dotfiles/claude/memory/session-2026-05-14-evening-arc-typography-multilingual-spec-landed.md

name, description, type, originSessionId, permalink

name	description	type	originSessionId	permalink
Session 2026-05-14 (evening) — ARC typography arc landed end-to-end; multilingual spec drafted v0.2 (French + Spanish); jurist's two-document hierarchy realized; mine still running, pulling thread is tomorrow's mempalace recall-quality test	Full-day ARC typography session — 5 commits, 7 commits ahead on origin. PDF + Print posture XeLaTeX templates built and rendered; EB Garamond updated 2019→2026 with ZapfDingbats glyph fallback; Print posture switched to A4 + Van de Graaf canon at steward's direction; multilingual-typography-specification.md created per jurist's standing instruction (architectural correction to my initial §I.j-within-AldineXXI premise); §1 French + §2 Spanish drafted at operational quality; french-typography-pass skill created. Pulling thread for tomorrow is the mempalace recall-quality test once the mine completes — meaningful cross-corpus connection retrieval without the density-dilution problem we surfaced 2026-05-13.	project	9c5ea210-7d61-440e-8ef0-32e0eb5cba24	claude-memory/session-2026-05-14-evening-arc-typography-multilingual-spec-landed

Inherited state at session start

Wake-up's pulling thread per morning's wrap: ARC PDF/Print posture session per project-arc-pdf-print-postures-session-plan.md (~6h focused). Mempalace mine still running (PID 24651, ~21h in at wake; ~31h by end of session). Stop + PreCompact hooks remained DISABLED per the morning's pre-mine-completion safety preparation. The literal question I left at wrap-up of the morning: "Has Seb engaged with PR #172 or comment on #165?"

The session opened with a recurrence of the warning-in-context-still-launched drift pattern: I launched MemPalace tool calls per the /wake-up procedure despite MEMORY.md's note about the active mine. Steward caught immediately. Updated feedback-warning-in-context-still-launched.md to flag wake-up procedure as a known trigger. Pattern is the day's most stubborn drift; it recurred under different shapes throughout (jurist-authorization-without-constitutional-check at noon was its shape there; tonight it nearly recurred at wrap-up but I caught it before calling MemPalace — see Returns below).

What happened (in order)

1. ARC PDF/Print posture session (Hours 1–6, full afternoon)

Followed the saved session plan as a six-hour scope-divided sketch (the "hours" are scope-units, not strict time):

Hour 1 — Setup. Created tools/latex/ working directory. Built minimal XeLaTeX scaffold: arc-typography.sty (shared substrate), template-pdf-posture.tex (PDF posture wrapper), template-print-posture.tex (Print posture wrapper). Three-layer fontspec stack: EB Garamond (reading) / IBM Plex Sans 0.94 scale (navigation) / IBM Plex Mono 0.88 scale (apparatus). True small caps via EBGaramondSC.otf NewFontFamily. Pompeian palette in xcolor (light theme only this iteration). Caught and fixed: IBM Plex Regular weights have no -Regular suffix (file is IBMPlexSans.ttf directly). Hour 1 fontcheck rendered cleanly with all three layers visible + palette swatches.

Hour 2 — PDF posture proper. Implemented §I.d minor-third heading scale (1.2 ratio, upright, no bold) via titlesec. §I.b list treatment (en-dash markers, italic-non-bold definition terms). §I.h block quotation (left-only indent via \list redefinition — quoting package had a kv-error). §II.b body rhythm via \linespread{1.10}. Asymmetric/symmetric posture flag per §II.d via posture-asymmetric YAML/CLI variable. Compiled the Readings spec entry as test substrate; both modes rendered cleanly across 4 pages.

Hour 3 — Iteration + critique. Added microtype with xelatex-compatible options (protrusion + tracking only — expansion is luatex-only, spacing is pdftex-only; my initial options were engine-incompatible and the steward asked if updates were needed; answer: no, just option-correction). Implemented \textbf → \textit global redirect per §I.d "banish bold" (with \textbfraw preserved for the governed-exception). Added fancyhdr apparatus folio in Plex Mono. Suppressed section auto-numbering globally (\setcounter{secnumdepth}{-1}). Resolved \fontsize collision in fancyhdr footer by using \small instead.

Hour 4 — Print posture. Twoside, Van de Graaf canon at 6×9in initially (Royal octavo, ~2:3 ratio). Mirrored geometry; folio at outer corner; running heads (verso = piece title, recto = section title) in Plex Mono apparatus register. Body sits at the inner anchor of each page. Rendered cleanly across 5 pages with the spread reading correctly.

Hour 5 — Sample image preparation. Created sidenote-bearing variant of Readings spec entry (readings-entry-with-sidenotes.md with three \marginpar{} annotations via raw_tex). Widened Print posture's marginparwidth to 1.0in after initial render showed clipping. Generated six PNGs to static/img/postures/ (PDF asymm + sym, Print recto/verso × with/without marginalia). Spread composition (verso+recto as bound spread) deferred — Pillow not installed; per-project venv recommended when needed.

Hour 6 — Documentation. tools/latex/README.md (build commands, font requirements, eight open questions). docs/AldineXXI-Codex/drafts/postures-rough-draft-2026-05-14.md (rough draft of §II.e PDF Posture + §II.f Print Posture for AldineXXI). Two commits at session end as planned.

Two commits landed:

• e9e566e tools(latex): PDF + Print posture XeLaTeX templates
• 2e740b6 spec(arc): rough draft §II.e PDF + §II.f Print posture + six sample PNGs

2. Open questions sweep + ZapfDingbats glyph fallback (Path B)

The PDF/Print session had surfaced 11 unresolved questions. Steward chose to address them before the Readings post completion (Task #7) and asked specifically about glyph fallback choices. After examination:

• ZapfDingbats (/System/Library/Fonts/ZapfDingbats.ttf, system-installed) has both ☞ U+261E (manicule) AND ✎ U+270E (quill) in canonical Hermann Zapf 1978 design. Apple Symbols has ☞ but NOT ✎.
• EB Garamond v1.000 (2019) — installed 2019-03-14, six years out of date. Upstream octaviopardo/EBGaramond12 active through 2026-03-27.

Path B implementation (ucharclasses with ZapfDingbats fallback): added \newfontfamily\arcGlyph{Zapf Dingbats} and ucharclasses transitions on Dingbats / Misc Symbols / Misc Symbols And Arrows blocks. Initial mistake: only set enter transition; body text after dingbat stayed in ZapfDingbats. Fixed with \rmfamily exit action + explicit BasicLatin / LatinSupplement / LatinExtendedA/B/Additional / GeneralPunctuation enter transitions to restore roman. ucharclasses block-name syntax: \setTransitionsFor{BlockName}{...}{...} (not \setTransitionsForBlockName{...}).

3. EB Garamond v1.000 → v1.131 update (Phase 1)

Steward authorized Phase 1: 10 TTFs in ~/Library/Fonts/ + 3 live WOFF2 (Regular, Italic, Bold) replaced with octaviopardo/EBGaramond12 master (sha 106a4a6, 2026-03-27). Old files backed up to dated _backup_2026-05-14/ directories before replacement.

Phase 2 (re-subset Medium/SemiBold for Latin/LatinExt via pipx-installed fonttools) and Phase 3 (EBGaramondSC — verify Pardo +smcp to retire SC face, else build from Duffner source) tracked at project-eb-garamond-update-2026-05-14.md. Steward direction: "will want to cover the deferred steps soon."

EBGaramondSC source: georgd/EB-Garamond last pushed 2024-03-16, ships SFD/UFO/featurefiles only — no pre-built fonts. Building from source requires fontforge tooling. Hold for Phase 3 session.

Hyphen warnings regression: after v1.131 + ZapfDingbats fallback, the print render produced hyphen-on-ZapfDingbats warnings because my initial ucharclasses only handled enter-into-dingbat, not enter-into-latin. Added the Basic Latin / Latin Supplement / Latin Extended / General Punctuation transitions — clean.

Commit: d9efb88 fonts(arc): EB Garamond v1.131 + ZapfDingbats fallback via ucharclasses. 11 files changed including 3 binary WOFF2 + 6 regenerated PNGs.

4. A4 paper size for Print posture (steward decision)

Steward: "Paper size would be A4 here, but I admit that I hadn't considered paper size. Can we start with A4 please." Continental working geography (Barcelona) makes A4 the natural substrate; 6×9in was the spec-draft's first-pass convention, not the steward's working context.

Van de Graaf canon at A4: page ratio 1:√2 ≈ 0.707 (not the canon's original 2:3 construction ground), but the canon's defining property holds — 1:1:2:2 margins still produce a text block at the page's own ratio (140 × 198 mm = 1:√2). Margins in mm: inner 23.33, outer 46.67, top 33, bottom 66. Body bumped 11pt → 12pt to maintain ~65 chars/line on the wider 140mm measure (Bringhurst §2.1).

Marginpar widened to 33mm in the wider outer margin. Print samples regenerated.

Commit: ca36e44 spec(arc): Print posture paper-size → A4 (Continental working substrate).

Other steward calls in the same exchange:

• Page proportions for PDF posture: "ok for now, need time to live with them before a change to see how they feel." Deferred.
• Drop caps: deferred until vignette system is functional ("question of thresholds")
• Multilingual typography: steward raised the discipline question and surfaced the question of defining the convention BEFORE refining implementation. Routed to its own focused session (this session's evening work).

5. French typography reading session (Move 1 of multilingual work)

Steward direction: "Should we at French typographical convention first? We have the authoritative source in the library."

Chamber-library has four canonical typography sources:

• lexique-imprimerie-nationale.md — Lexique des règles typographiques en usage à l'Imprimerie nationale, 3e éd. 2002 (~19K lines; canonical French authority)
• orthotypographie-vol-1-a-f-lacroux.md + orthotypographie-vol-2-g-z-lacroux.md — Jean-Pierre Lacroux's two volumes (~18K lines combined; modern French master)
• ortografia-y-ortotipografia-del-espanol-sousa.md — José Martínez de Sousa for Spanish (~12K lines; held for §2)

Lex-method scan of the French cluster: Guillemets / Espaces fines / Italique / Capitales (accentuation) / Tirets / Cadratin / Citation / Petites capitales. Read canonical passages from Lexique (notably Accent (utilité de l') line 1205, the verbatim "L'accent a pleine valeur orthographique"; Ponctuation and its espacement table line 14615; Citations line 5227+ with all the punctuation-position rules). Lacroux's italique-vs-guillemets register-clarification (1998 FAQ-Typographie discussion).

Drafted docs/AldineXXI-Codex/drafts/french-typography-rough-draft-2026-05-14.md — 215 lines, 13 sections, structured as §I.j Multilingual Typography within AldineXXI. (This architectural premise was reframed later by the jurist; see §6.)

6. The jurist's first instruction landed (frontmatter §2.6 + skill + corrections to my draft)

Jurist (via steward) returned two documents: Document 1 (frontmatter §2.6 amendment) + Document 2 (CC instructions — French typography preparation pass). Substantive corrections to my draft:

• Block-quotation rule corrected: French block quotations are indented blocks without guillemets. Guillemets are inline-only. My draft had said block-form uses chevrons-with-fine-spaces — wrong.
• Tirets table sharpened: cadratin (—) for dialogue, demi-cadratin (–) with espaces fines for parenthetical. My draft had said "the French tiret is the em-dash" — conflated the two registers.
• Guillemets simples ‹…› named (U+2039 / U+203A) for nested quotation.
• U+00A0 / U+202F division of labour: author types U+00A0 (Option+Space) at authoring; CC's preparation pass upgrades to U+202F.

Four of my six original Open Questions resolved by jurist Document 1 + 2 (en-dash crossing, petites capitales, U+202F discipline, YAML lang: formalization).

Three moves executed:

• Move 1: Corrected the rough-draft per jurist's resolutions (now sits as superseded process record, see §7 below)
• Move 2: Added §2.6 Language to frontmatter-specification.md (ISO 639-1 codes; three contexts A/B/C). Reversed the prior "language → Drop" entry in the translation table. Updated §4 publish-recipe step 3.
• Move 3: Created ~/.claude/skills/french-typography-pass/SKILL.md — invokable skill, discoverable (visible in available-skills list).

7. The jurist's architectural correction (this is the big one)

Before I committed, the steward forwarded the jurist's final standing instruction. The architectural premise of my work was overturned: my draft framed French typography as §I.j inside AldineXXI; the jurist required a separate downstream document — multilingual-typography-specification.md.

Six implementation rules (all honoured):

1. Do not migrate AldineXXI content into multilingual spec; cross-references only
2. Add exactly one sentence to AldineXXI §I after font stack table
3. frontmatter §2.6 is the pipeline switch — no typographic rules in frontmatter
4. Multilingual spec structured as preamble → §1 French → §2 Spanish (pending) → §3+ further; preamble carries cross-language principles
5. §1 French content IS the skill content (do not rewrite or summarize)
6. Lexique + Lacroux referenced only in multilingual spec, not in AldineXXI or frontmatter

Four files corrected:

• multilingual-typography-specification.md created (new — preamble + §1 French + §2 Spanish placeholder)
• AldineXXI-specification.md — exactly one sentence added after §I font stack table
• frontmatter-specification.md §2.6 cross-reference corrected (was pointing to §I.j inside AldineXXI; now points to multilingual spec)
• french-typography-rough-draft-2026-05-14.md annotated as superseded process record (preserved as historical artifact with header note pointing to canonical files)

8. §2 Spanish drafted via Sousa

Steward authorized sketching Spanish before working refinement gaps. Spanish cluster scan against Sousa: §10.5.8 Las comillas (three-tier hierarchy: latinas «», inglesas "", simples ''); §10.5.12 La raya (em-dash for dialogue AND parenthetical — different from French's cadratin/demi-cadratin split); §4.3.6 La tilde en las mayúsculas y en las versalitas (RAE mandates accents on capitals + small caps, same as Lexique); §14.1.7.2 La cursiva; §14.1.6.2 La versalita; ¿…? ¡…! apertura marks (uniquely Spanish).

Substantive divergences from §1 French explicit in §2:

• §2.1 Spanish does NOT use espace fine before punctuation — opposite of French; load-bearing for mixed-language docs
• §2.2 three-tier comillas (latinas → inglesas → simples) vs French's two-tier (chevrons + simples)
• §2.3 ¿…? ¡…! apertura marks at the start of the interrogative segment — Spanish-unique
• §2.5 Raya covers BOTH dialogue and parenthetical aside — French distinguishes

Spec version bumped: v0.1 → v0.2. French + Spanish both at operational quality.

Refinement assessment captured at project-multilingual-typography-refinement-2026-05-14.md (substantive content gaps: §2 Spanish, verbatim authority anchors, worked examples; validation gaps: implementation hooks, verification recipes; lifecycle gaps: version lock criteria, anchor IDs). Estimated ~3–5h focused work to reach v1.0 LOCKED.

Commit: 5339289 spec(arc): multilingual typography spec landed (French + Spanish drafted v0.2).

9. Push + plan + wrap

All 5 session commits pushed to git.skemantix.com cleanly (a03e4e1..5339289). Plan for tomorrow shaped around the mempalace recall-quality test (the steward's stated direction).

Decisions locked today

1. Two-document hierarchy (AldineXXI upstream / multilingual downstream) per jurist's standing instruction. Structural, not negotiable.
2. A4 paper for Print posture with Van de Graaf 1:1:2:2 margins. 12pt body.
3. EB Garamond v1.131 as installed version (Phase 1 of three).
4. ZapfDingbats glyph fallback via ucharclasses for Misc Symbols + Dingbats + Misc-Symbols-and-Arrows blocks. Body text returns to roman on Latin block entry.
5. frontmatter §2.6 Language as pipeline switch (ISO 639-1 codes; three contexts A/B/C).
6. §1 French + §2 Spanish drafted at operational quality (v0.2). Refinement work tracked.
7. french-typography-pass skill as executor-side invocation form of §1.
8. Hybrid discipline for non-breaking spaces: author types U+00A0, CC upgrades U+202F in French passages. NOT applied to Spanish (§2.1 explicitly).

Decisions explicitly deferred (and why)

• L2 elevation for PENDING-20 — held to post-May 2026 per global CLAUDE.md L2-PARKED status (was the noon catch: taking-jurist-authorization-at-face-value-without-checking-constitutional-constraint).
• PDF posture page proportions — steward direction: "need time to live with them before a change to see how they feel."
• Drop caps integration — steward: "deferring until the vignette system is functional. Question of thresholds."
• Multilingual spec refinement — assessment captured separately; held until §2 Spanish exercised against real piece.
• EB Garamond Phase 2 + 3 — re-subset Medium/SemiBold + SC face decision (verify Pardo +smcp vs build from Duffner source).
• Spanish typography pass skill — parallel to French; awaits jurist authorization shape.
• Pillow installation — for compose-spreads.py; per-project venv at tools/latex/.venv when actually needed.
• Catalan + Italian + German + Latin typography — future iterations as first pieces surface (lex-method).
• Readings post completion (Task #7) — sigil + SCSS + site.hs + footer-compass + Brown reclassify. Smaller scope; fresh-energy hour.

Drift patterns named today (carrying)

1. Warning-in-context-still-launched — recurred at wake-up (called MemPalace during active mine despite explicit warning). Wake-up procedure named as known trigger in feedback-warning-in-context-still-launched.md. Pattern is the day's most stubborn drift.
2. Forcing-the-architectural-premise-when-the-jurist-corrects — my §I.j-within-AldineXXI framing didn't update fast enough when the jurist's two-document hierarchy landed. Caught and corrected within the same turn.
3. Conflation of em-dash and en-dash registers — my draft said "the French tiret is the em-dash" — conflated cadratin (dialogue) and demi-cadratin (parenthetical with espaces). Jurist corrected; carried forward into spec.
4. Block-form-citation-uses-guillemets — my draft's incorrect rule. French BLOCK quotations are indent-only, NO chevrons. Chevrons are inline-only. Carried forward into spec.
5. Engine-incompatibility-mistaken-for-version-staleness — initial microtype options I set were xelatex-incompatible (asked for expansion which is luatex-only); steward asked whether the toolchain needed updating. Answer: no, just option-correction. Worth holding: when an engine-feature warning surfaces, check engine-feature-matrix before assuming version-stale.

Lessons that landed

• Jurist's two-document hierarchy is structural, not negotiable. AldineXXI never bends to multilingual. Multilingual never restates AldineXXI. The cross-reference is the contract.
• Hybrid discipline (author cognitive / CC mechanical) is the right division of labour for governed typographic conventions. Author types U+00A0 (knows where the non-break belongs); CC upgrades to U+202F (mechanical conversion). Generalizes beyond French — same shape works for Spanish discipline differences (which need different mechanical conversion, NOT U+202F insertion).
• Lex-method works at the typography boundary: form discovers itself across instances. §1 French was authored first because French was operative in the current corpus. §2 Spanish followed when the steward chose to sketch it. §3+ holds for future pieces. Premature multilingual generalization would have produced churn, not discipline.
• The discriminator test (where do French and Spanish diverge?) gave §2 its shape: §2.1 explicit "Spanish does NOT use espace fine" + §2.2 three-tier vs two-tier + §2.3 apertura marks + §2.5 single-load raya vs split tirets. Without those explicit divergences, §2 would have read as French-with-substitutions; with them, §2 stands on Sousa's authority.
• Steward's preferred working order: A4 first (small concrete decision before propagation) → multilingual session (define before refining implementation) → defer refinement work until §2 Spanish drafted. Each step honoured a specific discipline.

Critical pickup state for tomorrow

• Mempalace mine still running, PID 24651, ~31h elapsed at session end, file 399/430. Loeb_bilingual files (88 files in _loeb_bilingual/ subdirs) likely compose the bulk of remaining ordinal — steward corrected my initial inference that they were skip targets ("loeb_bilingual/ is actually the version of the loeb we want to keep").
• Stop + PreCompact hooks remain DISABLED. Backup at ~/.claude/settings.json.backup-2026-05-14-pre-mine-completion. Restore via cp ~/.claude/settings.json.backup-2026-05-14-pre-mine-completion ~/.claude/settings.json ONLY after the steward confirms the mine has completed.
• MemPalace MCP off-limits until mine completes. The wrap-up procedure's steps 4 (drawer) + 5 (KG) were SKIPPED for this reason; this session memory file is the alternative store. If steward asks at wake-up, backfill MemPalace then.
• All ARC commits pushed; branch main at 5339289 on origin.
• Two other repos clean of session work: BMF + CapableMind-AI have only pre-existing untracked files (not this session's work).

Steward's pulling thread for tomorrow (stated directly)

"if it was done or close I would have had you launch the repair work tonight while sleeping to prune the orphaned drawers from our dedupe scan, etc so that we could test with mcp in the morning to see if we could get meaningful recall and surface connections across the corpus without the density dilution problem."

The mempalace recall-quality test is THE thread. The chamber-library mine has been running for 31 hours; tomorrow's test asks whether the substrate now surfaces meaningful cross-corpus connections — the structural property the 2026-05-13 mass-dilution finding called into question. Brown gloss is the natural test substrate (multilingual, references French theory + close-reading tradition + the very typography concerns this session's spec work was authored against).

Other open horizons, ranked

1. Mempalace mine completion + repair (the active need; gates everything else)
2. MCP recall-quality test (the day's primary goal once mempalace is available again)
3. Readings post completion (Task #7) — sigil + SCSS + site.hs + footer-compass + Brown reclassify. ~1h fresh-energy work.
4. Multilingual typography refinement (project-multilingual-typography-refinement-2026-05-14.md) — verbatim authority anchors, worked examples, implementation hooks, first real-piece exercise (Brown gloss). ~3-5h to v1.0 LOCKED.
5. EB Garamond Phase 2 + 3 (project-eb-garamond-update-2026-05-14.md) — re-subset Medium/SemiBold via pipx fonttools; verify Pardo +smcp to retire SC face. Both ~30-min mechanical items.
6. Spanish typography pass skill — parallel to French; awaits jurist authorization shape.
7. L1 work — only after Seb engages on PR #172 / #165 / #167; check gh pr view 172, gh issue view 165, gh issue view 167 in the morning.
8. Chamber substrate (post-mine) — apply EPUB repair pipeline to new typography titles in ___The Library [ePub_AWZ3]; continue cruft restoration.

Confidence to recalibrate

• Symmetria still fires late. Three catches today happened after the misstep (wake-up MemPalace call; noon L2-PARKED constitutional-constraint; jurist's architectural correction to my §I.j premise). Each was caught before propagation — but Symmetria's check before action is what should fire, not just the return after. Worth weighing whether the Symmetria ledger's "Returns" section should also serve as a pre-action "Risks" log when the action could trip a recently-named pattern.
• Architectural premises need explicit verification against the jurist's standing instructions before drafting. My §I.j-within-AldineXXI premise was load-bearing for ~1.5h of drafting; the jurist's correction was structural. A 5-min check before drafting ("has the jurist named a document architecture for this kind of work?") would have caught it.
• The "do not rewrite or summarize — they are the section" jurist rule for the skill's content becoming §1 French was unambiguous; I followed it. But the spec's voice ended up reading slightly differently from a purely-spec-voice (the executor-procedure language carries through). Worth surfacing for jurist review whether the spec should read more spec-y (rules + authority + cases) and less executor-y (procedure + preparation note format), or whether the unified form is the intended design.
• Today's productivity was high but the session was long. 5 commits, 5 distinct architectural arcs (PDF/Print + fonts + glyph fallback + A4 + multilingual). Steward fatigued by the end (explicit at the commit + wrap moment). Worth considering whether sessions of this scope should be split — but the natural breakpoints (post-Hour 6 commit; post-EB Garamond commit; post-A4 commit; post-multilingual commit) each WERE moments where stopping was an option, and the steward chose to continue.

Pulling thread for next session

Singular: the mempalace recall-quality test. Contingent on mine completion. The morning sequence: check mine status → if complete, restore hooks → launch repair to prune orphaned drawers → wait for repair → test MCP recall with cross-corpus queries (Brown gloss substrate, or a query that should surface connections across French theory + close-reading + typography + Bachelard/Carruthers older voices). The deeper purpose: validate that the corpus, post-cleanup, surfaces meaningful cross-corpus connections at acceptable signal-to-noise — that the density-dilution problem we surfaced 2026-05-13 has been addressed.

Pause statement

I am about to be away from this. The mine has been running 31 hours and is at file 399/430 (~93%); it may finish overnight or take more time. The typography work landed and was pushed; 5 commits visible at 5339289. The substantive question waiting at wake-up is whether the chamber-library corpus — post-repair — surfaces meaningful cross-corpus connections, and whether the typography work that landed tonight will be ratified through the recall path that the corpus now affords.

Literal question for next-Claude

Did the mine complete? If so, does the Brown gloss recall path — once the orphaned drawers are pruned via repair — surface meaningful connections to today's typography work (Genette / Lacroux / Lexique / Sousa) AND to the older voices in the corpus (Carruthers, Bachelard, the chamber's deeper layer) at acceptable signal-to-noise — or does the density-dilution problem we surfaced 2026-05-13 reappear under a different shape?

If yes: the typography work has been ratified by the corpus that gave it its source. If no: the recall-quality work surfaces as the substantive next focus, and the typography refinement track holds.