evidence(bakeoffs): Verdana, Tahoma, Trebuchet MS vs the open corpus - #25
Merged
Conversation
Three high-value unresolved Word/UI sans fonts run against the full discovery corpus. The answer holds: no metric clone exists - every top candidate is visual-tier (best mean 0.6-1.8%, but max 32-85%, so one glyph always reflows). What the broad corpus DID surface: closer-by-mean visual leads than the registry's current top_candidates - e.g. Tahoma -> Viga (0.58% mean) vs the prior Open Sans (4.85%), Trebuchet MS -> Ruda (1.06%). Useful leads for a reviewer, but still not metric-safe; nothing is promoted - records.json is untouched.
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Runs the bakeoff (#24) for three high-value unresolved Word/UI sans fonts against the full Google discovery corpus, to test whether the broad corpus changes our current "no good substitute" answer.
It does not - the answer holds. Every top candidate for all three is
visual-tier: a low-ish mean but a max of 32-85%, so at least one glyph always reflows. Nodirect/likelycandidate exists, matching the existingvisual_onlyverdicts. Nothing is promoted;records.jsonis untouched.What the broad corpus surfaced, vs the registry's CURRENT committed
top_candidate:So the bakeoff confirms Tahoma's existing pick (Viga) and surfaces two new leads (Ruda, LINE Seed JP) the registry doesn't currently name. These are leads for a reviewer, not promotions.
Note on the numbers: the bakeoff uses docfonts' own prose-weighted mean + strict per-glyph
LATIN_COREmax, which differs from the apryse summary the committedtop_candidateswere measured with (e.g. the apryse Viga figure is 1.53% / 7.15%). So compare which font is closest, not the raw deltas across methods - and the huge max deltas are exactly why none of these is a metric candidate.Data-only PR: three committed
BakeoffResultJSONs (each withtargetFace), validated by the existing bakeoff tests (face-scoped, ranked, public-safe, no verdict field).Verified locally: bun test -> 114 pass / 0 fail; tsc clean; biome clean.