Skip to content

ISSUE 428 — THE ONE PERCENT - #70

Draft
isaacsight wants to merge 1 commit into
mainfrom
claude/youtube-ai-slop-detection-ckgsez
Draft

ISSUE 428 — THE ONE PERCENT#70
isaacsight wants to merge 1 commit into
mainfrom
claude/youtube-ai-slop-detection-ckgsez

Conversation

@isaacsight

Copy link
Copy Markdown
Owner

What

Ships ISSUE 428 — THE ONE PERCENT (一パーセント), a dispatch on the video platform's cluster-termination sweep against AI-slop content farms, and the false positive that hid inside its error bar. Adds the issue file, its mandatory artifact edition, registry wiring, and the PUBLISHING.md hygiene pass.

Why

Filed from a reader-supplied transcript of a published video commentary: the Google SCCTS research paper (~50,000 coordinated clusters identified, on the order of 130,000 channels terminated, appeal-overturn rate under 1%), Kurzgesagt's public account of being misread as synthetic slop and throttled (25M subscribers, human-made for over a decade, repaired via direct platform contacts), and the access asymmetry for creators without those contacts. The doctrine the dispatch files: a detector's honesty is measured at its appeal desk, not its precision figure.

Fixes #

How

  • src/content/issues/428.ts — dispatch spread, five propositions (FIELD → METHOD → CAUTION → ACCESS → DOCTRINE), bridge to 417, wire bulletin, — 30 — terminator. Identity: ledger stock + ledger-rule layout + graphite accent per 372's audit-register precedent, differentiated from 401 (ivory/asymmetric-left/pool). Claims printed as reported, hedges kept; provenance filed in the header comment.
  • artifacts/428-the-one-percent.html — artifact edition drafted first per artifact-language: an operable sweep field of 400 seeded channels (deterministic from printed seed 428-0805), a roving-tabindex threshold radiogroup carried into every sweep, authored hit rates disclosed in the colophon, a claims register citing public reporting at the floor, timer-robust sweep animation, reduced-motion instant resolution, print keeps the field/register and hides the controls. The reduction the spread gives up is filed in the issue header.
  • src/content/issues/index.ts — import + ALL_ISSUES registration; LATEST_ISSUE flips to 428.
  • src/content/issues/PUBLISHING.md — §IV dispatch example → 428, footer refreshed.
  • SCRATCHPAD.md — session memory updated.

Nothing publishes from this branch; only main deploys.

Testing

  • npm run build passes
  • npm run test passes (145 registry/accent tests)
  • npx tsc --noEmit passes (with the repo's local TypeScript 5.9.3; a globally-resolved 6.0.2 emits pre-existing tsconfig deprecation notices unrelated to this change)
  • Tested manually with kbot in dev mode — n/a, magazine-only change

Screenshots

n/a — preview locally with npm run dev at / (new cover) and /issues/428.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EcHbd4KB1YxK3HudJEsiuc


Generated by Claude Code

A dispatch on the video platform's cluster-termination sweep and what
its error bar hides. The Google SCCTS paper claims ~50,000 coordinated
clusters identified and on the order of 130,000 channels terminated,
with an appeal-overturn rate under one percent; in the same season the
platform's AI detectors misread Kurzgesagt — 25 million subscribers,
human-made for over a decade — as synthetic slop and throttled the
channel, repaired only through direct platform contacts. Five
propositions from sweep to doctrine: measure the detector at its
appeal desk, not its precision figure.

Identity: ledger stock + ledger-rule layout + graphite accent, per
372's audit-register precedent — the sweep is an accounting register;
differentiated from 401 (ivory/asymmetric-left/pool). Seal TERMINATED
· OVERTURNED; bridge to 417 (the same adjudication running in reverse
at industrial scale). Artifact edition drafted first at
artifacts/428-the-one-percent.html: an operable sweep field of 400
seeded channels (seed 428-0805 printed at the masthead), a threshold
radiogroup carried into every sweep, authored hit rates disclosed in
the colophon, and a claims register citing public reporting at the
floor. Reduction filed in the issue header. tsc clean, vite build
clean, 145 registry tests pass. PUBLISHING.md hygiene pass included.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcHbd4KB1YxK3HudJEsiuc
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants