The standing post-feature audit over a7f0cfe. Correctness (data): splits
become photo-scoped store records so splitting one edition no longer
force-splits same-named editions, and renaming a split copy migrates its
protection to the corrected title instead of silently re-merging copies.
Correctness (web): edit scoping now counts siblings by NORMALIZED title
(matching how stored edits apply), same-title-same-photos edits are
refused rather than corrupting the sibling entry, split copies serve
their real per-photo cues to the edit form instead of blanks, and a
split whose row vanished underneath returns 409 instead of a false 200.
Silent failures: replay_titles refuses to rebuild from a PARTIAL raw
cache (fresh clone + one --only extract would have truncated the
committed titles.json); the edit endpoint writes in crash-safe order
(cull, record, replay); corrupt curation stores fail loud naming the
file; retried edits don't double-record. Review-decision durability:
drop_rows never drops dedupe_veto rows — a rename retitles them in
place — and writes through a no-reload path so a concurrent rewrite
can't silently discard the cull. Style: catalog action cells get their
own class (.rowactions' flex display broke table alignment), editor
inputs match the design system and stop overriding the global
focus-visible outline, EditBody's clear-semantics docstring scoped to
cue fields, "nothing to change" derived from the record itself.
Tests: 8 new (photo-scoped splits, veto preservation, photo-narrowed
drops, 409s on both curation endpoints under a running job, partial-raw
replay guard, rename-keeps-protection lifecycle, corrupt-store error,
cue-field editing) and the dead edition_hint key in the edit test now
exercises real cue fields. 259 passing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016jXZFSTZQKzAC8fqpWSz9g
Full artifact reset and re-extraction: 132 titles (was 133), 62 reshoot
sightings, 17 titles resolved from stub cache, 115 parked for the
token. Several old misreads self-corrected on the fresh pass
(Hebarceos->Herbaceous, THE BEST OF RA...->CHEAPASS GAMES, Mostly
Original->The Ain't It Cool Trivia Game) and box subtitles now read
fuller, which shifts many stub cache keys — those titles wait for real
data rather than resolving against stale stub queries. Review state
starts clean by design; splits and dismissals are Eric's to redo in
the UI.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Both resolve as unmatched by design — it's an RPG (geekdo rpgitem), so
the boardgame search realistically returns nothing; empty-search stubs
cover all five queries incl. truncation heads. Resolve also merged six
duplicate reads across photos (Hebarceos -> Herbaceous et al.).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The vision prompt now returns {titles, unidentified}: boxes that look
like games but can't be confidently titled are reported (location
relative to identified neighbors, partial text, art notes) instead of
silently omitted. They land in data/unidentified.json keyed by photo,
and the end-of-run summary lists them — plus low-confidence reads —
with instructions to retake a closer photo and re-run. New --force flag
re-extracts everything; pre-feature raw caches (bare arrays) still
parse. Live run on IMG_4499 confirmed the flow and the low-confidence
list correctly flags the known "Hebarceos" misread.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Long transcribed box titles that defeat search now retry with shorter
heads (pre-separator, pre-"Game ..." descriptor, first-two-words from
the pre-subtitle part) matched exact-only against the head — fuzzy
thresholds stay untouched. Tie-breaks gain a publisher pick: when the
box showed a publisher and exactly one exact-named candidate is from
that publisher, it wins (SPI's Sorcerer 1975 now beats the more-owned
White Wizard Sorcerer 2019). The extract prompt excludes thematic/
subject years from year_hint; re-extracting Flat Top's photo drops the
bogus 1942, and a regression test pins that a wrong year can never
drive version selection. Re-extraction also drifted two transcriptions
(DUNGEON!, and Herbaceous misread as "Hebarceos") — fixtures added; the
misread demos review's re-search rescue in the end-to-end run.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First live vision run: 4 shelf photos -> 27 unique titles with edition
cues (publishers, Bookcase Game/Designer's Edition wording, art notes);
cross-photo dedupe merged repeat sightings. data/titles.json now holds
real pipeline data, so the hand-typed resolve test list moves to
tests/data/titles.json (tests and record_fixtures updated), and the
matches.csv generated from stub fixtures for that list is removed —
the real resolve run will regenerate it once the BGG token arrives.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
bggpipe resolve works end to end: search -> exact/fuzzy candidate
scoring -> auto/ambiguous/unmatched classification with owned-count
tie-breaks (mixed base/expansion candidates never auto-match), version
scoring from edition cues (never guessed; no cues -> version_unknown),
idempotent matches.csv appends.
Discovered mid-build: BGG now requires registered-application Bearer
tokens on the XML API (2025 policy change) and returns 401 otherwise.
Client sends Authorization from BGG_API_TOKEN and raises an actionable
BGGAuthError; CLAUDE.md and the bgg-api skill are updated to match.
Live fixture recording is blocked until registration is approved, so
tests replay hand-crafted stub fixtures via a network-refusing
transport; scripts/record_fixtures.py re-records real XML under the
same cache keys once a token exists. One live read-only smoke test is
skipped unless --run-live.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Eric Wagoner