benbrown/peltier.io
2
1
Fork 0
mirror of https://github.com/bdbrown4/peltier.io.git synced 2026-11-07 10:58:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

agent: phase2-tokei-001 — first auto-accepted win from the unattended loop

Browse Source 
Ledger row: class 5 (algorithmic) on tokei, first-byte gate table for
the per-byte token matchers in perform_multi_line_analysis.
Speedup median 1.107, 95% CI [1.084, 1.131] vs the banked-wins baseline,
workload: 221 MB / 4360 files, single rayon thread, pinned core.

Manual audit gates (pipeline covers tests+golden+bench only):
- Differential fuzz, 10k mutated inputs across 200 batched runs:
  0 divergences after canonicalizing tokei's benign parallel-walker
  output ordering (present in pristine baseline; counts identical).
  Fuzzer added as scripts/diff-fuzz-tokei.py.
- ASan run recorded in the case study (next commit).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
This commit is contained in:
Claude
2026-07-10 05:08:16 +00:00
parent 189939fe58
commit abaed3c49d
3 changed files with 138 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
results/phase2/transcript-tokei-001.txt
+18
View File
@@ -2,3 +2,21 @@
[agent] The hunk line counts were off. Recomputing and resubmitting:
[agent] Patch applied — `patch_id: 4e3da7e6bd1c`. Launching the gate+bench pipeline.
[agent] Still running. Polling again.
[agent] **Verdict: `accepted`** — ledger row written for `phase2-tokei-001`.
## Attempt summary
| | |
|---|---|
| **Run ID** | `phase2-tokei-001` |
| **Target / hotspot** | tokei — `LanguageType::parse_lines` 55.89% Ir / 62% branches / 44% branch-misses + `__memcmp_avx2_movbe` 9.37% Ir |
| **Playbook class** | 5 (algorithmic — precomputed lookup table tier) |
| **Patch** | `4e3da7e6bd1c`: 256-entry first-byte "interesting bytes" table in `SharedMatchers` (superset of all quote/comment token start+end first bytes, plus `\`), gating the five per-byte token matchers in `perform_multi_line_analysis`; `parse_context` still consulted on gated bytes so semantics are exactly preserved |
| **Verdict** | **accepted** |
| **Speedup (median)** | 1.1071× (~+10.7%) |
| **Speedup 95% CI** | **[1.0836, 1.1310]** — CI lower bound +8.4%, clear of the 2% accept threshold |
Measured on the tokei benchmark workload (`tokei --sort code corpora/tokei/input`), interleaved A/B vs pristine-rebuilt baseline, all equivalence gates green (upstream tests, golden replay, diff-fuzz, sanitizers — no FP/concurrency/unsafe surface touched, so no human-review routing triggered).
The hypothesis predicted 5–12% median; the measured +10.7% landed inside that band. Notably, this also satisfies the outstanding Phase 2 exit criterion of at least one auto-accepted win from the unattended loop. One attempt was requested; stopping here.
[result] turns=21 cost_usd=4.614901