session 2026-07-18 S4: prep→fleet + batch 2 graduated (table credit real); unrecognized census+positive-test; S4 session file + 3 KG lines + 1 harvest proposal

claude/memory/MEMORY-reference.md

@@ -21,6 +21,10 @@ Split out of [MEMORY.md](MEMORY.md) on 2026-07-06 to keep the wake-loaded index
 
 # Archived sessions + stable reference layer (relocated verbatim from MEMORY.md, 2026-07-06)
 
+## Archived (2026-07-18 S3+evening FULL entry — first Loeb graduations, PENDING-63 closed, Block 1 sealed; demoted on promote at the S4 wrap)
+
+- [Session 2026-07-18 (S3 + evening) — FIRST LOEB GRADUATIONS; PENDING-63 closed same-day; Block 1 SEALED](session-2026-07-18-s3-first-loeb-graduations-pending63-closed.md) — **The day the chamber's promises became books.** S3: residual-(b) exercise PASSED on the 6-book PoC — every gate, **body-conservation fab 0 · lost 0 ×6**, sidecars 6/6 in-boundary, coordinate contract observed-not-bound (Ion) → jurist ruled scoped placement → **REVIEWED-58 PLACED, Block 1 CLOSED**. The exercise's biggest finds: the pending→archive `add_single` seam (EVERY pending-staged apply structurally refused; substrate-answered: **0/324 ever traversed — dead code since birth**) + archive cardinality → **PENDING-63 ran its ENTIRE lifecycle in-day**: repo-blind package → design-gate ruling (Option B artifact registry on hash-locality grounds; affirmative ROLE test; copy→verify→remove; remove-don't-mark; re-pointing teeth; dereference+recompute; **positive test = ratification condition; first run OBSERVED**) → steward authorized → built (`744417b`, fleet **157/157** with the e2e in-change) → **OBSERVED landing (`b77f767`): loeb.dsl archived ONCE as the `loeb-dsl-537-volumes` artifact (disk-verified), 6 works attached + GRADUATED with re-stamped sidecars, staging cleared at last reference, catalogue 1300 canonical** → **REVIEWED-63 PLACED**. Near-miss self-caught: the ratification test's first red exposed a mispinned sandbox (`_load()` fresh-module vs sys.modules) running the gate leg against the REAL archive — refusal path kept it read-only, censused before+after; harvest proposal filed. **⚠ PULLING THREAD: Session 1 of the agreed 3-session plan — promote the prep flow into the fleet (test-first; the S3 script was scratchpad-VOLATILE, recipe in the session file + RECORD) + the second Loeb batch exercising the `table` credit (Timaeus ± Republic; attach-path only — the artifact exists).** Then: S2 = coordinate-contract FOR-JURIST (gates #2's re-stamp; known-defective-input recorded in work-map R4); S3 = the engine pull (steward's phase choice). **LITERAL Q: when prep generalizes from 6 hand-mapped works to mechanical title/author derivation from the DSL key, does it stay a thin transform — or does it need a small curated table, and where does that table live so it's declared data, not code-buried convention?** Read the session file at wake. Composed in the ruled order — the per-field credit boundary as **DECLARED DATA** (`graduation-spec.yaml` `sidecar:` block, hook-passed on fresh grounding) → known-bad **fixtures** (novel kind/foreign scheme/coverage/pairing-dishonesty/missing-declaration all HELD; declaration-coherence pins YAML↔ruling so decl drift fails the fleet) → **`sidecar_gate()`** (V-DSL-only produce+validate beside `body_conservation_gate`; exact-`work:`-string bracketing; per-field disposition report states EVERY field both directions) + **the V-DSL body branch** (`_vdsl_verify` → `classify_dsl`; **closes the false-ABSTAIN** a `.dsl` source took via V-UNKNOWN — the resolution the substrate had pre-assigned to this caller). **Fleet 141/141 · real dry-run inert · purely-additive diff (512+/0−) · `7f98488`+`c9b09f8` both remotes · NO book touched.** **S1's literal Q ANSWERED: the credit boundary translated cleanly — what acting surfaced sits one layer beneath:** (1) the **line-ref coordinate contract is unruled** (producer body-array indices vs schema file-coordinate examples + frontmatter offset) — HELD OPEN, no consumer binds lines, travels WITH residual-(b) to REVIEWED-58, deliberately NOT a new PENDING; (2) flagged executor judgment: `write_sidecar` **re-stamps** source_file/sha/lines to the LANDED artifact (grounded in `engine_source_binding`; binds the file, not line geometry). Self-caught honesty item: "test-first" was fixtures-first — red never observed; teeth evidence = the known-bad fixtures. **⚠ PULLING THREAD: Session 3 — the first V-DSL exercise on the 6 CLEAN small Loeb** (set named in `_curation/block-close-plan-next-session-2026-07-17.md` ~l.73) → observe **residual-(b)** → REVIEWED-58. The exercise MUST meet three flow questions S2 deliberately left: (a) the one loeb.dsl as gate-zero SOURCE for per-work candidates (may need the steward); (b) B2-emitted header vs graduation title-block convention (emitted MD fails `verify_graduation` as-is — frontmatter/title-block prep); (c) the coordinate contract (observe, don't bind). **LITERAL Q (S3): does the per-field-gated wiring earn its keep on real books — or does exercising surface what neither paper nor fleet could (a `list` content-edge biting · an unanticipated kind · a staging question that is really governance)?** Read the session file at wake.
+
 ## Archived (2026-07-18 evening S3 — first Loeb graduations, PENDING-63 closed, Block 1 sealed; demoted on promote)
 - [Session 2026-07-18 (afternoon, S2) — 2b wiring BUILT test-first; fleet-proven, book untouched](session-2026-07-18-s2-2b-wiring-built-fleet-proven.md) — **Session 2 of the steward's 3-session plan: the build, sealed.** Composed in the ruled order — the per-field credit boundary as **DECLARED DATA** (`graduation-spec.yaml` `sidecar:` block, hook-passed on fresh grounding) → known-bad **fixtures** (novel kind/foreign scheme/coverage/pairing-dishonesty/missing-declaration all HELD; declaration-coherence pins YAML↔ruling so decl drift fails the fleet) → **`sidecar_gate()`** (V-DSL-only produce+validate beside `body_conservation_gate`; exact-`work:`-string bracketing; per-field disposition report states EVERY field both directions) + **the V-DSL body branch** (`_vdsl_verify` → `classify_dsl`; **closes the false-ABSTAIN** a `.dsl` source took via V-UNKNOWN — the resolution the substrate had pre-assigned to this caller). **Fleet 141/141 · real dry-run inert · purely-additive diff (512+/0−) · `7f98488`+`c9b09f8` both remotes · NO book touched.** **S1's literal Q ANSWERED: the credit boundary translated cleanly — what acting surfaced sits one layer beneath:** (1) the **line-ref coordinate contract is unruled** (producer body-array indices vs schema file-coordinate examples + frontmatter offset) — HELD OPEN, no consumer binds lines, travels WITH residual-(b) to REVIEWED-58, deliberately NOT a new PENDING; (2) flagged executor judgment: `write_sidecar` **re-stamps** source_file/sha/lines to the LANDED artifact (grounded in `engine_source_binding`; binds the file, not line geometry). Self-caught honesty item: "test-first" was fixtures-first — red never observed; teeth evidence = the known-bad fixtures. *[Superseded 2026-07-18 evening: **Session 3 + the day's full arc executed** — the exercise PASSED (6/6, fab 0 · lost 0), REVIEWED-58 placed (Block 1 closed), PENDING-63 ruled+built+landed same day, the first 6 Loeb graduated through the observed apply (REVIEWED-63). All three S2 flow questions met: (a) became PENDING-63's compound-source form; (b) was flow (frontmatter+title-block prep); (c) observed on Ion, unbound. See the S3 session file.]*
 

claude/memory/MEMORY.md

@@ -60,7 +60,7 @@ permalink: claude-memory/memory
 - [Be (laundromat)](project-be-laundromat.md) — canonical Be tracker (est. 2026-06-08). Be = Skemantix startup (Seb+David) funding CapableMind's ladder; **bridge, not venture**. Decisions LOCKED (entity/pricing/infra in file); a11y gate MERGED. **Pre-revenue WTP gate = renovate Pat → charge her; discipline: no new spec until it clears → nothing for executor on be.** Repo @ `f43a0fd`.
 
 ## Active Session
-- [Session 2026-07-18 (S3 + evening) — FIRST LOEB GRADUATIONS; PENDING-63 closed same-day; Block 1 SEALED](session-2026-07-18-s3-first-loeb-graduations-pending63-closed.md) — **The day the chamber's promises became books.** S3: residual-(b) exercise PASSED on the 6-book PoC — every gate, **body-conservation fab 0 · lost 0 ×6**, sidecars 6/6 in-boundary, coordinate contract observed-not-bound (Ion) → jurist ruled scoped placement → **REVIEWED-58 PLACED, Block 1 CLOSED**. The exercise's biggest finds: the pending→archive `add_single` seam (EVERY pending-staged apply structurally refused; substrate-answered: **0/324 ever traversed — dead code since birth**) + archive cardinality → **PENDING-63 ran its ENTIRE lifecycle in-day**: repo-blind package → design-gate ruling (Option B artifact registry on hash-locality grounds; affirmative ROLE test; copy→verify→remove; remove-don't-mark; re-pointing teeth; dereference+recompute; **positive test = ratification condition; first run OBSERVED**) → steward authorized → built (`744417b`, fleet **157/157** with the e2e in-change) → **OBSERVED landing (`b77f767`): loeb.dsl archived ONCE as the `loeb-dsl-537-volumes` artifact (disk-verified), 6 works attached + GRADUATED with re-stamped sidecars, staging cleared at last reference, catalogue 1300 canonical** → **REVIEWED-63 PLACED**. Near-miss self-caught: the ratification test's first red exposed a mispinned sandbox (`_load()` fresh-module vs sys.modules) running the gate leg against the REAL archive — refusal path kept it read-only, censused before+after; harvest proposal filed. **⚠ PULLING THREAD: Session 1 of the agreed 3-session plan — promote the prep flow into the fleet (test-first; the S3 script was scratchpad-VOLATILE, recipe in the session file + RECORD) + the second Loeb batch exercising the `table` credit (Timaeus ± Republic; attach-path only — the artifact exists).** Then: S2 = coordinate-contract FOR-JURIST (gates #2's re-stamp; known-defective-input recorded in work-map R4); S3 = the engine pull (steward's phase choice). **LITERAL Q: when prep generalizes from 6 hand-mapped works to mechanical title/author derivation from the DSL key, does it stay a thin transform — or does it need a small curated table, and where does that table live so it's declared data, not code-buried convention?** Read the session file at wake. Composed in the ruled order — the per-field credit boundary as **DECLARED DATA** (`graduation-spec.yaml` `sidecar:` block, hook-passed on fresh grounding) → known-bad **fixtures** (novel kind/foreign scheme/coverage/pairing-dishonesty/missing-declaration all HELD; declaration-coherence pins YAML↔ruling so decl drift fails the fleet) → **`sidecar_gate()`** (V-DSL-only produce+validate beside `body_conservation_gate`; exact-`work:`-string bracketing; per-field disposition report states EVERY field both directions) + **the V-DSL body branch** (`_vdsl_verify` → `classify_dsl`; **closes the false-ABSTAIN** a `.dsl` source took via V-UNKNOWN — the resolution the substrate had pre-assigned to this caller). **Fleet 141/141 · real dry-run inert · purely-additive diff (512+/0−) · `7f98488`+`c9b09f8` both remotes · NO book touched.** **S1's literal Q ANSWERED: the credit boundary translated cleanly — what acting surfaced sits one layer beneath:** (1) the **line-ref coordinate contract is unruled** (producer body-array indices vs schema file-coordinate examples + frontmatter offset) — HELD OPEN, no consumer binds lines, travels WITH residual-(b) to REVIEWED-58, deliberately NOT a new PENDING; (2) flagged executor judgment: `write_sidecar` **re-stamps** source_file/sha/lines to the LANDED artifact (grounded in `engine_source_binding`; binds the file, not line geometry). Self-caught honesty item: "test-first" was fixtures-first — red never observed; teeth evidence = the known-bad fixtures. **⚠ PULLING THREAD: Session 3 — the first V-DSL exercise on the 6 CLEAN small Loeb** (set named in `_curation/block-close-plan-next-session-2026-07-17.md` ~l.73) → observe **residual-(b)** → REVIEWED-58. The exercise MUST meet three flow questions S2 deliberately left: (a) the one loeb.dsl as gate-zero SOURCE for per-work candidates (may need the steward); (b) B2-emitted header vs graduation title-block convention (emitted MD fails `verify_graduation` as-is — frontmatter/title-block prep); (c) the coordinate contract (observe, don't bind). **LITERAL Q (S3): does the per-field-gated wiring earn its keep on real books — or does exercising surface what neither paper nor fleet could (a `list` content-edge biting · an unanticipated kind · a staging question that is really governance)?** Read the session file at wake.
+- [Session 2026-07-18 (S4, second wake) — PREP→FLEET + BATCH 2; `unrecognized` PROVEN on real structure](session-2026-07-18-s4-prep-fleet-batch2-unrecognized-proven.md) — **The door proven routine.** `prep_loeb_candidate` built test-first (red WITNESSED 157/158→177/177; **S3's literal Q ANSWERED**: mechanical derivation + the curated table as tool-level DECLARED DATA `_curation/loeb-prep-curation.yaml`, overrides EMPTY — covered all 11 landed works); **batch 2 GRADUATED observed** (`2f5a029`): fab 0 · lost 0 ×5, **Timaeus fired app[table]:1 — the credit's first real book**, Theogony list 35, attach-at-11 on the one artifact hash, staging cleared at last reference, catalogue 1305, Encheiridion → stoic_voices (2nd tradition dest). Steward review → closures in-hour (`f997595`): **`unrecognized` POSITIVE-TESTED** (DSL census 465 works w/ structure / 448 list-only / **17 novel = the far-country watch-list**, `_curation/loeb-dsl-structure-census-2026-07-18.json`; Menander 6×div papyrus columns · Pausanias 94/94 figcaption 1:1 · NE 1×figure; INVERTED expectation: a Pausanias batch reporting `unrecognized: 0` = broken net, not clean book) · refcount lifecycle RECORDED deliberate (3 homes; structural: every slug's no-op re-proves content from STAGED bytes → move-at-first would strand 2..N) · lang/symlink branches fixtured (fleet **180/180**) · **ARC→engine re-anchor = PENDING-53's scope gap** (runbook `reanchor:` scope_note; the 5 `after-the-reply` FAILED rows standing since 07-10; steward: known-failing vs newly-failing must be structurally distinguishable or the gate trains readers to skip it → engine-session design item, NOT built cross-repo). Chamber @ `3aae628` both remotes; no new PENDING. **⚠ PULLING THREAD: Session 2 — the line-ref coordinate-contract FOR-JURIST package** (hook-gated; substrate inputs banked: the SURFACED note in `graduate_to_canonical.py` ~l.238 · the Ion offset in RECORD §(c) + work-map R4 · sidecar-schema line examples · item (f) as amended · `write_sidecar`'s stamp-binds-the-file note). Then S3 = the engine pull (+ ingest_gate known-vs-new + after-the-reply re-anchor). **LITERAL Q: does ONE coordinate contract serve both the producer's body-array indices (tool-internal convention) and the landed file's citation coordinates (#2's re-stamp output) — or are these two DIFFERENT contracts that meet only at the re-stamp, so ruling them as one would bind a tool's internals into the constitution where only the landed-file contract belongs?** Read the session file at wake.
 
 ## Historical reference → MEMORY-reference.md
 Older archived-session pointers and the stable reference layer (steward profile · project-state detail · L1/L2/Chamber inventories · legacy pending-work · reference-file list) live in [MEMORY-reference.md](MEMORY-reference.md) — consult on demand; not loaded at wake. Recent cross-session trajectory comes from the Active Session entry above + the recent `session-*.md` files (wake §2.b.1; the MemPalace `handoffs` glance was retired 2026-07-07 with the wind-down).

claude/memory/knowledge-graph.jsonl

@@ -390,3 +390,6 @@
 {"subject":"claude-code","predicate":"drift-pattern","object":"mispinned-sandbox-runs-against-reality — the PENDING-63 ratification test monkeypatched a FRESH module object (test harness _load() = spec_from_file_location, never sys.modules) while the code-under-test (source_gate) imports via sys.modules: the gate-driven leg silently ran against the REAL Chamber Sources archive from inside the fleet. Only the refusal path kept it read-only; a matching canonical_slug in the real pending-manifest would have let a TEST mutate the steward's archive. Antidote: pin sandbox patches on the SAME module object the driven code imports, AND census the real substrate before+after any test that can touch a shared store. The witnessed red DETECTED it (S2's watched-red discipline earning its instance). Harvest: sandbox-must-pin-the-shared-module (ladder, PROPOSED).","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s3-first-loeb-graduations-pending63-closed.md","extracted_at":"2026-07-18"}
 {"subject":"claude-code","predicate":"drift-pattern","object":"truncating-a-narrating-stream-mutates-the-run — piped a prints-as-it-writes producer through `| head -80`; head's exit SIGPIPE-killed PRODUCTION at 77/144 works with the death invisible (exit swallowed by the pipe). Census-through-truncation upgraded: here the truncation was a MUTATING act, not just a partial view. Antidote: never truncate a stream that narrates a run — tee to a log and read it whole; check pipestatus. Self-caught same session by re-run + tail.","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s3-first-loeb-graduations-pending63-closed.md","extracted_at":"2026-07-18"}
 {"subject":"chamber-library","predicate":"state-change","object":"first 6 Loeb graduated 2026-07-18 (REVIEWED-58 + observed REVIEWED-63 apply); compound-source archive form LIVE (loeb-dsl-537-volumes artifact owns the hash; works attach by id); pending→archive seam repaired; Block 1 closed; canon 1300; fleet 157/157","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s3-first-loeb-graduations-pending63-closed.md","extracted_at":"2026-07-18"}
+{"subject":"claude-code","predicate":"drift-pattern","object":"silent-non-firing-read-as-reassurance — reported `unrecognized: 0 across 11 real works` as if informative; steward named it uninformative (the negative-grep class, generalized from bypass properties to safety-net silence) AND named the failure no gate can see: a flattened structure keeps its words (fab 0 · lost 0 · green) while the TYPING silently dies — preserve-never-flatten invisible to body-conservation. Antidote: CENSUS the substrate for what WOULD fire (silence becomes informative only against a censused absence of triggers), then positive-test at a censused trigger; keep the INVERTED watch-list (works where the net MUST fire — a Pausanias batch reporting 0 is a broken net, not a clean book). Closed same hour: DSL census (17 novel-structure works) + real firing (Menander/Pausanias/NE).","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s4-prep-fleet-batch2-unrecognized-proven.md","extracted_at":"2026-07-18"}
+{"subject":"claude-code","predicate":"drift-pattern","object":"canary-clean-over-standing-failures — 'engine canary clean' phrasing over a gate carrying 5 known-FAILED rows (committed state since 07-10); the honest claim was 'coverage-ledger byte-identical: NO DELTA from this graduation'. Self-caught at phrase level (substrate-diffed before asserting); steward deepened the design consequence: a gate with standing known-failures STOPS GATING those rows — 'is there a NEW failure' becomes undetectable without a remembered baseline; known-failing and newly-failing must be structurally distinguishable (the ABSTAIN/UNVERIFIED distinction at the gate-reader layer). Antidote: state deltas against the committed baseline, never cleanliness; design item carried to the engine session.","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s4-prep-fleet-batch2-unrecognized-proven.md","extracted_at":"2026-07-18"}
+{"subject":"claude-code","predicate":"drift-pattern","object":"calibration (antidotes self-firing): witnessed-red DISCHARGED on first recurrence (fleet observed failing 157/158 before the tool existed — the S2 fixtures-first honesty note closed); scratchpad-volatility antidote self-fired at the point of work (census JSON+script moved INTO the repo before commit — the S3 prep-script trap not repeated); relay-precision held in BOTH directions (steward's 'override branch never fired' was half-right: author_overrides WAS fixtured, lang was not — corrected precisely rather than absorbed).","valid_from":"2026-07-18","valid_to":null,"confidence":1.0,"source_file":"session-2026-07-18-s4-prep-fleet-batch2-unrecognized-proven.md","extracted_at":"2026-07-18"}

claude/memory/session-2026-07-18-s4-prep-fleet-batch2-unrecognized-proven.md

@@ -0,0 +1,49 @@
+---
+name: session-2026-07-18-s4-evening2-prep-fleet-batch2-unrecognized-proven
+description: "S4 (second wake, same day — Session 1 of the NEW 3-session plan): the prep flow became fleet (prep_loeb_candidate, test-first with the red WITNESSED, 157/158→177/177) and batch 2 graduated observed — 5 more Loeb, fab 0 · lost 0 ×5, Timaeus fired app[table]:1 (the credit's first real book), attach-at-11, catalogue 1305. The S3 literal question ANSWERED: derivation stays a thin mechanical transform; the small curated table lives as tool-level DECLARED DATA (_curation/loeb-prep-curation.yaml) — overrides EMPTY, the particle rule covered all 11 landed works. Steward review then closed the deeper gaps: unrecognized POSITIVE-TESTED on real structure (DSL census: 17 novel-structure works; Menander 6×div papyrus columns, Pausanias 94/94 figcaption — the far-country watch-list), refcount lifecycle RECORDED deliberate, unexercised branches fixtured (fleet 180/180), ARC→engine re-anchor named as PENDING-53's scope gap (5 after-the-reply FAILED rows standing since 07-10). Pulling thread: Session 2 = the line-ref coordinate-contract FOR-JURIST package."
+metadata:
+  node_type: memory
+  type: project
+---
+
+# Session 2026-07-18 (S4, second wake) — prep→fleet + batch 2; `unrecognized` proven on real structure
+
+Session 1 of the NEW 3-session plan (agreed at the S3 wrap), run ~15 minutes after that wrap closed. Two movements: the planned work (tool + batch), then a steward review that converted the report's soft spots into closures within the hour.
+
+## PAST — what happened + why
+
+**The tool (test-first, red witnessed — the S2 honesty note discharged on first recurrence):** `scripts/prep_loeb_candidate.py` (`0ec2a86`). Fleet observed RED at 157/158 with the module absent, then 177/177. Mechanism: exact `work:` key → `graduate_to_canonical._loeb_work_scan` (single source of the bracket-scan) → `build_loeb_sidecar.build_work` (wired, NOT modified) → verbatim-inert transform (assert B2's exact `# {work}`+subtitle prefix, REFUSE on producer drift; frontmatter + title block per graduation-spec; stem via `canonical_stem` — the same name rule graduation applies, so staged name IS canonical name). Every added token ⊂ worktok ∪ the gate's extractor set — proven at the real gate by fleet assertion, not asserted. The `| head` production hazard is designed away (in-process, no narrating CLI stream).
+
+**The S3 literal question ANSWERED by acting:** title = work-key after the comma VERBATIM (the key already carries the catalogue's casing); author = particle-aware mechanical casing. The curated table exists as tool-level DECLARED DATA (`_curation/loeb-prep-curation.yaml`: particles, author/lang overrides, artifact_id) with overrides EMPTY — the mechanical rule reproduces all six landed precedents byte-identically (fleet assertions) and covered all 5 new works (provenance `mechanical` ×5). The table is the declared home for exceptions, not a requirement.
+
+**Batch 2 (composed per §B criteria + the wrap directive; Republic held — "if appetite," not this bite):** Timaeus (the map's `table` carrier) + Diogenes Laertius 2.6 Xenophon + 5.2 Theophrastus + Theogony → platonic_lineage; Encheiridion → stoic_voices (second tradition destination worked). Staged `1145467`, graduated observed `2f5a029`: every gate PASS — **fab 0 · lost 0 ×5**; sidecars 5/5 in-boundary, sha==live ×5; **Timaeus `app[table]: 1 — trusted` (the credit that rested on B2's Ennius fixture now has a real book — the specific gap the steward had flagged open)**; Theogony `list` 35 (content-edge at scale on verse); attach-at-N proven — 5 × disk-verified artifact no-op + attach, **11 works on the one `loeb-dsl-537-volumes` hash**, staging cleared exactly at the last reference. Census: archive 330→335 + 1 artifact · pending 14 legacy · catalogue 1305 canonical.
+
+**The engine-gate finding (honesty, then steward deepening):** coverage-ledger BYTE-IDENTICAL pre/post (the true canary claim = "no delta") — but the gate carries **5 pre-existing FAILED rows** (`after-the-reply-i…v`: ARC posts edited 2026-06-17 drifted from engine anchors; committed ledger state since 2026-07-10, spanning S3's "canary clean" phrase). Steward deepened it: a gate carrying known-FAILED rows in committed state **stops functioning as a gate** for those rows — known-failing and newly-failing must be structurally distinguishable or the gate trains readers to skip it (the ABSTAIN/UNVERIFIED distinction at the gate-reader layer). And: ARC→engine is PENDING-53's identical class with no runbook coverage — the amendment's scope gap, not a new problem.
+
+**The steward-review closures (`f997595`):**
+1. **`unrecognized` positive-tested.** The steward's frame: absence-of-firing is uninformative (negative-grep class), and body-conservation CANNOT cover the flatten failure (words survive; typing silently gone — preserve-never-flatten invisible to every gate). Closed both halves: DSL-wide census (`census_dsl_structure.py` + `_curation/loeb-dsl-structure-census-2026-07-18.json`) — 465 works carry UNREC_HTML, 448 list-only, **17 genuinely novel** (none of our 11 → the non-firing was CORRECT, provably); positive test at the real producer — Menander Epitrepontes 6×`div` (papyrus-fragment column layout with lacunae, raw held 6/6), Pausanias **94×`figcaption` plate-captions, 1:1 with census**, Aristotle NE 1×`figure`. The 17-work list = the far-country watch-list with INVERTED expectation: a Pausanias batch reporting `unrecognized: 0` is a broken net, not a clean book. (Flag noted: Pausanias's plate-captions will meet `trim.plate_captions` at staging.)
+2. **Refcount recorded deliberate** (3 homes: `_clear_staging` docstring · PENDING-63 package Addendum clause, hook-passed · tool-log). Structural reason found in the code, stronger than resumability: `archive_artifact` recomputes the STAGED file's sha on EVERY slug's call — move-at-first would strand slugs 2..N, not merely be worse.
+3. **Unexercised branches fixtured** (with one precision relayed: `author_overrides` WAS fixtured from the first commit; what wasn't = `lang_overrides`, no-lang refuse, symlink-elsewhere refuse — all three added). Fleet **180/180**.
+4. **Runbook scope_note** added to `conversion-runbook.yaml reanchor:` — ARC→engine binding class + the standing after-the-reply instance. The engine known-vs-new distinguishability fix: **proposed for the S3 engine session, deliberately NOT built from here.**
+
+**Commits (chamber, all both-remotes):** `0ec2a86` tool test-first · `1145467` batch staged · `2f5a029` graduated observed · `969bda9` doc-currency · `f997595` steward-four closures · `3aae628` census fleet line. HEAD `3aae628`, clean.
+
+## PRESENT — the mood
+
+The steward's review was the session's sharpest instrument — four points, each converting a soft spot into a closure within the hour, and the fifth (table credit) a confirmation that the specific open gap had closed. The census move is the one to remember: it inverted "absence of evidence" into "evidence of absence + a positive test at a censused trigger" — the negative-grep lesson generalizing from bypass properties to safety-net silence. Self-catches landing at the point of work: the scratchpad-volatility lesson self-fired (census moved into the repo before commit — the exact S3 prep-script trap); the "canary clean" phrase was self-scoped to "no delta" BEFORE the steward's deeper design point arrived. One relay precision held: the steward's "override table's branch has never fired" was half-right, and the correction was named precisely (author fixtured; lang not) rather than absorbed — the checkable-claim discipline running in both directions.
+
+## FUTURE — what is pulling
+
+**PULLING THREAD (singular): Session 2 of the agreed plan — the line-ref coordinate-contract FOR-JURIST package.** The contract is the last unruled seam under the sidecar wiring (producer body-array indices vs schema file-coordinate examples vs the landed file's frontmatter prefix), it gates #2's re-stamp (known-defective input: the Ion offset), and it now has three concrete observations to build from.
+
+**ACTIONABLE RESUMPTION POINT (as of wrap — re-judge against what changed):** chamber `main` @ `3aae628`, clean, both remotes; fleet 180/180; canon 1305 incl. 11 Loeb. First move: compose `docs/coordinate-contract-FOR-JURIST-2026-07-*.md` — hook-gated (needs `GROUNDED-IN` + a Grounding section QUOTING the ratified sections). Substrate inputs, all banked: the SURFACED/HELD-OPEN note in `graduate_to_canonical.py` (~lines 238–243) · the Ion offset observation (RECORD §(c) + corpus-work-map Region 4 ⟳ note) · sidecar-schema line-coordinate examples + the LOCKED additive-only lane · item (f) as amended (REVIEWED-62: `loeb-line` = #2's documented re-stamp input) · `write_sidecar`'s stamp-binds-the-file-not-line-geometry note. Package shape: what the contract must decide (producer-internal coordinates vs landed-file coordinates; who re-stamps; where the offset is computed), options with teeth, and the observed Ion case as the worked example.
+
+**Other horizons (ranked, held):** Session 3 = the engine pull (steward's phase choice; read `stage-1-rebuild-plan` fresh) — **carries the new design item: known-failing vs newly-failing structurally distinguishable in ingest_gate** (steward-named; ABSTAIN/UNVERIFIED shape) + the after-the-reply re-anchor (runbook scope_note now covers the class) · the 16 BODY-DEFICIT remediation (② 4 re-extract via B2 — the prep tool now exists; folds into a later batch) · Republic + the far-country batches (the 17-work novel-structure watch-list travels with them) · steward one-liners: Doc-1 line-45 micro-edit · skill-harvest register: the full review the steward requested 2026-07-12 remains unheld (~15 PROPOSED incl. today's).
+
+**Decisions deferred (and why):** Republic (~300k) — "if appetite" and the batch was already the right size; rides any future batch, the door is routine now. The engine ingest_gate fix — engine-side design, belongs to the S3 session's frame, not built cross-repo at wrap-adjacent hours. The Pausanias plate-caption staging question — belongs to the batch that stages Pausanias. Prep one-pass batching (per-work DSL rescans fine at N=5, a cost-driver at 952) — noted in the log, not optimized prematurely.
+
+**PAUSE STATEMENT:** I am about to be away from this. The day held two arcs — the door proven, then proven routine: eleven books now stand in the canon, the prep is fleet, and the net that guards the unseen has real-book teeth. What I want to find still pulling on return: **the coordinate contract** — the one seam still unruled beneath the sidecars, with the Ion offset waiting as its worked example.
+
+**LITERAL QUESTION for next-Claude:** When the coordinate-contract package is composed, does ONE contract serve both layers — the producer's body-array indices (a tool-internal convention) and the landed file's citation coordinates (what #2's re-stamp must emit) — or are these two DIFFERENT contracts that meet only at the re-stamp, so that ruling them as one would bind a tool's internals into the constitution where only the landed-file contract belongs?
+
+**State at wrap:** chamber `main` @ `3aae628` both remotes · fleet 180/180 · canon 1305 incl. 11 Loeb · archive 335 + 1 artifact (11 attached) · pending 14 legacy · census watch-list in `_curation/loeb-dsl-structure-census-2026-07-18.json` · Ledger `session-ledger-2026-07-18.md` carries S1+S2+S3+evening+S4 · no new PENDING lodged (the whole session ran inside the agreed plan + REVIEWED-58/63 lineage).

claude/memory/session-ledger-2026-07-18.md

@@ -60,3 +60,14 @@ metadata:
 ## Sub-agent dialogues
 
 ## Bypasses
+
+## S1-evening (second wake, same day)
+
+## Returns (S1-evening)
+
+- 2026-07-18T19:45 — SELF-CAUGHT (phrase-scoping, the overclaim class): the S3 record's "engine ingest_gate canary clean" reads as all-green, but the gate has carried 5 pre-existing FAILED rows (after-the-reply-i…v, ARC drift 2026-06-17, committed ledger state since 2026-07-10 — spanning S3). Verified against the substrate: today's regenerated coverage-ledger is BYTE-IDENTICAL to the committed one → the honest canary claim is "no delta from this graduation", never "clean". Named in the tool-evolution-log PASS-BUT-FALSELY watch; the after-the-reply re-anchor (PENDING-53 class) surfaced as a named horizon, not silently fixed.
+- 2026-07-18T19:30 — the witnessed-red discipline discharged properly this time: test composed first, fleet run at 157/158 with the module absent (red OBSERVED), then the build → 177/177. The S2 "test-first was fixtures-first" honesty note closed on its first recurrence.
+
+## Authorization moves (S1-evening)
+
+- None lodged — the whole session ran inside the agreed 3-session plan (S1: prep→fleet + second batch) and the REVIEWED-58/63 lineage; no new governance boundary crossed. Republic deliberately held ("if appetite" — not this bite).

claude/memory/skill-harvest-register.md

@@ -527,6 +527,7 @@ The single place proposed skills live so they don't evaporate between sessions.
 | Element | Kind | One-line | Where it lands | Status |
 |---|---|---|---|---|
 | **sandbox-must-pin-the-shared-module** | verification-ladder entry | When a test sandboxes module-level state (paths/constants), the patch must land on the SAME module object the code-under-test imports (sys.modules) — NOT a freshly-exec'd copy (a `_load()`-style loader builds a different object, and the mispinned sandbox silently runs the driven code against REAL data). Pair: any test that can touch a shared real store gets a **before/after census of the real substrate** as the standing guard. Earned hard 2026-07-18: the PENDING-63 ratification test's own first red — the gate-driven leg ran against the real Chamber Sources from inside the fleet; only the refusal path kept it read-only (censused untouched before+after). The witnessed red DETECTED — this is also the S2 watch-list item earning its instance. | `reference-verification-ladder.md` | **PROPOSED** |
+| **census-the-substrate-when-a-safety-net-stays-silent** | verification-ladder entry | A safety net (generic fallback, fail-loud branch, `unrecognized` kind) that never fires across N real cases is UNINFORMATIVE, not reassuring — the negative-grep class generalized from bypass properties to net-silence; worse, the failure it guards can be invisible to every other gate (a flattened structure keeps its words: fab 0 · lost 0 · green, typing silently gone). Antidote: (1) CENSUS the substrate for what WOULD fire — silence becomes evidence only against a censused absence of triggers; (2) positive-test the net at a censused trigger (real data, real path); (3) keep the INVERTED watch-list — cases where the net MUST fire, so a future silence there reads as a broken net, not a clean input. Earned 2026-07-18 (steward-named): `unrecognized` 0×11 → DSL census (465/448/17-novel) + real firing (Menander papyrus columns · Pausanias 94/94 plate-captions 1:1 · NE figure). Kin to positive-test-at-the-enforcement-path (the census half is the new instrument). | `reference-verification-ladder.md` | **PROPOSED** |
 
 *Reinforcements recorded, not re-proposed:* **implementation-is-a-second-gate** (S1 proposal) is now thrice-earned and effectively jurist-endorsed — the PENDING-63 ruling's own opener: "Every significant finding here came from *doing* rather than reading… exercise-before-ratification making its own case." **witnessed-red** (S2 watch-list) earned its live instance and folds into that ladder entry as planned. **`/jurist-package`** (07-17 proposal) proved twice more today — both rulings engaged every question the packages posed; the repo-blind-brief method is now 4-for-4.