Repository navigation
docs: correct the got recall figure, measured with a stale binary - #43
Merged
Merged
Conversation
BENCHMARK.md was last regenerated in #40 ("carry module path segments instead of re-deriving them") — the commit whose fix resolves exactly the TypeScript call sites this row was reporting as misses. The figure recorded there came from a binary built before that fix, so it credited the tool with 197 fewer resolved call sites than it had already earned: got ts 1723/2184 = 78.9% -> 1920/2184 = 87.9% misses 461 -> 264, and `request` falls from 91 unpinned to 49 Re-run on a clean release build. The corpus is untouched and still pinned to the same commits in bench/repos.tsv, so this is the same measurement on the same inputs — nothing about the tool changed, only the number was wrong. The tree-sitter 0.27.1 bump in #42 is not responsible, which is worth stating because the timing invites the assumption. A 0.27.0 binary built from this same source reports 87.9% as well, and an interleaved A/B over 20 runs per command shows the two versions indistinguishable: map 0.0675 vs 0.0679, doctor 0.0699 vs 0.0702, callers 0.0683 vs 0.0678. The other five rows and the whole grep-comparison table regenerate byte for byte, as do all eight regenerated blocks in EXAMPLES.md. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01VXJrNh9MMrV6WkJbdZnZyP
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
BENCHMARK.mdreportedgot(TypeScript) internal recall as 78.9% when the tool was already achieving 87.9%. One row changes:197 call sites that
ctxwas resolving correctly were published as misses.Cause
BENCHMARK.mdwas last regenerated in #40 — carry module path segments instead of re-deriving them — which is precisely the fix that resolves these TypeScriptrequestcall sites. The figure recorded in that commit came from a binary built before the fix it was meant to reflect, so the row has understated the tool ever since, in the one document whose whole purpose is calibration.The corpus is untouched and still pinned to the same commits in
bench/repos.tsv. Same measurement, same inputs, clean release build — nothing about the tool changed, only the number was wrong.It is not the tree-sitter bump
Worth stating, because #42 landed immediately before this and invites the assumption. A 0.27.0 binary built from identical source reports 87.9% too, and an interleaved A/B over 20 runs per command shows the two versions indistinguishable:
map --view skeletondoctorcallersScope
The other five recall rows and the entire grep-comparison table regenerate byte for byte, as do all eight regenerated blocks in
EXAMPLES.md— so this is the only published figure that had drifted.🤖 Generated with Claude Code
https://claude.ai/code/session_01VXJrNh9MMrV6WkJbdZnZyP