ISSUE 428 — NOTHING NEW WAS ADDED

[isaacsight]

What

Presses ISSUE 428 · MAR 2027 — NOTHING NEW WAS ADDED (何ひとつ新しくはない), a compare spread on YouTube's 15 July 2025 "inauthentic content" rename, plus its artifact edition.

• src/content/issues/428.ts — the issue: cream stock, classic layout, brick accent, coverSeal NO NEW POLICIES · III·27. Six externally-sourced facts read under two lenses (THE LETTER / THE QUEUE), a verdict, a references block.
• artifacts/428-nothing-new-was-added.html — the artifact edition: THE REVIEW QUEUE, a six-stratum descent from the uploaded file to the rule that judges it.
• src/content/issues/schema.ts — CompareSpread gains an optional references?: SpreadReferences.
• src/components/CompareFeature.{tsx,css} — renders that block under pop-compare-refs (ink-stock variants and print included).
• src/content/issues/index.ts — registration; LATEST_ISSUE flips to 428.
• src/content/issues/PUBLISHING.md — §III.2 compare row, §IV template pointer, and the "Last updated" line (the §IX hygiene pass).
• SCRATCHPAD.md — session entry.

Why

The material is a policy update that almost everyone read wrong: the platform renamed one guideline, named three ineligible categories, and stated plainly that no new policies were added to the Partner Program — and the creator internet read a ban on AI. Both halves of that are true and they pull against each other, which is exactly the shape compare exists for.

Read as THE LETTER, nothing moved: a rename of a guideline older than generative video, tool-agnostic by design, with the reused-content policy (commentary, clips, reactions) untouched. Read as THE QUEUE, three new words entered a review stack running at platform scale, and two of them — "generic", "off-putting" — are affective calls a classifier cannot show its work on. There is no blended reading that is sixty per cent letter and forty per cent queue.

How

Why a second compare and not an essay. A dial holds N positions on one variable; these are two complete accounts of the same six facts. Per docs/interaction-language.md rule 7, instance two is the threshold at which a pattern may be extracted — this deliberately does not extract one. CompareFeature is reused as written.

Why this one gives a verdict and 406 did not. 406 declined because both readings had equal claim on the same day. Here they answer different questions — what the platform owes you versus what your upload meets on a Tuesday — so declining would have been a false symmetry rather than honesty. No reading is written as a strawman; THE LETTER is argued as the platform would argue it, and no fact is invented for symmetry.

Why the schema grew. Every fact is external and dated, so the spread needed somewhere to print its sources. references is optional, so 406 is unaffected; the CSS is self-contained in CompareFeature.css rather than borrowed from EssayFeature.css, so a reader landing directly on /issues/428 gets it styled.

Sources (all read before the issue was set, cited in the spread's own policy file): the platform's monetization-policy and additional-reviews help pages, Social Media Today, PPC Land, Search Engine Journal, Tubefilter.

Artifact edition (PUBLISHING.md §V.5, docs/artifact-language.md §III). The spread argues the text; the artifact lowers a probe through the machinery the text runs on — upload → classifier → three buckets → second look → icon → the rule. Named depth axis with a real gauge; carried context (the upload's character, three authored composites, re-inks every stratum below); a floor that resolves per carry; every stratum legible at rest, so the probe raises emphasis and never conjures a layer. Seed 428-0715 printed on the masthead, timer-robust clock via the rAF-vs-setTimeout race, reduced-motion collapse, print stacks all six strata with the controls removed. What the reduction to the spread gives up — the descent, the gauge, and the three-way carry — is filed in the issue's header comment.

Filed at artifacts/428-nothing-new-was-added.html; not published to the claude.ai artifact surface — that stays the editor's call.

Testing

• npm run build passes
• npx vitest run passes — 811/811 across 51 files
• npx tsc --noEmit passes (via npm run build; a bare global tsc reports pre-existing tsconfig.json deprecation errors on a clean tree too, unrelated to this change)
• node scripts/check-editorial.mjs — clean, 69 issues scanned (artifact-first, no POPEYE naming, no emoji)
• node scripts/check-adherence.mjs — clean, 0 raw hex across 32 files
• Visual preview at /, /issues/428, /issues/427, and ≤640px — not run in this environment

Not deployed. This is a feature branch; only a push to main publishes (PUBLISHING.md §VII).

Screenshots

None — no dev server was run in this session. The cover, the switch at both lenses, and the artifact's six strata are worth an eye before merge.

---

Generated by Claude Code

[isaacsight]

CI status — 6 of 7 green, e2e blocked on a pre-existing timeout

Green: Web — Type-check & Build, kbot — Build & Test, VS Code Extension — Build, notify-pr (plus two skipped notify jobs).

e2e — cancelled, not failed, and not caused by this PR.

.github/workflows/e2e.yml sets timeout-minutes: 15. The job started 12:27:08 and was cancelled at 12:42:21 — exactly the ceiling. The Playwright step got as far as:

Running 120 tests using 1 worker
Skipping free account auth — TEST_FREE_PASS not set
·Skipping pro account auth — TEST_PRO_PASS not set
##[error]The operation was canceled.

One test through, then ~13.5 minutes of silence. No playwright-report/ was produced, so there is no failure to read — the suite hangs before it reports.

This is chronic, not new. Every e2e run in the workflow's recent history is cancelled, across unrelated branches:

Created	Branch	Conclusion
2026-08-01	claude/youtube-monetization-rules-pv8pos	cancelled
2026-07-30	fix/remove-social-media-assets	cancelled
2026-07-30	feat/channel-intro-finish (×5)	cancelled
2026-07-23	feat/cinematic-video-system (×4)	cancelled
2026-07-21	claude/brokerage-system-problem-utqyqh	cancelled

What I verified locally instead

Ran the magazine-facing specs against a local preview build on this branch:

• e2e/tests/stacks.spec.ts — 4/4 pass
• e2e/tests/issue-415-close.spec.ts — 2/3 pass

stacks.spec.ts › the ledger lands on the real issue route hardcodes N°427, and 428 pushes 427 into the second volume (each .stacks-volume is min-height: 100vh), so it was the obvious suspect. It failed once, then passed 3/3 on re-run in the identical combination, and 3/3 in isolation — a flake in the walkable room's actionability wait, not a regression. Worth knowing it is brittle-by-construction, though: it will need its pin moved, or generalised to the latest issue, at some point.

issue-415-close › zero console errors on the page fails identically on the base commit (ad9098d) — local console errors from absent Supabase env secrets, not this diff.

Also green on this branch: npm run build, 811/811 vitest, check-editorial (69 issues), check-adherence (0 raw hex).

Not fixing here

Raising the e2e job timeout, or finding what makes the suite hang after the auth setup, is a CI change affecting every branch and well outside an issue press. Happy to take it as its own PR if you want it.

---

Generated by Claude Code