drowned-bell/ on content main holds only REMAKE.md, media/, out/, out-zh/ — no campaign source at all, which is why the campaign builds gate discovers 2 campaigns and not 4

So the remake's open ENGINE blockers are exactly two shapes, both in the grammar
vocabulary: an exit past a terminus (Z3) and any ascending route (Z7). Everything
else §7 lists is built.

Document corrections (each with the measurement behind it)

Document	Was	Measured
CLAUDE.md layout	crates/ … dsl / compiler / orchestrator / admit / schem / render	ls crates/ → 7 members; grammar was missing
CLAUDE.md layout	docs/reference/ listing six files	ls docs/reference/ → 7; worktree-bootstrap.md was missing
CLAUDE.md layout	prefabs/ # .nbt library + metadata (git-lfs)	find prefabs -name '*.nbt' | wc -l → 0; *.json → 0. 36 .nbt are in the content repo. prefabs/README.md already said so; the constitution did not
docs/reference/grammar.md §5c	REMAKE §n cited with no path	The file is at campaigns/the-drowned-bell/REMAKE.md, content main 795ef92. §3 = zones, §4 = lettered catalogue, §7 = prerequisites — all three resolve
spec-0027	Status: Proposed	AC1/2/4/5 shipped (#266/#277/#286); AC3 not built — the §4 mirror at staging.rs:1316 still stands in
spec-0028	Status: Proposed	refimg.py shipped (#334/#338); AC3 nowhere — similarity occurs only in the spec itself; no test of any kind for the tool
spec-0031	Status: Draft	All five primitives + AC9 merged; runtime death proof absent (task #68)
spec-0032	Status: Draft	Compile-time shipped (#355); AC9's bot-tier loop absent
docs/notes/capability-ownership-audit.md	"Binds today": A 11, D 7, E 10	Ran tools/check-capability-ownership.py on main: A 13, D 6, E 11. D's "7" counts a STRUCTURAL_TWINS tuple as a MODIFIER_HOLES key — the dict has six. A hand-copied binding count in the table whose whole purpose is stating binding counts, so the note now tells the reader to re-run the tool rather than quote the row

Status lines are not touched — they record approval. Each spec gains a
separate Implemented line that names its evidence and, deliberately, what is
still missing. spec-0029 is untouched: PR #357 owns that file.

Findings, reported rather than edited (code is wrong, not the doc)

1. DW0425's message names a field that does not exist. emit.rs:6165 tells
   the author "shortcut X declares an on_wrong_side answer" — on_wrong_side
   appears in no DSL schema (grep -rn on_wrong_side crates/dsl/src/ is
   empty). It is the vestige of an abandoned design; wrongside.rs:2's module doc
   repeats it. Worse, check_shortcut_sides raises DW0425 for every
   shortcut with an underivable side, whether or not any answer is authored — so
   the message is false for the common case.
2. v1.0.0's shelf is still empty although ADR-0017 §5's retag was performed.
   workflow_dispatch on engine-release.yml with tag: v1.0.0 fills it.
3. Three of the capability ledger's five binding counts had drifted below the
   build (A, D, E) — corrected in this PR, since the numbers are the document's
   only content.
4. The traversal ruling exists but its content is unrecorded — CLAUDE.md
   asserts it was ruled and nothing anywhere says what was ruled.

What CI proves

Doc-only; no file under any crate's src/ is touched. check-doc-dupes,
check-dw-codes, check-reference-versions, check-required-contexts,
check-capability-ownership, check-skill-version and check-storybook-version
all pass locally, on a tree that built first.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AjQ5p1Kv5MrkGPumi7yXWL