mirror of
https://github.com/bdbrown4/peltier.io.git
synced 2026-11-07 10:58:28 +00:00
Trust-layer generalization (zero-regression verified on tokei):
- TargetSpec [build] uses an explicit {out} isolation-dir placeholder
instead of injected CARGO_TARGET_DIR, plus optional sanitizer/
sanitizer_binary templates. Nothing assumes cargo anymore.
- verdict rebuild_pristine_baseline, sanitizer_check, and harnessd
run_verdict candidate build all use {out} substitution. A target with
no sanitizer build caps accepts at needs-human-review.
- tokei/comrak target.toml migrated to the {out} form (CARGO_TARGET_DIR=
{out} + nightly ASan sanitizer command).
cJSON C target (v1.7.19, MIT):
- harness.c bench/golden driver (trust-layer; agent can patch workspace
cJSON.c but not the rig), compiled with clang against the patched tree.
- 5.6 MB deterministic corpus (gen-corpus.sh) + MANIFEST/GOLDEN pins.
- gates = hermetic Unity/CTest suite (19 tests pass); sanitizer =
ASan+UBSan; scripts/diff-fuzz-cjson.py for differential fuzzing.
coz causal profiler integration (SPEC §12):
- harness.c optional -DHOTPATH_COZ throughput progress point.
- scripts/install-coz.sh, coz-profile.sh, coz-summary.py; just coz/install-coz.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
targets/
One directory per optimization target. This is the only agent-writable path in the repository (SPEC §10); everything else is trust layer.
Layout
targets/<name>/
target.toml # upstream repo URL, pinned commit, build & bench commands
equivalence.toml # optional; absent = byte-identical required (SPEC §3.2)
workspace/ # agent's checkout — patches land here via harness git-apply
Selection criteria (SPEC §6): CPU-bound (>70% user time), meaningful test suite, scriptable workload, permissive license, active but not hyper-optimized, builds cleanly in a container. Good hunting: codecs, parsers, serializers, image processing, compression, static-site generators, linters.
equivalence.toml
Default (no file) is byte-identical outputs. FP-producing targets may declare an explicit tolerance policy:
mode = "fp-tolerance"
abs = 1e-9 # max absolute difference per numeric value
rel = 1e-6 # max relative difference per numeric value
Declaring a tolerance covers output comparison only; FP flag changes still route to needs-human-review (SPEC §8).