Skip to content

Flat rebuild: mark superseded captures in the index (clean diff for evaleval#307) - #1

Closed
j-chim wants to merge 1 commit into
fix/generation-config-and-source-datafrom
feat/flat-index-supersession
Closed

j-chim wants to merge 1 commit into
fix/generation-config-and-source-datafrom
feat/flat-index-supersession

Conversation

@j-chim

@j-chim j-chim commented Sep 28, 2026

Copy link
Copy Markdown
Owner

Fork-internal stacked view: base is the branch of evaleval#306, so this shows only the supersession commit. The upstream draft is evaleval#307.

Corrected or re-scraped records arrive as new files and nothing retired
the old ones, so consumers pooled both. The index rows now carry
evaluation_id, retrieved_timestamp, cron_adapter, cron_run_date,
result_count, superseded_by and superseded_reason. A checked-in
flat/retired.jsonl is applied first; then, for cron-published rows only,
captures sharing (collection, adapter, evaluation_id prefix) whose id tail
equals retrieved_timestamp form a series and every capture but the newest
points at it. Series with two captures from one run date, or whose newest
capture has fewer results, are reported instead of marked; collections
still awaiting id backfill are skipped. Ids are backfilled once with a
bulk --bootstrap-ids download and trickled afterwards. data/ and
flat/objects/ are never modified.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@j-chim

j-chim commented Sep 28, 2026

Copy link
Copy Markdown
Owner Author

Closed: not presenting these as a stack. See evaleval#307.

@j-chim j-chim closed this Sep 28, 2026
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.

1 participant