Skip to content

Answer the mobility budget bullet and record two rejected arms - #252

Merged
aywrite merged 1 commit into
masterfrom
claude/slider-attack-sharing
Sep 15, 2026
Merged

aywrite merged 1 commit into
masterfrom
claude/slider-attack-sharing

Conversation

@aywrite

@aywrite aywrite commented Sep 14, 2026 •

Copy link
Copy Markdown
Owner

One commit, rebased onto master now that
#251 is closed.

The budget bullet asked whether mobility being over the 5% a leaf term is
allowed was worth fixing. It is answered: the term was made 6% cheaper across
the whole search and 6,000 games at 10+0.1 could not see it, so the 5% is a
rule of thumb rather than a constraint. The 8.48% the bullet quoted was stale
as well, predating the refit that priced all four kinds, and the term reads
13.5% of the suite.

Two arms from the same round join the rejected list. Sharing the slider attack
sets between the evaluation and the generator was bounded at 0.385% against a
storage cost of 0.781% and never built. Lazy mobility was built and played at
three margins, reading +1 ±8 over 4,000 games at the best of them.

The working, the per margin instruction counts and the pairs to resume any of
the three tests from are in the planning repository. This carries the verdicts.

🤖 Generated with Claude Code

@github-actions

github-actions Bot commented Sep 14, 2026 •

Copy link
Copy Markdown
Contributor

Speed against 81093868

Both sides built and run on this runner in this job, the way
scripts/speed.sh measures a perf commit. The change is between
medians, and the spread beside it is what the change has to be
read against. The node counts are the search's: they move when
the search does, and a speed change leaves them alone.

round     base nps  candidate nps
    1      8755701        8645413
    2      8992708        8951409
    3      8961298        8965223
    4      8541771        8387665
    5      8775501        8877631
    6      8751126        8917511
    7      8866664        8630891
    8      8936997        8782473
    9      8878384        8893098

             nodes    time  median nps  fastest nps
base       3868987  0.44 s     8866664      8992708
candidate  3868987  0.44 s     8877631      8965223
change                           +0.1%        -0.3%

the change is inside the spread, so the medians make no claim; the
fastest column is the steadier comparison when the machine was not
quiet

Speed: +0.1% (bench nps, 9 interleaved rounds vs 81093868, spread 6.5%)

@aywrite aywrite changed the title Record sharing the slider attack sets as measured and rejected Record two measured negatives from the mobility speed round Sep 15, 2026
… arms

The bullet asking whether mobility being over the 5% a leaf term is allowed
is a problem worth fixing has an answer now. The term was made 6% cheaper
across the whole search and 6,000 games at 10+0.1 could not see it, so the
5% is a rule of thumb rather than a constraint and the bullet says so.

Its 8.48% was stale as well. That figure predates the refit that priced all
four mobility kinds, and the term reads 13.5% of the suite at `378c148`.

Two arms from the same round join the rejected list. Sharing the slider
attack sets between the evaluation and the generator was bounded before it
was built and the bound came out negative, because the stand pat cuts three
quarters of quiescence evaluations before they generate anything. Lazy
mobility was built, played at three margins and read +1 with an interval of
8 over 4,000 games at the best of them.

The working for both is in the planning repository, which is where the
per margin instruction counts, the skip rates and the pairs to resume any of
the three tests from are kept. This file carries the verdicts and the two
branches that hold the code.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@aywrite
aywrite force-pushed the claude/slider-attack-sharing branch from cc90d7f to a247576 Compare September 15, 2026 08:03
@aywrite aywrite changed the title Record two measured negatives from the mobility speed round Answer the mobility budget bullet and record two rejected arms Sep 15, 2026
@aywrite
aywrite merged commit 1a5e362 into master Sep 15, 2026
19 of 20 checks passed
@aywrite
aywrite deleted the claude/slider-attack-sharing branch September 15, 2026 12:29
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