An adversarial audit found the enforcement story ahead of the code: gates
that were documented as hard gates were advisory, manual, or absent, and
several human-facing claims outran the machine record. This makes the
claims true.
High findings
- Differential fuzz was documentation, not enforcement: verdict hardcoded
fuzz_iters=0 and diff-test hardcoded the gate to Skipped, while the case
studies claimed 10k-iteration fuzz on the same wins. Fuzz is now a real
pipeline gate (strict FUZZ-RESULT contract, executed count recorded), and
it is inherently baseline-vs-candidate, so it runs on the accept path and
reports Skipped in the standalone `just gates` flow. New hard rule: a
machine `accepted` verdict is impossible unless the fuzz gate actually
passed — otherwise it caps at needs-human-review.
- The agent's system prompt claimed FP/concurrency/unsafe patches were
"routed to human review automatically". No such mechanism existed. Added
a conservative lexical risk classifier (crates/verdict/src/risk.rs) that
forces needs-human-review on any would-be accept carrying those signals;
fp-tolerance equivalence always signals.
- The local verdict path ran agent-influenced code unsandboxed. It now runs
under scripts/no-net.sh (network namespace) and fails closed. Upstream
test suites are hash-verified against corpora/<t>/TESTSUITE.sha256 when
present (scripts/pin-testsuite.sh generates them).
Also fixed, found while fixing the above
- The pipeline script's `set -e` was inert (brace group as the left operand
of `&&`), so a failing candidate build did not abort the run and verdict
could bench a stale binary. Now a scoped subshell whose status is branched
on at top level.
- A wrapper that exits before exec'ing (no-net.sh refusing, bad wrapper path)
stranded RUNNING.lock and left read_verdict polling forever. Failure
accounting now lives outside the wrapper.
- harnessd's patch allowlist ignored git rename/copy headers and --no-prefix
diffs; the rename-escape case is now a committed regression test.
- The workspace did not compile on Windows (ungated Unix APIs). cfg-gated;
71 portable tests now run on both platforms.
- pin-check regenerated the corpus manifest instead of verifying it; it is
now verify-only, with a deliberate `just pin-corpus` to re-pin.
- Max RSS is captured per run on Unix; PMU counters are opt-in diagnostics
(compare --perf-stat), never part of the accept decision.
Docs now match the code, including what is NOT yet enforced: no target ships
a TESTSUITE.sha256 pin, no target declares [build].tsan, and matmul is not
wired to the fp-tolerance gate — each is stated as mechanized-but-inert
rather than done. Corrected the stale "33 ledger rows" (34), the playbook's
class-3 cJSON win (rejected), adversarial-review.md's sanitizer-separation
claim (five accepted rows carry sanitizers_clean=false), the unsupported
"40,000 requests" figure, and the blanket "historical rows record
fuzz_iters=0" (8 of 10; phase0-tokei-002/-003 record 4,332).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- docs/: an mdBook documentation site — thesis, architecture, measurement
discipline, equivalence gates, agent loop, isolation/anti-reward-hacking,
playbook, ROI, reproduce, roadmap, and the five case studies (pulled from
their single source in results/ via {{#include}}). Builds clean (21
pages).
- .github/workflows/docs.yml: builds mdBook and deploys to GitHub Pages on
push to main (one-time setup: Settings → Pages → Source = GitHub Actions).
- README.md: rewritten as the project front page — the thesis, the verified
results table, layout, quick start, how it stays honest, links to the
book. Replaces the stale Phase-0-era text.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
- bench-runner service mode (SPEC §3.1 mode c): a coordinated-omission-
correct, open-loop, fixed-rate load generator; latency measured from
the intended send time so queueing is counted, not hidden. Interleaved
A/B of two server binaries, exact percentiles (bounded sample, no HDR
bucketing error), bootstrap p50/p99 CIs via the existing ratio-CI
machine. service-calibrate: A/A false-positive + injected latency-
regression detection (calibration in progress).
- targets/cjson/service.c: minimal single-threaded HTTP server wrapping
the patched cJSON.c (trust-layer, outside the workspace allowlist);
env-toggled per-request busy-wait for the injection calibration only.
- report generator (crates/report bin): reads a ledger row +
optional service-latency JSON, renders throughput→cores→dollars and/or
latency percentiles, every figure with its CI, pricing source, and the
methodology printed inline (SPEC §9); flags any non-accepted row so a
rejected change never reads as a committed saving.
- ledger::report_row; justfile service/service-calibrate/report.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
phase3-cjson-002 (class 5+3): stack-buffer number parsing + strtod
round-trip replacing sscanf. +8.85% median, 95% CI [+7.5%, +10.0%] on
the C target, banked. 100% audit: byte-identical golden replay,
independent ASan+UBSan clean, 10k-input differential fuzz 0 divergences,
patch bounds/free-gating verified by eye.
- verdict: an infrastructure sanitizer-BUILD failure (missing runtime,
broken toolchain) now caps the accept at needs-human-review and logs
it, instead of crashing the verdict and losing a measured bench.
- results/phase3/: four public case studies (tokei class-5 wins; build
flags + allocators; the comrak caught false-accept; the cJSON
cross-language proof) + README. Compound-speedup claim stated as an
approximation, not a measured figure (CIs don't multiply).
- .gitignore: cjson build-artifact dirs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
~22% of Ir in glibc float serialization (__printf_fp, cJSON print_number)
and parsing (strtod/str_to_mpn, cJSON parse_number), plus parse_value/
parse_string and malloc/free churn. Ample headroom (SPEC §6).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
The Claude Agent SDK loop (agent/hotpath_agent/loop.py, model
claude-fable-5) drives the seven harness tools over a standalone stdio
MCP server (mcp_server.py) and runs the full profile -> hypothesize ->
patch -> gated-verdict cycle on comrak with no human in the loop.
Three audited attempts, all honest rejections, ZERO false accepts:
- phase2-comrak-001 class 3: syntect buffer with_capacity, [0.971,1.007]
- phase2-comrak-002 class 5: gate dead VecDeque build, [0.982,1.026]
- phase2-comrak-003 class 6: hoist find_special_char check, [0.970,1.007]
The agent picked the cheapest untried class each time (3->5->6) and
TWICE refused to fabricate a verdict it couldn't read back, verifying
ledger integrity instead — the anti-reward-hacking behavior SPEC §10
targets.
Harness changes surfaced by the loop:
- run_verdict is now async: launches the multi-minute build+bench
detached (past the 60s MCP transport cap) and returns immediately;
new read_verdict op polls the append-only ledger row (ledger
verdict_summary). Fixes attempt 001's unreadable-verdict gap.
- read_target_source takes offset/limit line windows so 100KB+ files
page instead of overflowing — removes the agent's reason to reach for
a shell (attempt 003 paged a 2400-line file cleanly, no shell).
Honest boundary note (case study + code comments): in THIS dev
container the SDK tool allow/deny layer is defense-in-depth, not the
trust boundary — a nested CLI as root under a parent session doesn't
reliably gate built-ins. SPEC §10's boundary is OS process/user
isolation; that hardening is the remaining Phase 2 exit item
(docs/infra.md). What holds regardless: propose_patch is the only
mediated write path (path-allowlist + git apply) and the ledger is
append-only.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
Adds crates/verdict ('just verdict'): runs the core gates on the
candidate binary, then an interleaved A/B bench vs the baseline with
the CI-lower-bound rule, and writes the append-only ledger row with
per-side median CIs, workload statement, and env fingerprint. Gate
failure short-circuits to rejected-gate; --needs-human-review caps an
accepted verdict (never upgrades). CPU pinning moves into accept.toml
(pin_prefix, machine-specific, applied to both sides identically).
The smoke test caught a real trust gap: pointing --baseline-bin at the
workspace build (phase1-verdict-smoke-001) measured patched-vs-patched
and returned null, because gate runs rebuild the workspace binary from
the patched tree. Fix: --rebuild-baseline stashes the patch, builds
the baseline into isolated targets/<name>/baseline/, and restores the
patch — SPEC 3.1's pristine-checkout rule, now enforced by the
harness. Re-run (phase1-verdict-smoke-002) reproduces the Phase 0
result mechanically: +10.4% median, 95% CI [1.0932, 1.1137], accepted.
Both smoke rows are in the ledger, including the null one.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
Adds 'bench-runner calibrate': N A/A sessions counted against a <5%
false-positive bar, and N sessions with a synthetic slowdown injected
as a real busy-wait inside the candidate's timed window, counted
against a >=95% detection bar (detected = ratio CI entirely below 1.0).
Bootstrap seed varies per session so resampling noise is not
correlated across the very statistic being validated. JSON evidence
includes per-session CIs, injected wait, acceptance verdict, and env
fingerprint.
First run against the real comrak workload (20+20 sessions, pinned
core): false-positive rate 0.000, detection rate 0.950 -> PASS.
Evidence: results/calibration/2026-07-06-comrak-full-calibration.json.
Closes the SPEC 3.1 acceptance-criteria gap; remaining Phase 1 gaps:
no-net containers, perf profiling, diff-test orchestration, agent IPC.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
Attempt 4 (phase0-tokei-002, class 3, ACCEPTED — first auto-accept):
every file was read through encoding_rs_io's DecodeReaderBytes via
generic read_to_end, which zero-fills every grown buffer chunk (~221MB
memset/run, 8.5% of Ir) and grows without a size hint. Patch: fs::read
with a BOM-gated fallback to the identical decode path (a UTF-8-BOM
file transcodes invalid sequences, so the fallback is required for
equivalence). Gates: golden byte-identical; differential fuzz old-vs-
new binaries 4,332 iterations (BOM grid, invalid UTF-8, binary junk,
randomized) zero divergence; 250 upstream tests green; ASan+LSan clean
with ASan-run output hashing to golden. Bench: +9.9% [8.0%, 12.7%].
Attempt 5 (phase0-tokei-003, ship tree, ACCEPTED): stacks fat LTO+cgu=1
(real +1.6% standalone but under threshold, phase0-tokei-001) on the
accepted patch; benched as a whole vs pristine: +10.4% [8.5%, 12.0%].
Median clears the Phase 0 exit bar; CI lower bound 8.5%.
Case study written (results/phase0/case-study-tokei.md), ledger at 5
attempts, CLAUDE.md phase line moved to Phase 1, next hypotheses
logged (tokei class-5 byte-skip table; comrak class-3 arena).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
Commit the spec and scaffold the system it describes:
- SPEC.md: the full v0.1 engineering spec
- crates/: trust-layer Rust workspace, unit-tested
- bench-runner: interleaved ABAB scheduling with discarded warm-ups,
deterministic bootstrap CIs over the ratio of medians,
CI-lower-bound accept rule, env fingerprinting, and a CLI with
compare + A/A self-test modes (A/A verified passing end-to-end)
- diff-test: gate outcome model, sha256 corpus pin verifier that
refuses on mismatch, per-target equivalence policy parser
(byte-identical default, explicit fp-tolerance opt-in)
- ledger: append-only SQLite attempt schema with UPDATE/DELETE
refused by triggers, anti-double-attempt query
- report: cores-saved and dollars/yr ROI math over speedup CIs,
pricing.toml inputs
- playbook/: classes 1-7 with preconditions, procedure, verification
notes, and known failure modes per class
- agent/: Python package skeleton — six-tool contract and prompting
spine, harness IPC deferred to Phase 1
- config/: accept.toml (2% threshold, 30 runs/side), pricing.toml
- targets/, corpora/, results/calibration/: layout and policy docs
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EcKPWQ3MQD4NdEHwar96u2
benbrownandClaude Opus 4.8
