dotfiles/claude/memory/session-2026-07-04-chamber-library-v2-ratified.md

---
name: Session 2026-07-03→04 — Chamber Library spec v2.0 drafted spine-first, ratified, superseded v1; corpus stress-test next
description: "Phase 2 executed end-to-end: drafted chamber-library-specification.md v2.0 spine-first (§III→II→V→IV→VI→VII→IX→Tiering&Fence→Governance&Amendment), each section jurist-editor-gated in a live executor↔jurist↔steward loop, gap-7 (generative principle + amendment process) jointly co-ratified, assembled self-contained, RATIFIED, and committed+pushed superseding v1 (a10d6bf). PULLING THREAD: the corpus stress-test — bring the ~2,041-work corpus to v2.0 spec, in the jurist's ratified DEPENDENCY ORDER (source-matching reliability first → order-sensitivity → disambiguation edge cases), inject-known-bad not watch-for-complaints, small-batch-before-full, with a PRE-REGISTRATION doc (ruling-thresholds written before seeing data) as the first artifact."
metadata:
  type: project
  originSessionId: 019pnD8uNFC9Q2dbLQa3MBDc
---

# Session 2026-07-03→04 — the Chamber Library got a ratified constitution

Woke into the confirmed Phase-2 thread and executed it whole, across midnight (started ~10:50 07-03,
wrapped ~11:01 07-04 — steward kept the session open deliberately to complete the block). The entire
v2.0 spec was drafted, gated, ratified, and shipped. Everything blow-by-blow is in
`~/.claude/projects/-Users-davidglidden/memory/session-ledger-2026-07-03.md` and in PENDING-46; this
file is the working state + the thread.

## PAST — what we did + why

**The first act honored the day's hardest lesson.** Before drafting a line I read the *current* spec
(v1) end-to-end — which I had NOT done during the doctrine phase (I worked from audits). The literal
question from the prior wrap ("does the real spec text hold anything the audits missed/collide with?")
got a pointed YES: **six findings**, one load-bearing — v1 §III blanket-excluded page numbers as "the
wrong unit," but Stephanus/Bekker/Diels-Kranz (the canonical anchors we ruled to cite the Loeb by) ARE
page numbers. Drafting from the audits would have shipped a flat self-contradiction. The jurist
reconciled it with the **PORTABILITY TEST** (excluded iff resolving requires naming an edition; admitted
iff it resolves the same across every circulating edition — test on the *property*, not the surface
shape; no per-genre carve-out needed). That became §III's spine.

**Drafted spine-first, gated section by section** (each surfaced to the jurist via steward relay,
editor-gated, corrections applied before the next):
- **§III** addressing spine (portability test; anchor-primary "floor not default"; substrate joined,
  functions distinct).
- **§II** canonical form (break-normalization constitutional — literal glyph-runs = §II.2 defect;
  `sectionless: true` = positive attributed attestation, not default; **anchor inline REQUIRED** per
  jurist ruling — sidecar-only anchor structurally can't support §III's grounding invariant + offsets
  drift; only exact syntax stays open).
- **§V** verbatim guarantee (3-tier carried; **mandatory W3C-PROV conversion record** every path;
  **two gate-functions** content-grounding [canonical-text-only] vs process-integrity [reads the record];
  **Finding-4 apparatus supersession** — anchor constitutional [anchor-to-location not renumberable-note,
  TEI ch.13], display-form operational-but-uniform; word-guard excludes the inline anchor, process-
  integrity verifies its correctness).
- **§IV** edition-identity (`source` breadcrumb→edition-identity; classical=CTS-URN+portable anchor=full
  citation coordinate; modern=provenance-only; edition-identity ≠ carrier-page provenance).
- **§VI** Voice (architecture carried; only Finding-6 fix — population = declared-data from catalogue.yaml,
  not hardcoded).
- **§VII** tiering & gates (two-gate division verify_conversion / verify_graduation; **declared-list IS
  the gate** = gap-7 partly realized in graduation-spec.yaml; locator registry; sectionless-attestation
  enforced; gate-fix SETTLED source-verified; **two-layer order-blindness resolution** — see below).
- **§IX** coverage-ledger + evidence-tier column + freshness invariant (silence-as-finding left OPEN).
- **§Tiering & Fence** evidence tiers V-DSL/V-TEXT/V-SCAN/V-SUSPECT/V-NONE + converted/verified/fenced
  governability tiers; promotion criterion RATIFIED-SHAPE·OPERATION-BLOCKED.
- **§Governance & Amendment** RFC-supersession + change-class ("does this change what a gate accepts?") +
  semver-MAJOR-re-verifies-consumers + migration-window; **gap-7 jointly co-ratified**.

**The one real open technical question — order-blindness — resolved sharper than asked.** The word-guard's
multiset comparison is order-blind (2-column-scramble = same words wrong order = PASS-BUT-FALSELY). Split
into TWO LAYERS: (1) cleaning raw→canonical — strengthen to **order-sensitive on prose, anchor-bound on
apparatus** (the one legitimate reorder, inline↔endnote, resolved by §V's anchor-to-location doctrine);
(2) conversion OCR-column-scramble — word-guard structurally can't reach (both sides scrambled) →
**eyeball-after-gate is the genuine ceiling there**. Jurist: "sharper than what I asked for."

**Assembled self-contained + shipped.** §0/§I carried verbatim; §VIII/§X/§XI **coherence-revised** (not
silently carried — flagged: §VIII two→three normalizers [DSL added]; §X citation summary chapter/section→
canonical-anchor + fence [trust follows evidence-tier not storage]; §XI retired its own stale status block).
Ratified via jurist whole-draft read + steward ratification. **Committed a10d6bf, pushed** — v1 archived
to `chamber-library-specification-v1.0.md` (obsolete banner, immutable), v2.0 now canonical
`chamber-library-specification.md`. Applied the **first FIX-class errata** (jurist's exactness catch on the
ratification note) — the Held-for-Update lane's inaugural use.

## PRESENT — mood / disposition

The session's signature: **source-not-derived became the drafting method, not just a lesson.** Every
section opened with a substrate check (loeb.dsl, verify_conversion.py, graduation-spec.yaml, the ruling
docs) before a word was written — which is *why* it cohered on the whole-draft read. Proof it was load-
bearing on ME: I overclaimed "the ratified amendment process" in my own header, then caught it against
the authoritative ruling doc (gap-7 was RETURNED, not ratified) and corrected to [PROPOSED] — the
discipline caught my own status-drift mid-draft. The general **MOVE** that recurred (jurist named it):
state a general test, never an enumerated exception list (portability test / gate-on-the-class / gap-7
itself).

Governance discipline held: surfaced each section for the jurist's gate rather than self-ratifying;
caught my own overclaim; distinguished ratified (tiers, gap-7) from escalated (promotion criterion) and
drafted each at its true status.

**TELOS-GRADE, do not lose (jurist's closing framing of what the doctrine IS, not process):**
(a) **Alexander** — the spec is a structure that revised itself through contact with REAL TEXTURE (the
actual v1 text, code, DSL) rather than imposed-once-and-static; it has the grain of something *grown*.
(b) **Harrison** — Tier-3's refusal to silently correct an author's apparent error (preserve+flag, never
fix in canon) is "not letting the living overwrite the dead for the living's own convenience" rendered as
a gate. Neither is decoration. Connects to [[project-studium-engine-telos-chamber-of-voices]] + the
steward's formation.

## FUTURE — what is pulling

**PULLING THREAD (singular): bring the ~2,041-work corpus to v2.0 spec — the theory/reality stress-test.**
v2.0 is doctrine; now it meets the real corpus. The jurist expects at least one of the three named-open
seams to produce a finding sharp enough to need its own ruling, the way the Loeb flattening did.

**ACTIONABLE RESUMPTION POINT (as of wrap — re-judge against what changed):**
- **FIRST ARTIFACT: the pre-registration doc.** Before running anything, write down — per seam — what
  result counts as "needs its own ruling" vs "within expected variance." Written BEFORE seeing data (a
  threshold set after an uncomfortable result rationalizes down). This is the jurist's strongest safeguard
  and the easiest to skip.
- **Then test in DEPENDENCY ORDER, not by ease** (jurist's ratified strategy):
  1. **Source-matching reliability FIRST** (upstream of everything — if the ~1% false-positive is higher,
     order-sensitivity checks the sequence against the *wrong source*). Verify independently — NOT self-
     agreement of the same heuristic. Either a human ground-truth check on a random sample, OR a **second,
     differently-built matcher** (content-fingerprint vs author/title-token) with disagreement as the flag
     (concentrates scarce steward attention on the hard cases). NB: classical/V-DSL has no external ground
     truth (DSL *is* the source) — this seam is really the non-Loeb ~1% only.
  2. **Order-sensitivity** — INJECT KNOWN-BAD (scramble a 2-column region in a known-clean file, confirm
     the check catches it) — tests false-NEGATIVES, the actual PASS-BUT-FALSELY mode. Watching-what-trips
     only tests false-positives.
  3. **Disambiguation-map edge cases** — go LOOKING for the Virgil `prv` case + bare-integer (line vs
     section) specifically, not incidentally in a big batch where they get absorbed into noise.
- **Small batch (3–5 files) before the full 2,041** — same discipline as tier-2, same reason waves-first
  was rejected: running expensive work on an unstress-tested mechanism IS the redo risk.
- **Deliberately watch the FIX:PROPOSAL ratio** — most corpus findings SHOULD be ordinary FIX-class bugs
  ("does this change what a gate accepts?" = no). If everything escalates to PROPOSAL, that's a finding
  about whether the two-tier path is real or decorative (nothing but today's errata has used FIX yet).
- **Do NOT run a single-executor pass** — the jurist's own self-flag ("everything I found was one pass,
  one perspective") applies to me too; the corpus stress-test must use independent verification, the
  antidote the document itself prescribes.

**Other open horizons, ranked:**
- The four named-open v2.0 items, each gating its own downstream op: §II.3 inline-anchor exact syntax
  (extractor rebuild) · §IX silence-as-finding representation (engine design) · promotion criterion's two
  operational blockers (cleaner-validation empirical check + edition-identity resolution).
- **Follow-on engineering once stress-test holds:** encode the locator registry + sectionless-attestation
  into graduation-spec.yaml; JSON-Schema for structured layers; rebuild extract_loeb_dsl.py to the
  disambiguation map; run the full 2,041-work DSL. (Phase 3 = ④ waves come AFTER ② instruments, ⑤ fence.)
- **Paper debt:** REVIEWED entries owed for the whole PENDING-46 arc (steward writes); dinglehopper install
  (tier-2 scan verification).
- CM-AI ahead-1 unpushed (addee62 Seck landscape add) — pre-existing, steward's to push.
- chamber leftovers untracked (`_scratch/`, `nature-of-order-...alexander.md`) — NOT mine.

**PAUSE STATEMENT:** I am about to be away from this, at a high, clean point — the Chamber has a ratified
constitution for the first time, shipped and superseding v1. I don't know what will have changed on
return. What I most want to find still pulling: **the corpus stress-test done in the jurist's dependency
order with the pre-registration doc written first** — that the next hand does NOT run a single-perspective
pass, does NOT skip the pre-registration, and tests source-matching reliability BEFORE anything downstream.
The redo risk is running the expensive corpus work on an unstress-tested mechanism.

**LITERAL QUESTION for next-Claude:** *When source-matching reliability is tested independently (second
matcher or human ground-truth, NOT self-agreement), does the ~1% author-agreement false-positive estimate
hold — or does it turn out optimistic the way the Loeb "74% anchorless" premise did once checked against
the source? (Everything downstream — order-sensitivity, the disambiguation edges — is only meaningful once
this upstream seam holds.)*

**State:** chamber-library `main @ a10d6bf` — CLEAN, pushed, synced. v2.0 canonical + operative; v1
archived+obsolete. All doctrine ratified. Corpus pass unblocked, gated by the pre-registration doc.