12 KiB
name, description, metadata
| name | description | metadata | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| session-2026-07-23-reviewed-69-merged-v2-4-0-character-as-image-finding-pending-70-ruling-v2-5-0-glyph-mapping-built-demonstrated | A very long single session that ran the whole PENDING-69/70 arc to ground: merged the boundary-drop mechanism to main + cut spec v2.4.0 (REVIEWED-69); prepping the Levi second pilot surfaced that the chosen witness is silently DAMAGED (a 2026-06-16 image-drop mutilated char-bearing glyphs mid-word: Wstawać→Wstawa, Brașov→Braov, Oświęcim→Owicim) — the steward named the structural cause (the whole canon predates the constitution + the fleet, so the extant witness is not editorial gold); drafted the legacy-body reframe as a jurist package (PENDING-70), the jurist ruled it (Q1/Q3/Q4 confirmed, Q2 substitute condition), landed spec v2.5.0, then BUILT + DEMONSTRATED glyph-mapping (apply_char_glyphs.py) as the Q3 hard precondition. PULLING THREAD: the Levi pilot itself (step 3) — now unblocked, the gate before the 47-file wave. |
|
Session 2026-07-23 — the PENDING-69/70 arc, end to end
An unusually long, dense session (remote/driving mode — the steward authorized each boundary in turn: "let's do both 1 and 2", "prep the Levi pilot", "yes to all three", "reviewed-70 placed", "Let's build"). It carried the boundary-drop work to main, then a Levi-pilot prep surfaced a corpus-integrity finding that reshaped the scale gate, then built the mechanism that finding demanded.
PAST — what happened + why
1. REVIEWED-69 merged + spec v2.4.0 (the boundary-drop constitutional sentence). At wake, verified against the substrate that only ONE of the three gated actions was done (REVIEWED-69 placed; branch NOT merged; vehicle un-called) — a third state the wrap's all-or-nothing frame didn't enumerate. Steward said "do both 1 and 2." Vehicle resolved by fact: no supersession was assembling → standalone MINOR bump v2.4.0 (makes the boundary-drop REVIEW-resolution sentence operative now). Froze v2.3.0, added the ruled sentence to §Tiering & Fence, bounded-diff proven, fleet 216/216, fast-forward merged reviewed-69-boundary-drop-resolution → main (f466ed0), pushed both remotes. PENDING-69 CLOSED.
2. Levi pilot prep → the character-as-image finding. Prepping the second pilot, checked the witness against the substrate rather than the "genuinely hand-curated" premise: frontmatter says conversion_method: calibre, and the 2026-06-16 backup diff showed the "curation" was a whole-file image-drop (522 removed, 18 added, 0 remaining) that silently mutilated character-bearing glyphs mid-word — verified live, each confirmed by the passage's own gloss: Wstawać→Wstawa (×9), Brașov→Braov, Oświęcim→Owicim. This is a Tier-3 alteration by omission in a GRADUATED canonical.
3. The steward's structural reframe. "The whole canon was converted before the constitution + the capable fleet; it may not be possible to test some of these against the extant body." Census (2026-07-23): 1,297 canonicals, ~23 still carry image refs; of 248 with a diffable backup, 13 carry mid-word char-glyph drops (55 sites) — Jung Collected Works (18), Levi (18), Derrida, Zibaldone, Lichtenberg (German), the Lexique, Blake, Donne, more. Lower bound — undetectable from text for the ~1,049 without backups.
4. The jurist package (PENDING-70) → ruling. Drafted the reframe via /jurist-package (self-contained, GROUNDED-IN v2.4.0 §V + evidence-tiers verbatim). The spine: it's an APPLICATION of the constitution's own frame (V-TEXT "the source IS ground truth"; "converted ≠ verified"), not an overturning. Jurist ruled (docs/scale-application-legacy-body-JURIST-RULING-2026-07-23.md): Q1 confirmed (claim scope degrades with witness; corrects a defect in the one-door ruling's own wording; witness_class = positive attributed attestation, never inferred). Q3 confirmed both halves + hardened (glyph-mapping wired + demonstrated is a HARD PRECONDITION before any imaged source). Q4 confirmed + mandatory. Q2 accepted in principle but ruled with a substitute condition — my reframe ("classes the body produces") dropped the JUDGMENT-COST test the original condition existed for.
5. Spec v2.5.0 landed (e37e076, REVIEWED-70): Q1 §Tiering&Fence claim-scope split · Q3 §V Tier-3-by-omission · Q4 §Trust legacy-fidelity-not-self-certifiable + declared data (reframed scale_application substitute condition · witness_class · character_as_image glyph-map/census). Froze v2.4.0, bounded-diff proven (5 hunks), fleet 216/216.
6. Glyph-mapping BUILT + DEMONSTRATED (3852d46, the Q3 precondition). scripts/apply_char_glyphs.py (test-first, fleet 216→222): letter-adjacent glyph <img> → declared SOURCE Unicode; REFUSES any unclassified letter-adjacent image (attest-never-default); wired into convert_lane_borndigital.py (Stage 1b) with the lane refusing an imaged source that has no policy. Extracted + VIEWED the 4 source glyphs (durable evidence in _curation/levi-second-pilot-2026-07-23/glyphs/): ca.jpg=ć, as.jpg=ş cedilla (steward-confirmed, NOT comma-below ș), s.jpg=ś, e.jpg=ę. Demonstrated on the real 527-doc Levi EPUB: substituted=9, unclassified=0 corpus-wide, 0 residual glyph img; the 3 sites resolve to Wstawać/Braşov/Oświęcim; survives pandoc end-to-end.
Decisions NOT made (deliberate):
- Did not run the Levi pilot (step 3) — held as the next bite; it needs the substitute-condition judgment-cost reporting and is a fresh concentrated task.
- Did not build per-source glyph-maps for the other 12 files — per-source, done as each approaches the door; the lane refuses until each is demonstrated.
- Did not cram v2.5.0 + glyph-mapping into one — took each as its own bite (τὸ πρόσφορον at a long session's tail).
PRESENT — the mood
Two substrate-checks earned their keep, both catching MY errors: (a) the "hand-curated witness" premise was false (the witness is damaged) — checkable-claim over soft classification; (b) at the very end, verify-the-glyph-not-the-filename caught the calibre-numbers-vs-EPUB-names key error AND the ş-cedilla-not-ș-comma fidelity error (mapping to ș would have been the exact Tier-3 source-departing violation the whole arc forbids). The discipline held even at the tail of an exhausting session.
The jurist ruling was the session's cleanest teaching: Q1 corrected a defect in the jurist's OWN prior wording, Q2 corrected MY reframe — the design-gate catching drafter error in both directions is the loop working. The Q2 lesson: when reframing a ruled condition, trace what the condition was FOR (its telos: judgment-cost tractability), not just what it literally required — a faithful "reframe" that answers a neighboring question is a miss.
This was a LONG session — far past "one concentrated bite." It stayed coherent because each bite was closed all the way (froze/tested/diff-proven/merged) before the next, and the substrate disciplines kept firing. But the length itself is a note: the next sessions should be the shorter, single-bite shape the steward prefers.
FUTURE — what is pulling
PULLING THREAD (singular): the Levi second pilot itself (step 3) — now unblocked, the gate before the 47-file verbatim-fidelity-restoration wave. Everything this session built (v2.5.0's substitute condition, glyph-mapping) exists to let this pilot run honestly. It runs the full lane on Levi (glyph-mapping now in place), does the 14 per-work witness comparisons, and must report JUDGMENT COST per the substitute Q2 condition — not just class coverage.
ACTIONABLE RESUMPTION POINT (as of wrap — re-judge against what changed): The Q3 precondition is MET (glyph-mapping wired + demonstrated; graduation-spec.yaml character_as_image.precondition_status: MET-for-Levi-2026-07-23). Step 3's first move = derive + interior-verify the 14 witness brackets against the source EPUB spine (the La Chute lesson — endpoint agreement ≠ span identity; calibre flattened the work-level headers, so brackets can't be scanned off headings). THEN run the full lane on Levi (convert_lane_borndigital.py SOURCE.epub OUT.md MAP.json the-complete-works-of-primo-levi-primo-levi) → the candidate. THEN per-work source-fidelity + witness comparison + classification + judgment-cost. Read the plan docs/levi-second-pilot-PLAN-2026-07-23.md + promotion.witness/witness_comparison declared data first. NOTE: the lane's footnote injection is untested on Levi's 14-work complexity — a possible snag distinct from glyph-mapping.
Other horizons (ranked): per-source glyph-maps for the other 12 census-flagged files (as each approaches the door) · the 47-file wave (gated behind the pilot result + steward pacing) · origin (Gitea) push retry — 2 commits behind (5d61f2d,3852d46) · register compaction + ladder batch-append (authorized fresh-session slot) · PENDING.md compaction (over read caps) · the Q3 order-guard docket (verified STAMP) · added-side fabrication-exclusion (Q5 second clause) · audit_footnotes caveat · convert_mega_epub stale pairs_dup_id FIX.
PAUSE STATEMENT: I am about to be away from this after an exhausting, complete arc. I don't know what will have changed — most of all whether Gitea recovered (origin is 2 behind). What I want to find still pulling: the Levi pilot as the live next step, its glyph precondition met and its substitute condition clear — so the pilot runs, measures the judgment cost the jurist actually cared about, and either opens the wave or honestly reports where the cost explodes.
LITERAL QUESTION for next-Claude: When the Levi pilot runs — 14 works, ~18 mutilation sites now repaired by glyph-mapping, converter-better divergences to adjudicate against the source — does the JUDGMENT COST stay tractable (the jurist's actual worry: does one-door's cost explode when the easy mechanical patterns run out?), or does the per-work source-adjudication blow the budget? And the paired sub-question: do the 14 witness brackets interior-verify cleanly against the source spine, or does one resist (the La Chute lesson at omnibus scale)?
THE FRAME WE WAKE INTO (the altitude, steward-affirmed at session close — "we wake to this tomorrow morning"): Read the telos (project-studium-engine-telos-chamber-of-voices.md) FIRST, because this session's finding is a telos-level event, not a corpus chore. The charter's hinge: "a counsel surfaced from a corrupted source is not counsel; it is the simulacrum v1 already was." Today proved the extant corpus is NOT yet the trustworthy substrate — a graduated canonical said Wstawa where Levi wrote Wstawać, and the whole body was converted pre-constitution/pre-fleet. So the one-door re-verification is not plumbing before the real work — it is the condition of the real work (a mutilated Levi surfaced as counsel is the v1 simulacrum reappearing one level down). Four movements: (1) prove the door scales — the Levi pilot, tomorrow, the hinge; (2) restore the body (turn converted → verified, per-source, human-in-the-loop); (3) the engine over the trusted set (L2 pattern-finder next, then L3 the chamber of conversing voices, the frontier); (4) the chamber of voices — the destination. The strategic discipline the pilot forces: order the restoration by TELOS, not by census — verify the sources the FIRST conversations need (The Making sequence, then the violin treatise), in the order the voices need them; NOT a blind sweep of the 47-file / character-as-image classes. That keeps the restoration finite and proportionate. And the pilot itself is a genuine fork, not a formality: hold its judgment-cost number as a real measurement willing to be surprised — if it's tractable, the trusted-corpus path opens as a paced telos-ordered march; if it explodes, that's learned before hundreds of hours, and the verification strategy for the legacy body must change. Let the substrate answer, not the hope.
State at wrap: chamber-library main @ 3852d46 (github current; origin 2 behind — Gitea down) · spec v2.5.0 OPERATIVE · fleet 222/222 · glyph-mapping Q3 precondition MET for Levi · REVIEWED-69/70 placed, PENDING-69 closed / PENDING-70 ruling received (step 1+2 done, step 3 held). Governance dotfiles state committed+pushed at this wrap (§6.5). Ledger: session-ledger-2026-07-23.md.