Dw Title Repair
repo: ~/Projects/dw-title-repair · 9 commits · 0 in last 24h, 0 in last 7d ·
Skills used
- /ledger-supp1
- /ledger-record1
- /undo1
Creative ideas + design notes
Commits with substantial prose (≥120 chars) — the rationale behind each move.
24d7b63 · 2026-09-11 · TK-11240 part B: backfill dwc.pattern_name on 223 ACTIVE Hollywood products
Approved memo (yolo-queue/_approved/). 223 of 283 have an authoritative
versa_catalog source; the other 60 are left unfilled rather than derived from
titles, since title-derivation is the origin of the category-label pollution.
Live-verified 25/25. Verify-before-write: skips any product that already has a
value, so it cannot overwrite. Undo deletes only values this script wrote.
Part A (theme module) blocked on read_themes token scope, not approval.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
192fcfd · 2026-09-10 · TK-11376: derive ledger tag from the map filename
Both supplements matched MAP.includes('supp') and wrote into one shared
apply-supp-progress.jsonl (1,822 records = 920 + 902). Restore maps are separate
and authoritative so rollback was never affected, but per-supplement forensics
were muddied. Tag now comes from the map's own filename.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
85acf92 · 2026-09-10 · TK-11376 supplement 2: 902 more via clean manufacturer_sku metafield
Products with no dw_sku anywhere but a digit-bearing manufacturer_sku metafield
(rejects the junk values SANCAR/TRUE/Donghia). Drops any whose mfr_sku is shared
within its group, since a shared value cannot differentiate.
Applied the lesson from the 17-id overlap: checked PRODUCT-ID disjointness, not
just title uniqueness. Found and excluded 8 ids that overlapped the main map.
Verified after exclusion: 0 id overlap vs main and vs supp1, 0 title collisions
across all three maps.
Coverage 13,679 -> 14,581 of 17,341.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
36da240 · 2026-09-10 · TK-11376: add supplement rollback (t8) and document the 17-id map overlap
The supplement was validated for TITLE collisions against the main plan but NOT
for product-ID overlap. 17 ids appear in both maps with different targets. t3 ran
first and won; t2 correctly detected drift and skipped, so nothing was double-written
and all 17 are distinctly titled live. But rollback needed a supplement undo, and the
two must run in order t8 -> t9 or those 17 never return to their original titles.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2933b94 · 2026-09-10 · TK-11376: re-read 3x before declaring a product not_found
Observed mid-run: the verify read returned null for two products that
demonstrably exist and are ACTIVE (they had already been renamed in stage 1).
A transient null was being treated as 'deleted' and skipped. Harmless for an
already-applied product, but an unapplied one could have been silently missed.
Now re-reads 3x with backoff before believing it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
48ff5d4 · 2026-09-10 · TK-11376 supplement: recover 920 more products via dw_sku metafield (mirror column null); apply.mjs honours MAP=
The mirror column shopify_products.dw_sku is null on 4,583 affected products, but the
dw_sku METAFIELD holds a value for 920 of them. Metafield is truth (same lesson as the
mfr_sku pollution). Verified 0 collisions against each other or the main plan.
Coverage 12,759 -> 13,679 of 17,341.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
File tree
20 files tracked. Click any to browse the source at HEAD.
rendered in 0ms