Plan — Repo restructure

2026-refresh › repo-restructure · chore · 2026-08-05 · spec.html

Plan — Back-catalogue ops

2026-refresh › back-catalogue-ops · feature · 2026-08-05 · spec.html

Three phases: build the manifest and write all card copy (needs no design) · build the batch machinery and prove it on a pilot set · run the full catalogue with Asim spot-reviewing. F1 can start immediately; F2's card rendering waits on design-system's card template.

F1

Build the manifest and write every card's copy

Dossiers + triage + the August pivots reconciled into one machine-readable manifest (route per lecture); all ~35 card copy files written to the canonical register and fact-checked by execution.

Delivers J1.1 J3.1

Proved byThe manifest accounting for every kept lecture (zero unaccounted, counts per route stated) and a card-copy audit table: every factual claim in every card verified by a real run or TC39 status, zero failures.
the work + evidence drawers

Build

  • diffManifest generated from triage + dossiers + pivots: route (card · trim+card · upscale/level · as-is), inputs, dossier link per lecture — the KEEP-loudness question resolved by measuring the catalogue
    evidence
    On tick.
  • diffAll card copy files written (canonical register: what changed · the 2026 truth · why still worth watching), British English
    evidence
    On tick.
  • diffEdits log drafted: every trim's in/out timestamps from the dossier lines (Scribe word-precision), justification cited
    evidence
    On tick.

Prove

  • logManifest completeness check (zero unaccounted) + the card-copy fact audit table, all claims verified
    log
    On tick.
F2

Build the batch runner and prove it on a pilot set

One idempotent runner over the manifest (render card → trim → upscale → loudnorm → conform per route), proven on ~6 lectures spanning every route. Needs design-system's card template for the card route.

Delivers J1.2 J1.3 J1.4

Proved byThe pilot batch: ~6 finished lectures spanning all routes, every one conform-passing, and a re-run log showing the second invocation skips finished work (idempotence demonstrated).
the work + evidence drawers

Build

  • diffBatch runner in packages/scripts/: manifest in, finals out, per-route pipelines composed from the existing scripts, content-addressed skip logic
    evidence
    On tick.
  • logPilot set rendered: one card-only, one trim+card, one upscale-only, one as-is no-op, one async import, one card+upscale
    log — batch output
    On tick.

Prove

  • logconform-check over the pilot set (all pass) + the idempotent re-run log
    log
    On tick.
F3

Run the full catalogue and collect Asim's spot approvals

Every manifest row executed; review pages per batch; flags fixed and re-run; the manifest finalised as the publish inventory.

Delivers J2.1 J3.1

Proved byThe full-run report: every route's count rendered and conform-passing, Asim's spot-review approvals recorded per batch, zero manifest rows unaccounted — the inventory udemy-publish will consume.
the work + evidence drawers

Build

  • logFull batch run to courses/adv-js/media/out/; failures triaged, fixed, re-run
    log
    On tick.
  • diffReview pages per batch (card joins + cut points, first 60s clips); Asim flags handled as config fixes + re-runs
    evidence
    On tick.

Prove

  • logFull-run report + conform sweep + recorded approvals; manifest marked final
    log
    On tick.

Decision Defaults — standing autonomy for this run

  • Card copy wording within the canonical register is the writer's call — the fact audit and Asim's spot reviews are the gates.
  • Batch ordering, retry handling and skip-logic details are the builder's call, logged.
  • Not covered (always a stop): changing any dossier verdict or route · touching the evergreen KEEPs beyond upscale/loudness · editing narration · uploading anything to Udemy · recording a verdict without a verifier's report · amending this plan.

Decision Log

Empty — fills during the run.