Always-renders polish backlog from the #395 inventory (styling, CLI errors, lint-shaped gaps)

[abalter]

What

The #395 inventory's residue: failures that are already visible-or-loud but inconsistent, plus lint-shaped gaps. None violates always-renders on its own; together they make the error surface uneven. One backlog issue so they don't scatter:

Styling / render consistency

• assemble.js:118-123 missing-child note renders as a bare unstyled paragraph that reads like content — give it the error-block styling.
• parse-error / tag-error / .enscribe-error marker classes have no CSS rules (unlike ref-error/cite-error) — same error styling family.
• Deferred placement markers in a master (<toc>/<endnotes> in master scope) are dropped with a warning (assemble.js:139-141) — a small visible placeholder would be honest.

CLI ergonomics

• emit() write failures surface as raw ENOENT stacks (cli.js:570-576) — map to a clean CliError.
• --live build: a missing non-website master child throws mid-copy (build-live.js:222) where the nav-item path warns+skips — pick one policy.
• copyShellAssets require.resolve failure (unbuilt engine bundle) is a raw throw (build-live.js:57-64) — name the fix ("run the package build").
• Known --theme in the BROWSER bundle: getThemeCss is a Node fs read (index.js:275-281) — themes are sub-1KB; build-inline them.

Lint-shaped gaps (accepted-but-inert authoring)

• <library> with no src and no inline content: vfile-only note (library-load.js:186-189).
• Data-store <fig> carrying BOTH src= and a base64 body: src silently wins (asset-load.js:220-222) — one collision-style flag.
• Dataset→table hand-off with no format on either side (asset-load.js:343-345) — name the miss.
• ris / endnote-xml format words accepted by the gate but unwired in FORMAT_FORCETYPE (library-load.js:76-78) — falls to auto-detect, which does not detect them.
• CLI URL <table src> gap (browser preloads it, CLI sync path cannot) — small preload-collection fix flagged in the audit.
• DSL live-link CDN <script src> has no onerror hint (index.js:442-444) — offline reader sees un-activated source (visible degrade), a one-line hint would name it.
• Empty authored <bibliography> (zero resolved cites) is suppressed (bibliography.js:234-239) — same placement-marker-semantics question as the notes issue; settle there, execute here.

Source: #395 universality audit inventory (posted on #395). Each line cites its inventory row; verify against current code before fixing (verify-first).

[abalter]

Input needed (for Ariel)

The situation: The always-renders polish backlog (#413) has two leftovers only you can supply: a handful of citation/metadata facts (the software has no CITATION.cff yet), and a little prose that has to be in your voice. The facts unblock a mechanical slice right now; the prose blocks nothing and can happen on your own clock.

What's needed from you:

Part A — three facts (one line each; unblocks the CITATION.cff / metadata nibble):

• Citation name — the exact name/form the software should be cited by. Every package.json and the LICENSE currently say "Ariel Balter" — reply "as-is" to confirm, or give the preferred form. (Recommendation: as-is, unless you cite academically under a fuller form.)
• ORCID — yes/no, and the iD if yes. None is on file anywhere in the repo today (the only ORCIDs present are Jane-Goodall placeholder examples in the docs). (Recommendation: include it — it's the one durable author identifier a citation file can carry.)
• Sole author, or additional credits? — is the software citation just you, or are there other names to list?

Part B — reserved writing (your voice, no deadline — a punchlist for the content pass, not decisions):

• Book landing intros. All three generated books show the default "Select a chapter to begin reading." placeholder and have no intro paragraph: The Enscribe Authoring Guide, The eHTML Vocabulary, and The Enscribe Vocabulary (docs-source/{authoring_guide,ehtml,enscribe_vocabulary}/index.emd). Each wants ~1 short paragraph. (Heads-up: those three index files are machine-generated by docs-gen/generate-docs.mjs, so the slice threads your text through the generator — you write the prose, it wires it in; pasting into the file directly would trip the freshness guard.)
• STATUS.md narrative trim. Several checklist entries have grown into implementation stories of 2,000–5,900 chars (the "Live book edit loop" line alone is ~5,900; two others ~2,200–2,400), and each already links the issue/spec that owns the detail. An editorial pass in your voice moves the per-item narrative into those issues/specs and leaves STATUS as a checklist.
• Dev on-ramp paragraph. There's no "clone / install / test" block for someone working on the repo — README's Install section is for using the package, and CONTRIBUTING.md is a governance manual. A short "Working on the repo" block in your voice fills the gap.

Urgency: Part A (the three facts) is the only thing waiting — it unblocks the mechanical CITATION.cff/metadata slice. Part B blocks nothing; the polish slice lands its mechanical half without it, and the writing slots in whenever you get to it.

[abalter]

Hygiene-mechanical sweep — Part A still blocks; the code backlog is now triaged (verify-first)

The hygiene-mechanical slice swept this issue. It does not close here. Part A (your three facts) still gates the CITATION.cff nibble; Part B stays your writing pass; and the body's code backlog was verified line-by-line against current code and triaged below — a focused follow-up slice can land the mechanical ones, the rest need rulings only you can give.

Part A — the three facts (still the only thing blocking the CITATION.cff nibble)

Unchanged — none of the three checkboxes is answered, so the hygiene-mechanical slice skipped writing CITATION.cff rather than guess a legal name / ORCID. The moment you drop the three facts here, the file is a one-nibble mechanical add. (package.json + LICENSE say "Ariel Balter"; no ORCID is anywhere in the repo.)

Part B — reserved writing (your voice, no deadline)

Book landing intros, STATUS.md narrative trim, dev on-ramp paragraph — unchanged, blocks nothing.

The code backlog (body items) — verify-first triage

Every line re-verified against current code (the prompt's line numbers had drifted; corrected below). Dominant blocker across the cluster: one unsettled family policy — at each "accepted-but-inert" seam, is the right degrade a silent vfile note or a visible always-renders flag (the __asset-error/__library-error family)? That's the #395 always-renders scope boundary; deciding it per-item silently would fragment the family, so those are held for a ruling.

#	Item	Current state	Disposition
S1	missing-child note is a bare <p> (assemble.js:173)	present; code's own note calls robust reporting a "#190 later refinement"	📋 decision — needs a marker taxonomy (__master-src-error vs generalize __include-error); visible-output
S2	parse-error/tag-error/.enscribe-error markers have no CSS (default.css; siblings ref-error/cite-error/import-error do)	confirmed no rule	🟡 mixed — the inline color rule is mechanical (copy the import-error sibling), but the block-family look (.enscribe-include-error, .enscribe-asset-error, S1) has no sibling to copy and is a visual-design call. include.js:7 explicitly reserves "family CSS is #413's scope".
S3	deferred <toc>/<endnotes> markers dropped with a warn (assemble.js:196)	present	📋 decision — new placeholder semantics; tied to the placement-marker question (with L7)
C1	emit() write failures → raw ENOENT stack (cli.js:730, no try/catch)	present; readInput (cli.js:349) is the exact CliError pattern to copy	🟢 mechanical-ready — self-contained in cli.js, error-path only, no fixture impact
C2	missing non-website master child throws mid-copy vs nav-item warn+skip (build-live.js, unguarded copy at :232/:243)	asymmetric policy confirmed	📋 decision (policy) — warn+skip vs a clean CliError vs both-throw; a declared <chapter src> is structure, arguably fatal (lean: named CliError)
C3	copyShellAssets require.resolve of an unbuilt engine bundle → raw throw (build-live.js:57)	present; website caller already softens it	🟢 mechanical-ready — name the fix ("run the package build") in the message (a clean Error; a fully clean CliError needs exporting CliError, shared with C1)
C4	getThemeCss is a Node fs read reachable in the browser bundle for a known --theme (index.js:277)	confirmed real defect — a browser user picking a valid theme gets fileURLToPath is not available…, no CDN fallback	📋 decision (architecture) — build-inline the 3 theme CSS files following the proven hover-preview pattern (new module + .browser.js twin + package.json browser map + tsup define); multi-file, bundle-size, and #413-scope call
L1	empty <library> (no src, no content)	a vfile note already exists (library-load.js:257)	✅ verified-already-done (the literal "add a warning" ask); the "invisible-note → visible-flag" upgrade is the family policy
L2	<fig> with both src= and base64 body — src silently wins (asset-load.js:219)	no flag today; dup-id sibling shows a visible collision flag	📋 decision (family policy); mechanically shaped
L3	dataset→table with no format on either side → raw-HTML passthrough (asset-load.js:342 → table.js:596)	no warning today; a real silent miss	📋 decision (family policy) + light semantic call
L4	ris/endnote-xml accepted by the gate but unwired in FORMAT_FORCETYPE (library-load.js:127)	deferred by design — the code comment defers forceType "when those land"	📋 decision/roadmap — drop from accept-set, warn, or wire citation-js plugins
L5	CLI URL <table src> preload gap (cli.js:459 collects only library sources; collectTableSources exists+exported but unimported)	clean isolated gap	🟢 mechanical-ready — add collectTableSources to the cli import + the URL-filtered collect (mirrors browser.js:209); +1 CLI test, empty fixture diff
L6	DSL live-link CDN <script src> has no onerror (index.js:444 + website-shell.js:191)	no onerror convention anywhere	📋 decision — behavior/wording/surface undecided; spans two committed emitters (guarded by examples:check)
L7	empty authored <bibliography> silently spliced, no message (bibliography.js:256; minipage sibling does message)	present	📋 decision — placement-marker "settle there, execute here"; decide with S3

Mechanical-ready right now (a small focused slice): C1, C3, L5 — all error-path / isolated, no fixture or visual impact. C3+C1 share one small structural question (export CliError / a shared errors module for a fully-clean CLI surface). L1 is already done. Everything else needs a ruling: the always-renders note-vs-flag family policy (S2-block, L2, L3), the placement-marker semantics (S1, S3, L7), the missing-child policy (C2), the theme build-inline architecture (C4), and the RIS/EndNote roadmap call (L4).

(Full slice report + evidence: ~/enscribe-reports/slice-report-hygiene-mechanical.md.)