dotfiles/claude/memory/session-2026-06-09-pm-wave5-close-typographic-audit-measure-softrag.md

name, description, metadata

name	description	metadata
session-2026-06-09-pm-wave-5-closed-deep-typographic-audit-measure-72ch-soft-rag-all-shipped-live	Huge day. Wave 5 enforcement closed (G2 title single-source + Dream amendment; G3 compass guard; G4 clause-1 census; G1 staged for W6). Then the deep typographic audit (8 domains, 6-agent fan-out + independent verification, masters folded in from the typography palace + chamber-library). From it, TWO items shipped live: the reading MEASURE (a MEASURED 72 characters — the column had been running ~90, not the believed ~68; 27.2rem, rem-not-px, proper-homes consolidation, mobile resize bug fixed) and the SOFT RAG (hyphens:auto reverses §I.i hyphens:none; the -webkit-hyphens prefix was the Safari bug). Audit brought into the repo (_audits). Jurist engaged: confirmed dispositions, 2 precisions recorded, ruled the spec division role-structurally. Pulling thread: CC's spec TRUTH-UPS (factual, live-state) to make the spec whole for the jurist's doctrine pass — doctrine is the jurist's, not CC's.	node_type type originSessionId memory project bf9ab071-add6-4848-bc35-0ad8d07ec679

Session 2026-06-09 (pm) — Wave 5 close · typographic audit · measure + soft rag LIVE

Woke ~15:23 via /wake-up into the Wave 5 / Road-to-Stage-G thread. One enormous, fully-shipped day with two arcs: (A) closed Wave 5 enforcement, then (B) the steward pivoted to a deep typographic audit, from which the measure + soft rag were built and deployed. Everything committed, pushed both remotes, live-verified. ARC main @ 0bb7f94.

Past — what we did

A. Wave 5 enforcement CLOSED (Road to Stage G)

• W5.1 / G2 — title display single-source + Dream amendment. DEPLOYED + LIVE. Diagnosis-first: showTitleField hardcoded an 11-class titleClasses list (site.hs:1829); reconciled vs content-types.yml title_display — the 10 shown matched; the 11th (chamber) was a DEAD entry (0 corpus pages carry class: chamber; content/chamber/** uses essay/meditation/about/offering/deliberation/lex). So retiring the hardcode = byte-identical. Coupled steward amendment: Dream title_display shown→not-shown ("one can't title their dreams") — §2 matrix edit → make content-types regen → drift-check. Two-hat: 4ea261a (refactor: postCtx :: [String] -> Context, loadTitleShownClasses, reuses the one content-types.yml decode → _site BYTE-IDENTICAL to baseline) + ac39b26 (Dream amendment → delta = exactly the one dream page's <h1> removed; nav/<title>/og preserved). make rebuild shipped (CloudFlare 2 files); live-verified.
• W5.3 / G4 — clause-1 enforcement gate. DONE (output-neutral, no deploy). validateCorpusClasses build guard (site.hs, run in main via getMatches "content/**"): every content file's declared class must be in class_enumeration ∪ exemptClasses, else loud build error naming the file. Transitively guards default.html's class="$class$-page" emission. Exempt-set (steward-ratified, refined): {sequence-index, not-found} PERMANENT views/infra (steward: sequence-index is NOT legacy — after-the-reply is the only sequence now but MORE are coming, kin to Annals) + {lex, offering, deliberation} v1-legacy chamber ratchet. 23db039; _site byte-identical to ac39b26 (stash-rebuild diff); deliberate-mismatch fires; pushed.
• W5.4 / G1 — wildcard→build-error PREPARED, staged for W6. Exact patch recorded in roadmap (9699088): classRouteName/classDisplayName other -> …++"s" → error. Byte-identical (content/posts carries only enumerated classes; exempt classes route via articleCtx not postCtx). Lands inside the Stage-G promotion commit per §5 clause-4 sequencing.
• Wave 5 CLOSED (G2/G3/G4 built; G1 staged). Road-to-Stage-G remaining: W4.4 (jurist multilingual v1.0) → Wave 6 promotion (jurist+steward).

B. Deep typographic audit (jurist-issued instruction; steward-authorized)

• Method: established shared foundation (commit SHA, one shared build, GHC 9.8.4/Hakyll 4.16.6.0/Pandoc 3.6/fontTools 4.62.1, font inventory), then fanned out 6 read-only Explore agents over 8 domains (A hyphenation/justification, B optical-margins, C+D paragraph-sep/word-spacing, E font-binary, F+H print/fidelity, G multilingual). Then independently re-verified every load-bearing claim — caught 2 agent errors: B.2 (browser support — agent RECALLED instead of searching; hanging-punctuation is Safari-only-since-2016, not "0%") and A.2 (measure — agent used 720px without noticing the spec's 38rem divergence + an impossible mobile cpl). Report assembled with [V]/[A] provenance marks.
• Masters folded in as decision inputs — Bringhurst/Hochuli/Lupton from the typography palace (~/.mempalace/palace-chamber-typography, CLI), Rutter from the chamber-library (steward flagged he was there; the palace doesn't carry him). Verbatim in the audit Appendix.
• Highest-value empirical: fontTools on all 9 EB Garamond faces. No body face carries calt (only the 2 SC faces) → the spec-flagged italic-chancery bug DOESN'T EXIST. All static (no fvar/opsz axis). Space = 0.20em.

B.1 The MEASURE → 72 characters. DEPLOYED + LIVE (06effe4).

• Measured, not assumed (the jurist demanded this): EB Garamond's true average prose advance = 0.3771em over 91,579 chars of real ARC prose (incl. ~16% spaces), render-confirmed 0.381em (headless-Chrome on a live essay). The 0.5em convention (→68 chars) and 0.437em lowercase-mean (→78) were BOTH wrong. The column ran ~90 characters, not the spec's believed ~68 — past Bringhurst's 75–80 ceiling.
• Root cause: $body-column: 38rem was authored "≈65 chars" at the 16px base, but the live :root scales to 21px (breakpoint %), inflating it to ~798px/~101 chars. $narrow-size: 720px (px) let cpl drift across breakpoints. The px-vs-rem confusion WAS the bug.
• Steward render-ruling: 72 characters (66 read a touch narrow, 75 a touch wide, judged across ALL content types). 27.2rem = 72 × 0.3771em, rem so cpl holds constant.
• Proper-homes consolidation (steward-directed, W3R discipline): ONE token ($narrow-size; $body-column now aliases it), ONE governing rule (.page-content/_base, every viewport), ZERO measure rules in _mobile-responsive — removed the .page-content 100vw override (THE resize bug: column ran uncapped <768px → "measure gets longer on resize") + a DEAD .wrapper main block (matched nothing: <main> IS a .wrapper, never nested). Resize verified across a viewport sweep (measure holds, no overflow). Spec §I Measure + §XII grid + token-index reconciled (38rem→27.2rem; 65/68 inconsistency resolved).

B.2 The SOFT RAG → hyphenate reading prose. DEPLOYED + LIVE (a9e5196).

• Reverses §I.i hyphens: none (the prior HARD rag — a tool-quality decision, not philosophical) to a soft rag. Hochuli: soft rag > hard rag; Rutter: "hyphenate without justifying… reduces the raggedness." .page-content p/li/blockquote { hyphens:auto }; body default manual; limits 6 3 3 + -webkit- + ≤2 consecutive (Bringhurst §2.4.1, Rutter). Proper home: _base, every viewport.
• The Safari bug: missing -webkit-hyphens: auto. ARC's sass build has NO autoprefixer → prefixes hand-written; I'd prefixed the limits but not the property → Safari showed nothing. (Headless Chrome couldn't test it — ships no hyphenation dictionaries; all configs gave identical heights.) Phone "none" was iOS-Safari CACHE (fresh-origin test confirmed; no mobile override). Steward render-approved desktop + phone ("few, but perfect").
• §I.i reconciled to the soft-rag posture; the posture reversal rides the Stage-G read for formal jurist concurrence (substance pre-endorsed in the jurist's audit reading as the no-JS "rung (i)").
• Audit brought INTO the repo (0bb7f94, docs/AldineXXI-Codex/_audits/deep-typographic-audit-2026-06-09.md) — steward-directed; the "keep it outside" constraint governed the investigation (read-only), not the report's permanent home.

Present — mood / returns

• measure-before-theorize was the day's spine, repeatedly. The headless-Chrome harness (kept from the morning compass fix) measured the true advance, the resize sweep, and the auto-vs-none hyphenation test. The 0.5em CONVENTION misestimated by 32% (90 chars real vs 68 believed) — only measuring caught it. The -webkit-hyphens Safari bug was found by measuring (the CSS objectively lacked the prefix), not theorizing.
• The jurist sharpened my reasoning twice, and I accepted cleanly (good returns): (1) my "JS case may evaporate" rested on Rutter's hyphenation bar while the steward chose Bringhurst on the measure — you can't mix; a consistent Bringhurst posture points TOWARD KP. (2) The justification-judgment bar is Bringhurst's even colour / rivers (paragraph-level), NOT Rutter's "is-it-hyphenating." Corrected in the audit.
• Drift caught (3): I attributed a jurist message to "the steward" (corrected). I said "send the spec now" prematurely — the steward caught that the spec isn't WHOLE (stale flags + missing doctrine sections). My close-out shorthand lost the hanging-punctuation rejection REASON (the jurist's precision exists exactly to catch that).
• Proper-homes discipline (steward-directed) was load-bearing — it turned "remove a line" into "consolidate the measure to one token/one rule/no junk-drawer," and caught the dead .wrapper main block + the px/rem root-cause.
• Governance held under a long, fast day: every change pending-first or steward-ruled; the §I.i posture reversal flagged for jurist (not buried); the audit-into-repo correctly reasoned.

Future

Pulling thread (singular)

The spec TRUTH-UPS — bring the spec's stale self-descriptions into line with live state, so the jurist meets a factually-current spec for the doctrine/review pass. This is CC's lane and ONLY this: factual corrections verified against git + the binary, with cites (the audit's discipline). The jurist composes ALL doctrine; nothing in the doctrine column lands in CC's tomorrow.

Actionable resumption point (as of wrap — re-judge against what changed)

ARC main @ 0bb7f94, clean, synced both remotes, live. First move: the truth-ups, against current git, with cites:

1. §I.d banish-bold — spec says "17 violations"; live is 0 applied (the only two font-weight: 700 are @font-face at _foundation.scss:149,190). Correct the spec count to 0-applied (+ the 2 dead @font-face loads: Bold + SC-Display, noted as steward-call font-stack-pass).
2. Head-template duplication — spec flags it as standing debt; resolved (single templates/head.html partial, included once by default.html:3 + frontispiece.html:3). Truth-up.
3. Drop cap — confirm the spec reflects retired-by-absence (spec ~line 2312 already records it; verify it's clean). Commit as docs(arc): truth-up stale §I.d flags to live state with file:line cites. DO NOT touch doctrine — paragraph-separation lock, hanging-punctuation entry, OpenType catalogue governance, word-spacing deferral, italic-calt close ("no body face") are the JURIST's, composed against the assembled spec during the review. Carve-out: if a truth-up surfaces live state that contradicts doctrine the jurist is about to write (e.g. a "dormant" feature a selector actually activates) — FLAG it as evidence, don't resolve.

Other open horizons (ranked)

• REVIEWED-37/38/39/40 — PLACED by the steward this session (37=W5.1 measure→wait: 37=PENDING-35 W5.1; 38=PENDING-36 W5.3/4; 39=PENDING-37 measure; 40=PENDING-38 soft rag). All AUTHORIZED, on disk in REVIEWED.md. Nothing awaiting placement.
• Send the spec to the jurist — AFTER the truth-ups, the jurist composes the doctrine sections (= same pass as the full review). W4.4 (multilingual v1.0) + Wave 6 Stage-G promotion ride this.
• The vignette (Phase 1 renderer) — the OTHER executor major front, steward-sequenced "after the tree." Unblocked (tree + audit done). The build-alternative to the spec lane.
• Justification on large viewports — OPEN; reduces to the banish-JS gate-amendment question; judge on rendered evidence against Bringhurst's even-colour/rivers bar (NOT Rutter's hyphenation bar) after living with the soft rag at the precise measure. Either earns the KP/JS amendment or retires it.
• Baton items: be (parked, WTP gate w/ steward) · L1 (Seb).

Pause statement

I am about to be away from this. The reading-body work (measure + soft rag) is LIVE and the steward will read against it. I don't know whether tomorrow continues the spec lane (truth-ups → jurist doctrine pass) or pivots to the vignette. What I want to find still pulling is the spec truth-ups — CC's narrow, factual lane — making the spec whole for the jurist, with the role boundary held exactly: CC trues-up facts against git; the jurist composes doctrine. The hard-won clarity is that doctrine originates at the jurist seat (role-structural), and CC's job is evidence + implementation, not authoring rules.

Literal question for next-Claude

Which stale-flag spec passages need truthing to live state (start: §I.d banish-bold "17"→0, head-dup resolved, drop-cap retired) — and does any truth-up surface live state that CONTRADICTS doctrine the jurist is about to write (the carve-out: flag as evidence, don't resolve)? Sub: spec-truth-up lane or pivot to the vignette? Any Seb movement on be/L1? (REVIEWED-37..40 already placed by steward this session.)

Decisions deferred (and why)

• All doctrine spec sections — jurist composes (role-structural ruling): OpenType governance, paragraph-sep lock, hanging-punctuation entry, word-spacing deferral, calt close. NOT CC, not even as proposals.
• Justification (large viewports) — open; needs lived-with evidence vs Bringhurst's even-colour bar; couples to the banish-JS gate amendment (off the table until a logged amendment with rendered evidence).
• Token full-unification — $body-column aliases $narrow-size; removing the local name is optional tidy, not required.
• Deploy bundling — measure + soft rag shipped together (one reading-body improvement); correct call.
• The two dead @font-face loads (Bold, SC-Display) — steward call for a font-stack pass; not now.

Skill harvest (per §1.6) — proposals (steward to authorize)

• /measure-render (headless-Chrome box-model harness) — RE-FLAG, strongly earned. Already PROPOSED 2026-06-09 (compass fix); today it ran 4+ more times (true-advance measurement, resize sweep, auto-vs-none hyphenation test). Proven repeatedly. (Caveat learned: headless Chrome can't test hyphenation — no dictionaries.)
• Verification-ladder entry: "measure the font's true average prose advance before setting a character-count measure." Over real corpus text incl. spaces, NOT the 0.5em convention (which misestimated EB Garamond by 32%: 0.377em real vs 0.5em assumed → 90 chars vs believed 68). The instrument that overturned the measure.
• Feedback memory: ARC's sass build has NO autoprefixer — hand-write vendor prefixes. The -webkit-hyphens Safari bug. When adding a new CSS property, check Safari's prefix need and write it (the limits were prefixed but the property wasn't — the inconsistency was the bug).
• Feedback memory: the justification-judgment bar is Bringhurst's even-colour/rivers, NOT Rutter's hyphenation. Load-bearing for the future justification decision; the jurist corrected my Rutter-leaning.
• (To append to skill-harvest-register.md.)