dotfiles/claude/memory/session-ledger-2026-04-28.md

---
name: Session Ledger 2026-04-28
description: Practice-of-return ledger maintained by /symmetria — returns, open horizons, recalibrations, authorization moves, sub-agent dialogues, bypasses.
type: feedback
originSessionId: e5a1fa7c-cae6-49c4-8678-e27ce379639b
---
# Session Ledger — 2026-04-28

## Hakyll caveats noted

- 2026-04-28 — **`match "scss/main.scss"` doesn't track SCSS import dependencies.** Changes to `_base.scss`, `_sidenotes.scss`, etc. don't trigger Hakyll to recompile main.scss. Workaround: `touch scss/main.scss` before `make build`. Worth either documenting in ARC's CLAUDE.md as a build gotcha OR fixing in `site.hs` to match all `scss/**/*.scss` for change tracking. Small DX improvement — separate when convenient.
- 2026-04-28 — **No local-only build target in the Makefile.** Every `build*` target auto-deploys. To rebuild without deploying: invoke `stack exec animal-davidglidden-eu build --allow-different-user` directly. Steward authorized one-by-one builds in low-traffic context. Adding `make build-local` would close the gap.

## Returns

- 2026-04-28T~afternoon — **Cul-de-bouteille surfacing caught.** Steward asked for "items 2 and 3 above" (Lex transcription + cul-de-bouteille). Yesterday's record explicitly deferred cul-de-bouteille to "its own focused session" + the wake-up's literal question asked whether it surfaces earlier. Instead of executing both, surfaced the tension: Kai-lunch fragmented context probably isn't a design-session shape. Asked whether item 3 was list-position reference or actual now-call. Symmetria-fittingness applied: design work in fragmented attention is not proportionate.
- 2026-04-28T~afternoon — **After the Reply source location — wrong path claimed.** I located the essays at Obsidian `10. Ready to publish/After the Reply/`. Steward corrected: canonical source is iA Writer at `/Users/davidglidden/Library/Mobile Documents/27N4MQEA55~pro~writer/Documents/Essay Sequence on Reply, etc/`. Same drift family as `canonical-library-not-checked-first` and `working-from-mental-summary` — claimed-without-checking-canonical-edit-surface. Project memory corrected. Cousin pattern, not new shape.
- 2026-04-28T~afternoon — **iA Writer `+++` mis-read.** I read the `+++` separators in `After the Reply, Sequence.md` as iA Writer's transclusion / sequence-compile syntax. Steward corrected: *"the +++ in iA indicates page-breaks."* Same drift family as `bibliographic-claim-without-source-verification` (2026-04-25) — characterizing source-syntax from inference rather than checking iA Writer's actual specification. Project memory corrected. Restraint lesson: when reading a steward source-file in an unfamiliar tool's format, name the inference as inference, don't ship as fact.
- 2026-04-28T~afternoon — **Small-caps register error caught.** I proposed *"II of V — After the Reply"* in **small caps** as the per-essay sequence-marker. Steward flagged: *"I don't think we use small caps like that, let's confirm."* Spec confirms steward's intuition: AldineXXI §I (line 46) — apparatus (datelines, captions, locators) renders in IBM Plex Mono, NOT EB Garamond small-caps. Reading-layer small-caps in this codebase is for epigraph attributions + register markers within reading content, not for apparatus locators. A position-marker is apparatus by nature. Cousin to the canonical-spec-not-checked-first family — characterized typographic register from general design instinct rather than reading the spec's stated apparatus/reading-layer distinction first. Project memory corrected with two new shape options (apparatus block vs Roman numeral in H1).
- 2026-04-28T~afternoon — **Blame-the-exotic-before-checking-the-boring (recurring shape).** Two instances same afternoon: (a) sidenote layout — blamed `:has()` + subgrid browser support; actual cause was `.page-content { max-width: 720px }` cascade + my own `:has()` rule's specificity ambiguity. (b) old-style numerals not rendering — blamed font-subsetting stripping ONUM; actual cause was browser cache not picking up the new CSS rule (the rule was in deployed style.css; verified with grep). Same shape: when an empirical "X doesn't work" comes in, I default to the most exotic possible cause (browser feature gating, missing font feature) instead of checking the boring/likely path (cascade order, cache state). **Discipline:** before hypothesizing "feature missing / unsupported," verify the boring causes — inspect computed styles, hard-refresh fully, check that the rule is actually in the deployed CSS. Cousin to the contamination flag pattern (component self-assessment accepted at face value); here it's "my own hypothesis accepted at face value."
- 2026-04-28T~evening — **Steward call on register-mark form**: drop the trailing period. *"does the roman numeral need the period?"* My read: no — the period was load-bearing in the H2 form (`## I.`) to signal section heading; in register-mark anchor form, it's decorative. Mark IS the anchor; context provides what it anchors. Steward: *"yes please."* Sed across all 5 essays. Now `I` / `II` / `III` etc., bare Roman numerals. Sidesteps Pandoc list-detection entirely AND reads cleaner in marginalia.
- 2026-04-28T~evening — **Register-mark alignment fix**: margin-top: 0 (was 1rem from my earlier preemptive fix). My preemptive fix was based on the sidenote/blockquote case (blockquote has `margin: 1.5rem 0`); register-marks anchor to paragraphs which have `margin-top: 0` in `.post-content > p`. Steward verbatim: *"This is my fault for having asked you to preemtively make a fix based on the right rendering sidenotes that had blockquotes and were therefore too high."* (Steward's framing of fault is generous — the preemptive fix was my move, but the diagnosis is what matters.)

## Section-break-glyph-or-whitespace consultation (2026-04-28 evening)

Steward: *"the spec, section II also outlines a spacing scale. So the choice is either hedera or space, but I honestly don't know what the better choice is — We may need to deep deeper into our typography folder for the answer."*

Walk findings (chamber-library `converted_texts/collections/typography/`):

- **Aldine tradition (Margolis on Manutius)**: *"published with much white space and without commentary."* The Aldine norm was generous whitespace; the leaf was occasional and festive, not structural between sections.
- **Bringhurst §4.1.4 (Mark each beginning and resumption of the text)**: lists fleurons as one method among several (alongside small caps, versals, drop caps) for marking section openings. His "fanfare" prescription kicks in when no heading marks the section AND the page is busy. Our register-marks ARE the opening-markers Bringhurst's list calls for. Doesn't mandate hedera between sections.
- **Tschichold (*Form of the Book*)**: doesn't directly address; overall posture is restraint, "cut don't add" — consistent with the Harrison Knife direction.
- **Lupton**: slightly skeptical of white-space-as-virtue (cites Tufte's density argument); context is information design not literary essay; her concrete inter-paragraph advice — *"skipping a full line often creates too open an effect"* — points toward precise tokens, which the §II spacing scale provides.
- **Manutius the Younger (*Orthographiae Ratio*)**: covers punctuation hierarchy (comma/semicolon/colon/period), not section-break ornament-vs-whitespace.

**Synthesis:** The spec's existing answer (§IV Ornament Taxonomy revised 2026-04-22: "*Section breaks within a piece … use Separation-level silence per §II — whitespace alone, no glyph. Structural markers (Roman numerals, headings) do the reader's orientation work; silence completes it.*") is well-grounded in the formative voices. Aldine tradition supports it; Bringhurst's permissive frame doesn't contradict it; the Harrison Knife consolidation is faithful to it.

**Gap surfaced**: typography corpus doesn't deeply address the *specific* case of "section-breaks-with-margin-markers" (most discussion is chapter-opening or paragraph-leading). Spec choice was made on inherited principles rather than a single explicit source on this exact question. Worth noting for the spec's provenance log at next spec-revision moment.

**Implementation recommendation pending steward go**: `.section-break` empty div between sections, height `$space-xl` (2rem) per §II Separation-level silence. Spec-compliant; preserves register-mark alignment; retroactively fixes Essays I + II per spec's migration note (`.ornament.section` divs were marked for retrofit 2026-04-22).

**Steward go (*"yes, and the spec annotation"*) → executed:**
- SCSS: `.section-break { height: $space-xl }` added to `_sidenotes.scss` (verified `$space-xl` token defined in `_variables.scss`)
- Markdown: 28 section-break divs inserted via Python regex across all 5 essays (counts match expected: 1+10+7+6+4)
- AldineXXI spec annotated: new `### Provenance — typography corpus consultation (2026-04-28)` subsection in §IV preserving the walk findings (Margolis/Manutius, Bringhurst §4.1.4, Tschichold, Lupton, Manutius the Younger), gap noted, implementation reference. Future spec-revision moments inherit this provenance.
- Build + deploy: 6 files (5 essay HTMLs + style.css) live.

**After the Reply Tier 1 publication essentially complete**: 5 essays public with register-mark form, section-breaks for inter-section silence, alignment correct. Pending Round 2 (sequence-index page + sequence-marker apparatus + RSS check). Pending Round 2.5 (corpus-audit retrofit of Essays I + II's existing `.ornament.section` and `.ornament.completion` per spec migration note).

## Preface apparatus added (§XII.c) — 2026-04-28 evening

Steward question: *"Essay 2s first block of text is a sort of preamble... are these not meant to be in italics according to Brinhurst?"* My read: yes — Bringhurst on Aldine register-set, *"roman was used for the main text and italic for the preface, headnotes, sidenotes and for verse or block quotations"* (cited in §XII Sidenotes line 542).

Steward decision: *"A please. All three, and for V, we can remove the Prologue header since it will be redundant."* — option (a) `:::preface` div with CSS italic; apply to all three preface-bearing essays (II, IV, V); remove Essay V's `### Prologue` H3 since the italic treatment makes the label redundant.

Executed:
- **`.preface` SCSS rule** in `_sidenotes.scss` — `font-style: italic` + Bringhurst flip (`em, i, cite, h1-h6` revert to upright per §3.5.2)
- **Essays II, IV, V** — preface paragraphs wrapped in `:::preface` divs
- **Essay V** — `### Prologue` removed; `### The Question We Cannot Answer` preserved as preface's title (renders upright per the Aldine flip)
- **AldineXXI §XII.c Preface** — new spec section with Bringhurst source, source pattern, treatment rules, distinction-from-neighboring-forms table (preface vs epigraph vs sidenote vs register marker vs heading), implementation status. Differs from §XII / §XII.b in being reading-layer (not apparatus); no `site.hs` extraction step needed.
- Build + deploy: 4 files (3 essay HTMLs + style.css).

**Apparatus vocabulary now**: §XII Sidenotes (right-margin recognition) · §XII.b Register Markers (left-margin threshold-anchor) · §XII.c Preface (italic prefatory body) · `.section-break` (inter-section silence). Three reading-layer apparatus + one whitespace primitive — all spec-grounded, all exercised in production.

## Round 2 executed (2026-04-28 evening)

After the Reply audience-building infrastructure live:

- **Sequence-index page** at `/essays/after-the-reply/` (`content/essays/after-the-reply/index.md`, new `templates/sequence-index.html`, new `site.hs` match rule). Renders title block + epigraph + Aldine italic preface (`:::preface`) + ordered listing of all 5 essays with originating dates.
- **Sequence-marker apparatus** on each essay (post.html partial; renders when `sequence` frontmatter present). Form per locked option (i): `III of V · After the Reply` in IBM Plex Mono, sequence-name links back to index.
- **`sequence_order_roman` field** added to all 5 essays' frontmatter (Python regex insertion: I, II, III, IV, V).
- **site.hs route gotcha**: first attempt missed `setExtension "html"` — file emitted as `index.md`. Fixed: `route $ gsubRoute "content/" (const "") `composeRoutes` setExtension "html"` (matches lares index pattern).
- **Build**: `make rebuild` (clean) used because site.hs changed. 74 files uploaded first round (template propagation), 1 file second round (post-route-fix).

## SCSS refactor (steward catch)

Steward verbatim: *"why in sidenotes.scss?"* Across the evening I had been adding unrelated blocks (`.preface`, `.section-break`, `.sequence-marker`, sequence-index styles) to `_sidenotes.scss` — whose own header declared its scope as §XII Sidenotes + §XII.b Register Markers. I was kitchen-sinking by proximity-bias.

Recovery:
- Split into three new partials: `_preface.scss`, `_section-break.scss`, `_sequence.scss` — each with proper header comment naming its concern
- Restored `_sidenotes.scss` to its declared scope
- Updated `main.scss` imports at slots 7.b, 7.c, 7.d (preserving §XII numbering visually with section comments)

Saved as durable feedback memory: `feedback-scss-file-organization-not-kitchen-sink.md`. Cousin to blame-the-exotic (same proximity-bias shape, different domain).

## Round 2 small fixes + completion (2026-04-28 evening end)

- **Sequence-listing dateline grid bug**: third grid item (`<span class="sequence-listing-dateline">`) auto-placed into column 1 of next row (the narrow 3rem numeral column), causing aggressive vertical-text wrapping (*"Wien, / December / 2025"*). Steward caught: *"the index has the apparatus information below each essays title, but this looks like a wrapping issue."* Fix: explicit `grid-column: 2` on dateline. Small lesson: when you have N children in an M-column grid where N > M, the (M+1)th child auto-places into column 1 of row 2, not column M of row 2. Use explicit `grid-column` for non-trivial multi-row layouts.

- **/essays/ promotion executed (Round 2 final task)**:
  - `site.hs` essays.html rule: `loadAll "content/essays/*/index.md"` injected as `sequences` listField in context (using `defaultContext` for the per-sequence context — exposes title, url, subtitle, tagline, etc.)
  - `templates/post-list.html`: conditional `$if(sequences)$` block above `<section class="posts">` renders sequence callouts (title link + subtitle + tagline)
  - `scss/AldineXXI/_sequence.scss`: added `.sequences-listing`, `.sequence-item`, `.sequence-item-title`, `.sequence-item-subtitle`, `.sequence-item-tagline` rules — quiet but distinct from the post-item listing below
  - Generalizes: any future sequence dropped into `content/essays/<slug>/index.md` automatically appears on /essays/

- **Round 2 final state**:
  - ✅ Sequence-index page at `/essays/after-the-reply/`
  - ✅ Sequence-marker apparatus on each essay (`III of V · After the Reply` linking back)
  - ✅ Sequences callout on `/essays/` listing
  - ✅ SCSS refactor (per-concern partials: `_preface`, `_section-break`, `_sequence`)
  - ✅ Spec annotations (§XII.c Preface; §IV Provenance subsection)

The After the Reply publication arc is complete — public, findable from essays listing, navigable as sequence, each essay knows its place. Audience-building infrastructure now matches content publication.

## Late-evening polish (2026-04-28)

- **Mobile rendering**: steward initially saw register-mark spacing, preface italics, and old-style numerals failing on mobile. I confirmed via grep + Python regex over compiled CSS that all three rules ARE in deployed style.css with no mobile-specific overrides. Steward verified via private browsing window: rendering perfect. Cache was the cause. Steward verbatim: *"I reloaded in a private browsing window and the css is perfect. sorry for that"* (no apology needed — cache issues are real and worth flagging fresh-eyes).
- **Sequence-item-kind apparatus marker** added to /essays/ callout — small `SEQUENCE` label above the title, IBM Plex Mono uppercase + letter-spacing in gray. Steward call: *"yes, a sounds good"* (option a from three I'd proposed).

## Font feature audit (2026-04-28 evening, via fonttools)

Installed fonttools: `pip3 install --user --break-system-packages fonttools brotli`. Inspected webfonts in `static/fonts/`:

- **IBM Plex Mono Regular (`IBMPlexMono-Regular.woff2`) GSUB features**: `aalt, ccmp, dnom, frac, numr, ordn, salt, sinf, ss01-05, sups, zero` — **`smcp` and `c2sc` NOT present**. Real small-caps not available for the apparatus face in this codebase.
- **EB Garamond Regular (`EBGaramond-Regular.woff2`) GSUB features**: `aalt, c2pc, c2sc, case, dlig, frac, hist, hlig, liga, lnum, locl, onum, ordn, pcap, pnum, sinf, smcp, ss01-07, subs, sups, swsh, tnum` — **both `smcp` AND `onum` present**. Confirms why old-style numerals render universally (the body rule `font-variant-numeric: oldstyle-nums` works because the webfont supports it) and why reading-layer small-caps treatments using `@include small-caps` render correctly.

**Future-direction call surfaced**: if real small-caps wanted in apparatus roles, two paths — (a) load Plex Mono's full feature set (need to source a non-subsetted woff2 OR use Google Fonts CDN), or (b) revisit §I's "apparatus = Plex Mono" rule for cases where small-caps register matters more than locating-not-dwelling feel. Worth a future spec call. Comment in `_sequence.scss` flags the swap-point.

## Round 2 complete + closing state

After the Reply publication arc (Tier 1 + Tier 2 audience-building) is fully shipped:
- 5 essays public with register-mark form, section-breaks, prefaces, sequence-marker apparatus
- Sequence-index page at `/essays/after-the-reply/`
- Sequences callout on `/essays/` listing with `SEQUENCE` apparatus marker
- AldineXXI spec extended: §XII.c Preface (new), §IV Provenance subsection (new)
- SCSS architecture: `_sidenotes.scss` returned to scope; `_preface.scss`, `_section-break.scss`, `_sequence.scss` split out per-concern

**Pending tomorrow** (composition territory + small adjacents):
- Hearth landing prose + About *Hic* + About *Adsum* (steward composition; deferred from today per the wake's pulling thread)
- RSS check (jurist's "probably already half-exists")
- Round 2.5 corpus retrofit (Essays I + II legacy `.ornament.section` and `.ornament.completion` per spec migration note; Essays III/IV/V missing closing Close ornament)
- Refinements: register marks in faded red (rubrication, Bringhurst), Essay V H3 sub-section question, sidenote alignment refinement, real small-caps in apparatus (font feature path)

## Future-refinement notes

- **Register marks in faded red (rubrication)** — steward 2026-04-28 evening: *"by refinements, I was wondering if the register marks shouldn't be in a faded red color like Bringhurst mentions historically — but this is for another day."* Bringhurst §4.1.4: fleurons and versals "often printed in red, the typographer's habitual second color." Register marks are direct heirs of medieval rubrication translated through the Aldine and into AldineXXI. A faded-red treatment would honor the historical color-coding without being assertive; would also distinguish register marks from other small-caps elements (epigraph attributions, etc.). Defer to focused typography session; potential `--register-mark-color: var(--rubric-red)` token; respect dark-mode treatment.

- 2026-04-28T~evening — **Pandoc Roman-numeral list-detection gotcha caught (after deploy).** I authored register marks as `:::register-mark\nI.\n:::` for all 5 essays. Pandoc interpreted `I.` inside the fenced div as the start of an ordered list (`<ol type="I">`), not as literal text. Result: register marks deployed but invisible — empty `<ol>` elements where the Roman numeral should be. Steward caught: *"don't see the register marks."* Steward's clean fix when asked back: drop the period entirely (the period was load-bearing in the H2 form `## I.` to signal section heading; in register-mark anchor form, it's decorative — the mark IS the anchor; context provides what it anchors). Same shape as the blame-the-exotic-before-checking-the-boring pattern (didn't anticipate Markdown's standard list-detection behavior; would have been visible in any Pandoc test). For future register-mark work: use plain Roman numeral without trailing period (`I` not `I.`) — sidesteps Pandoc list detection AND reads cleaner in marginalia.

## Executor work — After the Reply publication (Tier 1 in progress)

- 2026-04-28T~evening — **Five essays in register-mark form deployed** (with the Pandoc-list bug; awaiting period-drop fix per steward call):
  - Essay I — `2025-06-17-the-ethics-of-the-reply.md` — register marks + sequence_slug + revisions entry logging the register-mark conversion
  - Essay II — `2025-07-11-the-ethics-of-the-reply-part-ii-on-abdication-and-fidelity.md` — title trimmed `"On Abdication and Fidelity"`, register marks across 11 sections, followed_by → `after-the-reply-iii`, sequence_slug, second revisions entry logged
  - Essay III — `2025-12-10-after-the-reply-iii.md` — new (Wien); "Noli timere" + Leopardi epigraphs; 8 sections in register-mark form
  - Essay IV — `2026-02-04-after-the-reply-iv.md` — new (Barcelona); 7 sections
  - Essay V — `2026-02-08-after-the-reply-v.md` — new (Barcelona); Prologue + 5 sections (H2 titles below register marks per steward call); H3 sub-sections in IV preserved
- Frontmatter chain integrity: I → II (grandfathered slug) → III (spec-canonical) → IV → V. All `sequence_slug: after-the-reply` for marker apparatus (Round 2).
- Inter-section ornaments dropped (register marks substitute the threshold function); opening/closing ornaments preserved.

- 2026-04-28T~afternoon — **The actual sidenote-layout cause: `<div class="lex-body">` template wrapper broke the `article.lex > .voice-block` direct-child selector.** Voice-blocks were at depth 2 (article > lex-body > voice-block) but SCSS `>` selector required depth 1. The CSS rule never matched in production. MVP HTML rendered correctly precisely because it had no wrapper. Found by comparing rendered-HTML structure to SCSS selectors directly. Fix: removed `.lex-body` wrapper from `templates/lex.html`. **Lesson:** when layout doesn't apply despite all rules being correctly compiled, inspect the rendered HTML structure vs the SCSS selectors first — `>` direct-child selectors break silently when intermediate wrappers exist. This single fix obviated all the wrapper-expansion + page-content overrides + subgrid replacement work; those changes happened to be improvements but weren't the cause. The actual cause was structural mismatch I never inspected. Sibling to blame-the-exotic-before-checking-the-boring (same afternoon).
- 2026-04-28T~afternoon — **Diagnosed `:has()` + subgrid as the layout blockers; both wrong.** Steward's first review of deployed lex pages: sidenotes still underneath in BOTH Safari 26.4 and Chrome desktop. I had blamed `:has()` browser support (false — Safari 26.4 supports it) and then subgrid fallback (false — Safari 26+ supports subgrid). The real blockers were: (a) `.page-content { max-width: 720px }` constraining article.lex regardless of wrapper width, never overridden; (b) `:has()` rule had ambiguous specificity competition with later `.wrapper` rules. Fix: body-class via default.html (`<body$if(class)$ class="$class$-page"$endif$>`) → reliable cascade `body.lex-page .wrapper { max-width: 70rem }` + `body.lex-page .page-content { max-width: none }`. Plus: replaced subgrid with explicit `grid-template-columns: $body-column $gap-column $sidenote-column` (works in all browsers; removes the `display:block` fallback path entirely). **Lesson:** when a CSS rule appears not to apply in a fully-supporting browser, the cause is usually cascade/specificity, not browser-feature gating. Should have inspected computed-styles before blaming browser support.

## Open horizons

- 2026-04-28T~wake — Composition-in-dialogue trio (Hearth landing, About *Hic*, About *Adsum*) — steward-led; executor as registrar; do not pre-empt.
- 2026-04-28T~wake — Lex cluster registrar transcription (Posture + Marks) — recognitions LOCKED in `project-arc-rework.md`; queued as honest registrar work if composition stays held.
- 2026-04-28T~wake — Cul-de-bouteille design — flagged secondary in yesterday's literal question; surface only if steward names it.
- 2026-04-28T~wake — Class-field reconciliation across about-pages — small mechanical sweep when convenient.
- 2026-04-28T~afternoon — **After the Reply publication + sequence-indication architectural gap.** 5 essays in Obsidian "10. Ready to publish/After the Reply"; 2 published (I, II); 3 to publish (III, IV, V). Frontmatter schema for sequence (`sequence`, `sequence_short`, `sequence_order`, `preceded_by`, `followed_by`) populated on Essays I + II — but templates (`post.html`, `default.html`) and `site.hs` render NOTHING from these fields. Data layer ready; visual layer empty. Three rough shapes proposed (header epigraph / closing block / standalone index page); steward call pending.
- 2026-04-28T~afternoon — **CLAUDE.md drift quiet flag**: `~/CLAUDE.md` Active Projects table still reads *"L1 is PRIMARY focus through end of May 2026"* + *"L2 PARKED through end of May 2026"*; actual focus has been ARC architecture work for at least the last week. Not surfaced to steward today (not load-bearing in this session); flag for when/if next CLAUDE.md edit comes up.
- 2026-04-28T~afternoon — **Old-style figures still not rendering after `font-variant-numeric: oldstyle-nums` added to body.** Most likely cause: locally-hosted `EBGaramond-Regular.woff2` was subsetted at the Google Fonts download step and stripped the `onum` OpenType feature. Need to either inspect the font file (fontTools / pyftsubset) and replace with a complete EB Garamond, or load EB Garamond from Google Fonts CDN where ONUM is preserved. Pending steward direction.

## Authorization moves

- 2026-04-28T~afternoon — Proposed order for fragmented before-lunch slice: (1) Posture transcription, (2) surface essays III/IV/V mechanically gated on sequence-indication call, (3) steward selects sequence-indication shape from three options, (4) cul-de-bouteille held for own session. Awaiting steward selection of starting point + sequence-indication shape.
- 2026-04-28T~afternoon — **Refined recommendation after PDF reference seen** (`After the Reply, iA version.pdf`, 52pp): minimum (C) standalone sequence-index page at `/after-the-reply/` + (A) small `II of V — After the Reply` header marker per essay. (B) closing prev/next deferred — PDF-restraint as design instruction (no running heads, no inter-essay nav in canonical form). **Coupled question surfaced:** does Essay II rename from *"The Ethics of the Reply, Part II: On Abdication and Fidelity"* to PDF-canonical *"II. On Abdication and Fidelity"* when sequence-indication lands? My read: yes, the "Part II" was a workaround for absent indication. Steward call pending.
- 2026-04-28T~afternoon — **Three architectural locks** after spec consultation: (1) Form 1 — individual essays + sequence-index gathering page, no composite long document; (2) "Trust the reader" per steward verbatim *"I guess I will just have to trust the reader. It is their loss if they don't..."*; (3) No retractation territory per steward verbatim *"nothing is being retracted yet."* All three settled before publish work begins.
- 2026-04-28T~afternoon — **Confirms + open calls.** Title-trim scope: Essay II only (Essay I title already canonical stem). URL `/after-the-reply/` confirmed. Marker shape pending (apparatus block vs Roman numeral in H1; my recommendation: H1 form, closer to PDF). Slug strategy pending (steward leaning spec-canonical for all 5 — *"we make the first two conform?"* — but no redirect mechanism in repo; three options surfaced: Hakyll redirect-stubs, CF `_redirects`, or grandfather I+II).

## Confidence to recalibrate

- 2026-04-28T~wake — Yesterday's architectural rigor needs to land in the writing. Risk: rigor doesn't carry through if writing-attention diffuses. Hold this.
- 2026-04-28T~wake — Patterns to hold today (from yesterday's five returns): bulk-scaffold instinct → "would Alexander do it?"; over-decorating-as-proving → surface simpler form; preserve-don't-delete on retirement; ARC territory steward+executor only (no Seb); count-from-1.

## AldineXXI spec commitments (steward, 2026-04-28 afternoon)

After visual review of first-production-exercise of §XII Sidenotes:

- **Old-style figures globally to body — reading-layer default.** Steward verbatim: *"globally to body — this way arabic numerals are integrated into the text, not placed upon it."* Implementation: `font-variant-numeric: oldstyle-nums` added to `body` in `_base.scss`. IBM Plex Mono apparatus elements unaffected (Plex Mono has no oldstyle; renders as lining naturally). Spec-grade change to AldineXXI §I implicit (reading-face default = oldstyle figures). Wants integration into AldineXXI-specification.md at /wrap-up.
- **Wrapper expansion for lex pages via `:has(article.lex)`.** Steward chose option (a). The §XII grid (38rem body + 2rem gap + 14rem sidenote = 54rem; up to 66rem with register-marker column) doesn't fit `.wrapper { max-width: 890px }` at desktop (21px base font scale). Implementation: `.wrapper:has(article.lex) { max-width: 70rem; }` added to desktop block in `_sidenotes.scss` with comment explaining grid requirement + `:has()` browser support (Safari 15.4+, Chrome 105+, Firefox 121+). Wants integration into AldineXXI §XII at /wrap-up.

## First-production-exercise pattern noted

The first production page exercising §XII Sidenotes (Posture + Marks lex pages, 2026-04-28) surfaced **four spec/implementation gaps** that wouldn't have been visible without the live page:

1. Wrapper width too narrow for the §XII grid at desktop scale (architectural — spec assumed wrapper expansion that didn't exist)
2. Leading-caps in source materials render as web-shouting unless wrapped in `<span class="small-caps">` (transcription craft)
3. Old-style figures not auto-applied to body — print-typography commitment latent in mixin but never wired to reading layer (spec implicit, never made explicit)
4. Over-reduction risk: my Manutius Younger compression dropped both the English glosses AND the four mark-hierarchy quotes. Steward asked English back; mark-hierarchy still pending judgment.

**Methodological lesson:** the §XII spec was authored 2026-04-24 with the SCSS infrastructure built 2026-04-27, but no production page exercised it until today. Specs need first-exercise to discover the implicit assumptions. Cousin to the canonical-tracker discipline pattern (writing-without-using = silent drift).

## Executor work completed

- 2026-04-28T~afternoon — **Posture + Marks transcriptions written.**
  - `content/lares/posture.md` — 8 voice-blocks per the 2026-04-24 committed gather (mirrors the MVP Posture HTML form). Stub replaced.
  - `content/lares/marks.md` — 24 voice-blocks + 25 recognitions per `~/Downloads/lex-marks-gather.md` (paired-form for #13 violist as two-paragraph sidenote). Stub replaced.
  - **First production exercise of `:::voice-block` / `:::sidenote`** — Pandoc fenced-divs already enabled in `site.hs`; SCSS fully built per §XII. No prior production page exercises the apparatus. Build-verify wanted before commit. `make build` auto-deploys per ARC CLAUDE.md, so `make serve` (port 3000) is the safe path for steward visual review.
  - **Judgment calls flagged for steward review:**
    1. Manutius Younger reduced to dedication-Latin only — 4 mark-hierarchy quotes (comma/semicolon/colon/period definitions) dropped because the recognition responds to the dedication argument. Easy expand-back.
    2. Bridging registrar prose dropped — followed MVP Posture pattern (quote + cite + recognition only). Bridges crystallized into recognitions.
    3. Paired-form #13 violist as two-paragraph sidenote — render question surfaces only at build time.
    4. Attribution as inline `<span class="small-caps">` — matches Essay I convention; production CSS doesn't auto-small-caps blockquote cite.
    5. Honest absences kept as italic footer note on marks page — provenance honesty, not preface (spec line 501 disallows prefaces).

## Substantive context received from steward (durable, saved as user memories)

- 2026-04-28T~late-afternoon — **Steward shared morning jurist conversation in full** (the "why" behind the day's publish-After-the-Reply velocity). Two durable user memories created:
  - `user-formation-out-of-place-as-authority.md` — steward verbatim *"I've spent most of my life feeling a bit in the wrong place at the wrong time in terms of how I see the world"* + jurist's reading of this as both accurate data AND authority.
  - `user-care-over-traffic-skills-as-accountability.md` — steward's reframing of an earlier "reckless" comment: care is constitutive of ARC, not instrumental; skills/agent-circles serve quality+consistency+accountability, not scale.
- 2026-04-28T~late-afternoon — **Jurist's substantive read of After the Reply** (saved into `project-after-the-reply-publication-state.md` at /wrap-up): Pandora as structural analysis not decorative classicism, Vico-Harrison thread inhabited not cited, form-enacts-argument observation about boredom essay, address-to-the-unborn as the only honest grammatical position, sequence narrowing the space of evasion. Registral placement: Berger's *Ways of Seeing*, Weil's political essays, late Sontag.
- 2026-04-28T~late-afternoon — **Steward decision (strong)**: register marks across all 5 essays — *"I want these to bear the register marks (all of the five essays)—I think they would flow more freely."* Form: Roman numerals as `:::register-mark` content, replacing `## I.` headings. For Essay V's named sections, headings preserved BELOW the register mark; if unsatisfactory, fold into prose. Aligns with "flow more freely" diagnosis; avoids codex-style heading-as-rupture on web.
- 2026-04-28T~late-afternoon — **Marker apparatus copy locked**: `III of V · After the Reply` (apparatus middle option). My recommendation accepted by steward.

## Sub-agent dialogues

## Bypasses