Repository navigation
Conversation
bajtos
added this pull request to stack #73
October 2, 2026 17:18
bajtos
force-pushed
the
claude/read-rates-verdicts
branch
from
October 5, 2026 17:41
3f82be8 to
b99064b
Compare
bajtos
force-pushed
the
claude/read-rates-verdicts
branch
from
October 5, 2026 18:03
b99064b to
78bde27
Compare
A pull request that changes ingot's read path should see whether read-back or restore moved, not only ingest. experiment.py adds each run's read rates to its status entry and compares read-back and restore the way it compares ingest: the median of the branch runs' medians against main's, judged against that stream's noise band. The ingest comparison keeps the top-level keys, so a status written before this change still parses; read_back and restore objects of the same shape sit beside them. A stream is null when a run has no positive median for it, and its p5 delta is null when a run lacks the p5, as for records from before the harness recorded read p5. No committed band covers the read streams yet, so they take fallbacks. Read-back takes ingest's ±3.5% median and ±11% p5, since it tracks the ingest median within about 2% in every tier 3 run. Restore takes ±34% on the median, twice the 17.2% coefficient of variation of the 19 valid tier 3 runs, and no p5 band, because no run has recorded a restore p5 yet. The comment renders a row per stream with the median and p5 deltas, the median noise band and the verdict, and the state line names all three verdicts. pr_run.py accepts a status with or without the read objects. Refs FIL-810 Assisted-by: Claude:claude-opus-5-5 Signed-off-by: Miroslav Bajtoš <oss@bajtos.net>
bajtos
force-pushed
the
claude/read-rates-verdicts
branch
from
October 5, 2026 18:23
78bde27 to
16ddd30
Compare
This branch has not been deployed
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
1 participant
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.
Written by Claude.
Give the
/forge-perfcomment a verdict for read-back and restore beside ingest.experiment.pyadds each run's read rates to its status entry. It compares each read stream the way it compares ingest: the median of the branch runs' medians against main's, judged against that stream's noise band.read_backandrestoreobjects of the same shape sit beside them.pr_run.pyrenders one row per stream (median delta, p5 delta, median noise band, verdict), and the state line names all three verdicts. It accepts a status with or without the read objects.No committed band covers the read streams yet, so they use fallbacks:
Read-back tracks the ingest median within about 2% in every tier 3 run. Restore's ±34% is twice the 17.2% coefficient of variation of the 19 valid tier 3 runs. Those runs used different sets, so it is an upper bound.
Refs FIL-810