Repository navigation
Advice margin 5, tested on fresh seeds - #138
Merged
Merged
Conversation
Rule written first: margin 5 beats margin 2 head to head on seeds 10001..=10800 with a 95 % lower bound above 0. Result +0.245 (lower +0.187). Adds the margin example. Co-Authored-By: Claude <noreply@anthropic.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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
ADVICE_MARGINgoes from 2 to 5: the computer commander takes the lookahead's proposal only when it gains more than 5 points. The number was found by looking at seeds 1..=400 (plans/actors.md, step 4), so it was tested on seeds nobody had looked at.Rule, written before the run
Adopt 5 only if, on seeds 10001..=10800 (800 seeds, both sides, 1600 games a row), a commander with margin 5 beats one with margin 2 head to head, paired seeds, with a 95 % lower bound (margin - 1.96 error) above 0. Do not pick another number from the fresh table; report another margin that looks better as a lead for another fresh test.
Fresh seeds (
cargo run --release -p cognition_towers --example margin 800 10000)Verdict: the rule passes (lower bound +0.187 > 0), so the margin is 5. Margin 10 is level with 5 against no advisor and was not tested against 5 head to head: a lead for another fresh test, not a change. (Row 3 and row 7 beat margin 2 head to head while their score against no advisor is not above margin 2's; that is noise of about 1 sigma each, not explained.)
Before and after
advice 400(seeds 1..=400), margin 2 then margin 5:judge --game captain --champion timid --challenger default --seeds 24: before +0.384 (lower bound +0.048, promoted, 30 won 18 lost); after +0.453 (lower bound +0.215, promoted, 32 won 16 lost).Changes
05-commander,18-advice, glossary,domains/recreation/README.md,plans/actors.md.commander.rslikewise.margin [SEEDS] [OFFSET] [MARGIN...].Checks
cargo test --release -p recreation_towers -p cognition_towers -p recreation_towers_ground,cargo test --release -p recreation_towers_server --bins,cargo run -p expectation -- check expectations(0 errors),cargo fmt --checkpass.cargo clippy --workspace --all-targets: only vendor warnings (gpui, xattr), none in the repo's own code.🤖 Generated with Claude Code