Record the pre-#906 100-document baseline of the typed-graph bench, with tokens by phase - #917
Merged
Merged
Conversation
…ith tokens by phase Two groups on dev at 961c3c0 with gemini-3.5-flash, judged and with errata: precision 91.0% / 88.5% before errata and 96.0% / 96.5% after; same-sentence gold recall 11.3% / 13.2% before errata and 18.6% / 22.4% after, with the errata retractions mostly wrong on this binary (72/98, 164/196), which is what #906's two-vote retraction addresses; 158k and 183k tokens per document. The token table by phase shows where they went: the alignment window carried the rule proposals' per-item property tables, the two-vote phrase alignment, and the governance agent deciding the same pairs from up to ten concurrent runs (9481 decisions on 1421 pairs), fixed in #916. The bench judge now fails fast when it would read a sealed key from llm_settings instead of sending the ciphertext and collecting 401s. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Signed-off-by: Wayland Yang <wayland0916@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two full groups (100 documents, judge 200, errata) on dev at 961c3c0, the binary before #906, written into the bench README as the 100-document counterpart of #906's 20-document table. Precision 91.0% / 88.5% before errata and 96.0% / 96.5% after; same-sentence gold recall 11.3% / 13.2% before errata, 18.6% / 22.4% after, with retractions mostly wrong on this binary (72/98, 164/196); 158k and 183k tokens per document. The by-phase token table attributes the bulk to the alignment window: per-item property tables in rule proposals and two-vote phrase alignment (cut in #906), and the governance agent deciding the same pairs from up to ten concurrent runs (fixed in #916).
Also: the bench judge fails fast when it would read a sealed key from llm_settings (
enc:v1|…) instead of sending the ciphertext and collecting 401 per document, which is what happened on the first attempt.🤖 Generated with Claude Code