Files

T

David F GliddenandClaude Opus 4.8 0a424aa6b3 session 2026-07-12: PENDING-56 (canonical-format decision, jurist-ratified) + in-flight sidecar build report + session memory/ledger/REVIEWED-drafts

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015jYS1tpR4P4NBxui4FSgfv

2026-07-12 12:15:15 +02:00

4.6 KiB

Raw Blame History

In-flight work report — 2026-07-12 (unsupervised, steward travelling to BCN)

Authorized scope: read-only / reversible / dev-only build + docs; wrap when done. Nothing pushed, no corpus mutated, no spec/graduation-gate changed, no REVIEWED placed. Eight tasks, all complete.

Done

Sidecar schema — DRAFT v0 (chamber-library/docs/sidecar-schema-DRAFT-v0-2026-07-12.md). Extends the engine's studium/meta@1 (aligned, not forked); TEI-mirroring vocabulary (front/body/back · note @place · milestone/anchor · reserved app slot · pb/provenance); the two cardinal principles baked in. Marked DRAFT — schema-lock is yours + the jurist's, not the executor's (as the jurist required).
Extractor built + proven — chamber-library/scripts/build_sidecar.py (v0 PROTOTYPE, NOT in test_tools). Recovers structure from the source's OWN metadata (not post-processing Docling's flattened MD).
- Core thesis PROVEN: montaigne (EPUB2) 16/17 sections declared (guide+filename); jacobs (EPUB3) 37/37 declared, 0 flagged (nav-landmark propagation); provenance 400 pages (jacobs pagebreaks); notes 56 anchors (montaigne). The front/body/back partition — the A1 drop-vs-keep answer — is a declaration, not a heuristic.
- Honest v0 limits (characterized, not hidden): the Docling content-match line-join is brittle where the EPUB opening and Docling's rendering diverge (montaigne 2/17: drop-caps T HERE↔THERE, arabic-vs-roman headings; jacobs 34/37 after a <head>-strip fix). Note-anchor scheme is per-publisher (jacobs 0 vs montaigne 56). → the real fix is architectural: convert Docling PER-SPINE-FILE (exact line-ranges by construction) + reuse inject_epub_footnotes for the note-layer — deferred to the wired build.
- Drafts written to chamber-library/_curation/sidecar-drafts/ (dev-only).
Tool-review logged to chamber-library/_curation/tool-evolution-log.md (2026-07-12 entry; PASS-BUT-FALSELY watch = the v0 drafts carry per-publisher quirks → curator approval mandatory, per the schema).
conversion-runbook.yaml — added machines: (the explicit M4/M1 split) + pipeline_direction_pending_56: (the Docling→sidecar→verify→graduate direction) + updated the A3/apparatus known_gaps.research + the top changelog. Stamped "per PENDING-56 (ratified; REVIEWED-56 pending); existing pipelines stay authoritative until the amendment." (Note: the runbook was never strict-YAML — the pre-existing decision_tree → arrows; nothing yaml-loads it; my additions parse clean.)
chamber-library CLAUDE.md — Docling + build_sidecar.py in the tool fleet, the machine split, the pipeline pointer. Stamped PENDING-56.
studium-engine CLAUDE.md — noted D-2 is now realized chamber-side; the sidecar aligns with studium/meta@1 (don't-let-them-drift); heavy conversion on the M4.

Decisions I hit — SURFACED, not answered (yours / the jurist's)

Q-A. Schema-lock. The DRAFT needs your + the jurist's review before it's locked (1,284 books write into it). Specifically confirm: the placement (front/body/back) + source_declared section additions; the notes/provenance/anchors shape; and that app stays RESERVED (interior TBD at B2). No change until you lock it.
Q-B. Real-build architecture. I recommend Docling per-spine-file (exact line-ranges) over the v0 content-match join, and reusing inject_epub_footnotes as the note-layer. Confirm before I build the wired extractor.
Q-C. Fleet-promotion. build_sidecar.py → scripts/ + test_tools.py fixtures, after the schema locks. (It's in scripts/ now but flagged NOT-in-fleet.)
Q-D. The graduation-spec amendment (step 6). Wiring Docling+sidecar as the graduated structural layer changes what graduation accepts → PROPOSAL, runs the amendment process. Not started (correctly — needs you + the jurist).
Q-E. The [c darkmagenta] content question (Loeb B2). Whether it's apparatus-variants vs bilingual source-text — resolve by sampling volumes when B2 is built. The preserve-never-flatten principle covers both meanwhile.

Still owed by you (unchanged)

File REVIEWED-55 + REVIEWED-56 (paste-ready: REVIEWED-drafts-for-filing-2026-07-12.md).
Relay the PENDING-56 §8 closure to the jurist (or note it's already ruled).

Next session (you said: clear + `/wake-up` fresh)

The pulling thread on return: Q-A/Q-B (lock the schema, confirm the per-spine-file architecture) → then build the wired extractor and the Docling-per-file join. Everything constitutional (amendment, REVIEWED filing) waits for you. The forest view (chamber-library/docs/chamber-program-open-work.md) + this report are the frame.

4.6 KiB Raw Blame History