dotfiles/claude/memory/session-2026-08-08-voice-is-the-convocation-key.md

name, description, metadata

name	description	metadata
session-2026-08-08-voice-is-the-convocation-key	Disposition (vi) decided, placed (REVIEWED-97) and committed — but the ruling that landed is not the one any single party drafted: the jurist's two-value collapse would have made `traditional` one convocable speaker spanning Old Norse verse and Sanskrit epic, because `voice:` is the retrieval key, not a classification field. Corrections ran in all three directions in one day. PULLING THREAD: the `quotation-in` × `translation-of` composition package — (vi) is decided but INAPPLICABLE until it is ruled, and it gates the Mauss remediation and V2's fr gold behind it.	node_type type originSessionId modified memory project 7d08dad4-626a-484c-870b-8f1a9674db7a 2026-08-08T10:50:21.626Z

Session 2026-08-08 — voice: is the convocation key, not a classification field

Woke to (vi) 10.7 h after the previous wrap. Drafted the disposition, and the draft survived contact with nobody — steward, jurist and substrate each moved it.

PAST — what moved, and why

Grounded first, then read the substrate rather than its descriptions. D-4, §4, §4.1 and REVIEWED-96 verbatim; then all twelve blocks and Mauss's own lead-ins from the source file, not from the sidecar titles. That ordering is what produced every finding below — three of the day's errors were labels read in place of text.

The disposition, derived from consumers not from tidiness. chunker.py:136 is sec.get("voice", catalog.get("voice")), so omission resolves to the host — option C ("omit voice, let the relation carry it") is not weaker, it is refuted: it restores the exact defect under correction. Explicit null differs (returns None) but is unreachable by any convocation, since SQL = never matches NULL, defeating the D-4 mechanism Q1 was authorized on. Control 4/4.

Surah LXIV, not CXIV. The sidecar says "Surah CXIV vv. 15-18"; Mauss L1508 says "la fameuse Sourate LXIV, « déception mutuelle »". at-Taghābun, not an-Nās. The label had already travelled into REVIEWED-96 (Q4), PENDING-113 and session memory. Consequence was not cosmetic: the jurist's worked provenance note was built on the "Say" (قُل) formula, which opens an-Nās and appears nowhere in the passage Mauss quotes. Evidence withdrawn, not transplanted; the general ground untouched.

The steward's correction on language found a third defect in 118f411. I wrote that the corpus text "is French in every case" as though that carried inference — but French is the original language of the Essai. Being corrected sent me back to [^101] ("M. Maurice Cahen a bien voulu faire pour nous cette traduction" — commissioned for this book), and from there into the fence, where L850 reads "M. Cahen nous signale aussi la strophe 145 :" — Mauss's own prose, fenced inside the Havámál block. 118f411 also withdrew a line of Mauss from citability inside his own book. My purity check had passed it on "0 prose-shaped lines (>150 chars)" — length as a proxy for authorship, which a nine-word connective walks under. There is a curator precedent (L1043's "Et ainsi de suite :" is declared, deliberately) so it is a real choice, not an oversight to repair — but one is declared and one was silent.

The jurist exchange, and the correction that ran upward. The jurist's structure — a small enforceable primitive plus a place for the fuller account — is right, and the argument that a scriptural bucket repeats the flattening one level up (śruti ≠ revelation-through-a-Prophet ≠ living-Guru) is right. But the first draft put the two category values in voice:, and voice: is the convocation key (retrieve.py:216 filters d.voice = ?). Measured: glidden spans 5 sources, weil 2 — correctly, one person behind each. So voice: traditional would have made the Havámál and the Mahābhārata one speaker, with no mechanism to separate them: the same flattening, committed in the mechanism instead of the taxonomy. Jurist accepted, and sharpened the reason — individual-author voices aggregate at the person because a person is real and singular; traditional matter has no such person, so identity lives at the work. Four slots, each one job: identity voice: · relation quoted_by: · category (closed pair, out of the key) · per-source prose note.

Numbering caught before it landed. The drafted entry was REVIEWED-113, matching its PENDING. PENDING-110 rules the sequences independent and REJECTED renumbering to align; 113 would have skipped 97–112 and entrenched the false expectation. Filed as REVIEWED-97 (PENDING-113) per that item's convention (b). Steward placed it; verified clean — single heading, no double-header, all seven load-bearing parts present.

Filed, committed: docs/voice-non-individual-origin-2026-08-08.md (reasoning of record) + docs/vi-disposition-DRAFT-2026-08-08.md (marked superseded-in-part, keeps the population read, consumer measurements, step-3 defects, control record) — 824139d. PENDING-114 (Harrison/Mark) → steward AUTHORIZED (b)+(c) → REVIEWED-98 drafted and placed same day (verified clean). PENDING-115 (two step-3 defects) filed — checked first: 0 prior mentions in the register.

PRESENT — how it stood

A fleet test has been red for a day and nobody looked. Ran the suites to answer the steward's base-rate question: 202/203, one failure — test_navigate.py asserts div:mauss-essai-sur-le-don/body/s4/s4.1/s4.2, but 118f411 split body into body-01…13, so the node is now …/body-04/…. Verified the invariant itself still holds (Scolie [1259,1286] inside host [1234,1294]) — stale fixture, not tree damage. But the check fails at its existence precondition, so the containment invariant it guards has been unverified for a day, and it surfaced only because the assertion was written as a hard check rather than a skip. Fourth defect traceable to 118f411.

My own error reached the register and I caught it late. PENDING-114 cited harrison-dominion.md. That file does not exist — harrison-dominion is the manifest id; the file is the-dominion-of-the-dead-harrison.md. Found by a link-resolution canary after it was filed. Third instance today of citing a derived label instead of the substrate: sidecar title → surah · length → authorship · manifest id → filename.

Two proxy controls failed, and the ladder is what diagnosed both. The cruft scan reported 0 hits in the blocks when the scanner finds 0 anywhere in the file — vacuous; replaced with a positive control (known-bad string, 1 hit per signature) which established the real result: step 3 introduces no cleaning-gate exposure. The fence purity check is above. Both are REVIEWED-83 A1's shape — a control must sit at the layer the defect lives in — and that lesson was already banked and is what let me name them. The gap was firing, not knowledge.

FUTURE — what pulls

PULLING THREAD — the quotation-in × translation-of composition package. All twelve (vi) blocks are translated matter and role is single-valued, so each is quoted-in and translated-from at once. REVIEWED-96 filed this as an edge case against Ungaretti-in-Harrison; it covers the entire (vi) population. (vi) is decided but inapplicable until it is ruled — an unstable state — and behind it sit the Mauss remediation and V2's fr gold, the thread this all started from. PENDING-111's amendment is placed, so REVIEWED-96's sequencing gate on Q3 is clear; this composition is what is left.

ACTIONABLE RESUMPTION POINT (as of wrap — re-judge against what changed):

0. Nothing half-finished. studium-engine main [ahead 7], clean tree.
   Steward agreed the order: composition package FIRST, then PENDING-114 (b).
1. Draft the quotation-in × translation-of jurist package (/jurist-package skill).
   Repo-blind, everything inlined. The four questions it must put:
     - can one section carry two relations, or does role need to become plural?
     - is `translated-from` a property of the section or of the span (Q2's
       span-layer route, DEFERRED at REVIEWED-96, may answer both at once)?
     - what does a citation from a doubly-related span assert?
     - Ungaretti-in-Harrison (Italian verse in an English book) as the second case.
2. THEN PENDING-114 (b): validation phase first — hand-score the known-positive
   (Harrison/Mark 16:7-8) against the known-negative (Weil's Gita/Upanishad
   MENTIONS, not quotations) BEFORE any corpus claim. (c) attached as ruled.
3. [DONE 2026-08-08] REVIEWED-98 placed + verified clean (L1080, 8/8 parts).

Other open horizons, ranked:

• [load-bearing, mechanical] test_navigate.py's stale Mauss node id — one-line fix, but the finding is the proposal below, not the fix.
• [load-bearing] PENDING-115 (a1)+(b1) — block step 3 regardless of any ruling. ⚠ translation carries the identical ROLE_CLASS gap, latent by absence: 0 role: translation sections exist corpus-wide. First Loeb bilingual trips it silently.
• [owed, cheap, open since 2026-08-07] Q5 conversion-signature survey across 14 sources (REVIEWED-96 made it BLOCKING before sizing) · file the PENDING-112 ruling verbatim as a repo doc (live PENDING-108 instance).
• [steward, small] the Havámál L850 call: split as Mauss, or declare as L1043 was.
• [owed] the twelve per-source provenance notes — unblocked for writing (only tagging is gated); six need "Ibid." chains resolved inside the fenced apparatus.
• [open] V2 fr gold needs ~8 stratum-A pairs authored; nobody is scheduled.
• [evidence] attach REVIEWED-96 and the surah correction to PENDING-86 — the jurist could not have caught a Qur'ánic mislabelling without substrate access.

PAUSE STATEMENT: I am putting this down at a genuine close rather than mid-arc — the disposition is decided, placed, reasoned, committed, and both of its blockers are in the register. What I want to find still pulling is the composition package, because (vi) being decided but inapplicable is the unstable state, and a ruled thing left unapplied quietly becomes a stale thing. The unease I carry is not about today's work but about today's fourth defect in 118f411 — a commit already corrected twice was still hiding a red test, and the only reason I found it was that the steward asked an unrelated question about instrument reliability.

LITERAL QUESTION for next-Claude (checkable — the record answers it, not introspection): Was the test fleet re-run after the last change to a sidecar or corpus artifact, and did it catch anything? Today the answer was no, and yes — 118f411 left test_navigate.py red for a full day, through two rounds of correction to that very commit, and it surfaced only by accident. If the answer is no again, the remedy proposed at §1.6 (extend the studium-engine pre-commit hook to run the fleet when corpus/ or corpus/sidecars/ changes) is what to build, and this is its second data point.

---

CODA — after the wrap (the session did not end where the wrap did)

REVIEWED-98 placed and verified (L1080, 8/8 load-bearing parts, no double-header).

The steward pressed on my base-rate answer — "How?" — and it was rhetoric. I had written "act, don't wait" and named three remedies. Under challenge: one was a proposal awaiting authorization, one was a one-shot I had described as standing, and one (the ladder-ritual trial) is measurement, not remedy at all. Naming a remedy class plus three examples read as deployed when nothing was. Recorded as a drift pattern — rhetorical closure substituting for mechanism.

Then the steward asked the better question: "do we need so many single-use items? This seems to flow against our prime directive — but I honestly don't know." Answer banked at feedback-one-shot-instruments-are-proportionate: no — a one-shot is proportionate to a question asked once, and its counterfactual is an assertion, not a durable tool. The violation is re-writing what is already banked (same day: I typed a link-resolution canary inline that lives in /wake-up AND on the ladder). Too few promoted, not too many built — and only the first is actionable.

Applied (FIX lane, three instruments discharged): /wrap-up §8 now carries a standing Instruments field — N run · M with a control written before first execution · K duplicating something already banked. The K column is the steward's question turned into a series rather than left a worry.

Filed: PENDING-116 [PROPOSAL] — run the fleet on the change that breaks it. Design derived from reading the hook: core.hooksPath is ~/dotfiles/git/hooks, so the hook is tracked and travels but is global to every repo → the trigger must be repo-declared (generative-from-spec, as the chamber already does). Costs stated up front: slower triggering commits · --no-verify bypasses it (a tripwire, not a boundary) · ⚠ it does not close the cross-repo half — a chamber-library edit can invalidate engine fixtures with no engine-side commit at all, so no hook fires on either side. That larger hole is named as an open follow-on, deliberately not absorbed.

A fourth check-vocabulary failure, same day. My verification that the FIX-lane index line had landed grepped Instruments field; the line reads **Instruments** field, so it returned 0 against a line that was there. The write was fine; the check was wrong — again.