chore: gitignore run traces + add startup benchmark harness#128
Merged
Conversation
Agent transcripts and the ghost compose packets (built from kypp memory) can carry prompt/memory content that must not go public; findings are distilled into docs/memory, not the raw runs. Ignores NEW *.jsonl/*.log/*.json under scripts/eval/segmentation/results and scripts/ghost. Already-committed result jsonls stay tracked (single-level globs don't cross into subdirs).
scripts/bench/startup.py drives configured startup cases (docker/libkrun × pty/ server agents), reads host-emitted session.started lifecycle events, and summarizes startup_ms + per-stage p50/p95. Stays outside pillbox proper so the measurement surface is stable across optimization PRs. --dry-run, --json, --csv, per-case warmup/repeat; auto-removes each --detach run after capture.
These result jsonls are {task,cond,trial,score,cost} scalars over public Exercism
katas — not traces — so they belong in the evidence archive alongside h1-h4, not
gitignored. Narrows the ignore to just run *.log + ghost scratch.
- control-serial-glm52-pov: refutes the H5 "inversion" — glm-5.2 chained σ̂ = 0.000
serially (== H4); the campaign's 0.317 was a parallel-driver artifact.
- h5-* (deepseek/kimi/glm-5.2): cross-model runs; σ̂ is a parallel-contention upper
bound (README caveat), but the mean lift is robust across all three (+0.41/+0.11/+0.41).
Also drops a 1-line aborted enum-control stub.
Homebrew's runner image now sets HOMEBREW_REQUIRE_TAP_TRUST, so `brew install libkrun` fails with 'Refusing to load formula libkrun/krun/libkrun from untrusted tap' — deterministic, breaks the libkrun job on every PR, unrelated to any code change. Add `brew trust libkrun/krun` between tap and install (Homebrew's own recommended fix; non-interactive; forward-compatible when tap-trust defaults on).
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.
What
Triage of the uncommitted
bench/eval/ghostpile.chore: gitignore eval + ghost run outputs — agent transcripts + ghost compose packets (built from kypp memory) can carry prompt/memory content that must not go public. Ignores NEW*.jsonl/*.log/*.jsonunderscripts/eval/segmentation/resultsandscripts/ghost. Already-committed result jsonls stay tracked (single-level globs don't cross subdirs).feat(bench): startup-timing benchmark harness —scripts/bench/startup.pydrives startup cases (docker/libkrun × pty/server), reads hostsession.startedlifecycle events, summarizesstartup_ms+ per-stage p50/p95. Stays outside pillbox proper so the measurement surface is stable across optimization PRs.Also deleted a stale root-level
PLAN.md(Ghost v1 plan, Status 2026-06-14 — superseded by memory + docs; it was untracked, so no diff).Note
Trace outputs are preserved locally, just no longer git-tracked. Findings are distilled into docs/memory, not raw runs.
Test
python3 -m py_compile scripts/bench/startup.py→ OK