Experiment: 014a8a0
rad experiment show 014a8a0e6d4546a6a5bfefcde4d33d52c56418c8
by ee@did:key:z6MkvsiybCuk1WDZhh2aXDx7NTiZejKgWZygMNsUuXDMNvVQ · Apr 11 15:03 2026
cache config_at_commit+is_ancestor_of per tip in branches_count loop
Measurements
MetricBaselineThe benchmark measurement of the unmodified codeCandidateThe code with the proposed optimization appliedDeltaThe performance change from baseline to candidate
index_page_latency (ms) primary1546.931 ms (n=1)1559.577 ms (n=1)+0.81% REGRESSED
index_page_bytes (bytes) secondary83287.000 bytes (n=1)83283.000 bytes (n=1)0.00%
Annotations
hypothesisbranches_count loop re-reads optimize.yaml and runs is_ancestor_of redundantly across peers; cache per tip
next_action_hintstop optimizing branches_count until we profile where real time is going. Start with a coarse timing split: wrap the repo loop vs branches_count vs ticker assembly in Instants and print to stderr once. Or just try hoisting obvious allocations in the inner experiment loop (name/description clones per ticker item).
rollback_reasonlatency regressed +0.8% (1559.6 vs 1546.9 ms); variance tripled. Cache miss rate must be near 100% — local 51-repo profile has distinct branch tips per peer.
what_workednothing
Base CommitThe starting commit before the optimization was applied6b56f68f5d6ede368965f6b7839f77105d126a14
Candidate CommitThe code with the proposed optimization applied6b56f68f5d6ede368965f6b7839f77105d126a14
Configoptimize.yaml
Diff
No changes
Environment
Buildtrue
Teststrue
Archaarch64
OSDarwin 14.6.1
CPUApple M2
Agentpi-autoresearch / pi
Files