Commit Graph

Select branches

Hide Pull Requests

main

archive/engine-change-scoped-evidence-pass

archive/engine-typescript

archive/fix-mechanisms

archive/trained-classifier

archive/unseen-eval

v0.1.0

v1.0.0

v1.1.0

v1.2.0-alpha.1

v2.0.0

v2.0.1

Mono Color

• 03d925fffa spec/steering: the decision strategy corrected: software features do not decide safety; the trace does, the pool measures and never trains main Marius CiepluchandClaude Fable 5.1 2026-10-14 19:40:12 +02:00
• 03b114cbeb feat(prover): delivery D: control flow decides dominance; the trace carries each guard's relation, effect and the operation's control context Marius CiepluchandClaude Fable 5.1 2026-10-14 19:01:07 +02:00
• 6c3d4a7d72 spec(access-control-prover): a guard whose passing branch alone holds the operation discharges it (effect branches) Marius CiepluchandClaude Fable 5.1 2026-10-14 18:28:10 +02:00
• 2f628b921b feat(prover): delivery C: the ordered trace with argument provenance for every examined pair Marius CiepluchandClaude Fable 5.1 2026-10-14 18:22:45 +02:00
• 52ecbb76ab spec(access-control-prover): control flow decides dominance; the trace carries each guard's relation, effect and the operation's control context Marius CiepluchandClaude Fable 5.1 2026-10-14 18:07:09 +02:00
• 12d09faeed spec(access-control-prover): the ordered trace with argument provenance is the prover's output for every examined pair Marius CiepluchandClaude Fable 5.1 2026-10-14 17:56:50 +02:00
• 17e07b36c4 feat(prover): delivery B: handler-named operations, Symfony, Livewire, Remix and Tornado entries, enforcement witnesses, memoised path walk Marius CiepluchandClaude Fable 5.1 2026-10-14 17:31:07 +02:00
• 7f1b820d5a style(prover-arm): format Marius CiepluchandClaude Fable 5.1 2026-10-14 17:03:35 +02:00
• 5980bf8377 fix(prover-arm): the byte table holds only files that were read, so a zero-byte read stays on the record and the summary refuses it Marius CiepluchandClaude Fable 5.1 2026-10-14 17:00:13 +02:00
• beb272a5ec feat(prover): access-control prover, delivery A: path records, Stapler and Jetty entries, Java operation rows, the prover and its scoring arm Marius CiepluchandClaude Fable 5.1 2026-10-14 16:50:04 +02:00
• e514551cb6 spec(access-control-prover): the checker gains an opt-in premise and the unguarded_path label Marius CiepluchandClaude Fable 5.1 2026-10-14 15:54:29 +02:00
• 74a529825d spec: access-control prover (requirements, gap analysis, design, tasks) Marius CiepluchandClaude Fable 5.1 2026-10-14 15:51:25 +02:00
• 9feb8f1e33 fix(hunt): a row's entry points are the ones in its file or a caller's file, not the whole repository's list Marius CiepluchandClaude Fable 5.1 2026-10-14 14:47:08 +02:00
• 81ae206978 fix(hunt): build output gets no candidate rows, a minified line is not scanned once per function, the screen checkpoints every change Marius CiepluchandClaude Fable 5.1 2026-10-14 13:24:45 +02:00
• 4cf9c85d22 perf(hunt): the caller search prepares each product file once per change Marius CiepluchandClaude Fable 5.1 2026-10-14 09:46:46 +02:00
• 86d70831c8 perf(hunt): the source producer walks and reads the product tree once per change, not once per row Marius CiepluchandClaude Fable 5.1 2026-10-14 09:35:16 +02:00
• 4063fd1562 feat(semantic): the hunter's semantic producer covers Java and PHP Marius CiepluchandClaude Fable 5.1 2026-10-14 09:35:16 +02:00
• 6fe46c3062 data: access-control training pool refrozen (e6a67487): 23 fix changes, 93 accepted locations, 23 repositories, 20 ordinary pushes per fix, floor 20, walk cap 1500; 11 repositories left out for short history Marius CiepluchandClaude Fable 5.1 2026-10-14 09:31:56 +02:00
• 30ad197d3f fix(bench): an unparsable file in one ordinary commit no longer ends a repository's history walk Marius CiepluchandClaude Fable 5.1 2026-10-14 08:42:25 +02:00
• 351d73079a ci: the host job keeps the hunt and semantic extras; a benchmark main no longer leaks GIT_NO_LAZY_FETCH into later tests Marius CiepluchandClaude Fable 5.1 2026-10-14 08:08:32 +02:00
• 17194b3df9 ci: pytest runs with the hunt and semantic extras kept; sandbox tests skip by cause where no bwrap exists Marius CiepluchandClaude Fable 5.1 2026-10-14 07:54:09 +02:00
• b35137ed1b ci: install the hunt extra so pytest exercises the hunter tests Marius CiepluchandClaude Fable 5.1 2026-10-14 07:36:16 +02:00
• 436e83de5e data: access-control batch 2 spot-checked; all 69 passing locations accepted (27 changes, reviewer mc) Marius CiepluchandClaude Fable 5.1 2026-10-14 07:36:16 +02:00
• ae837a6aa4 fix(types): numpy, pydantic and litellm are absent in CI's type-check environment Marius CiepluchandClaude Fable 5.1 2026-10-14 03:05:12 +02:00
• c8f7610313 fix(types): mypy clean again over src; stub-less libraries and the benchmark imports handled in pyproject Marius CiepluchandClaude Fable 5.1 2026-10-14 03:03:12 +02:00
• 0d7d3f0dc9 chore: format the whole tree and lint the scratch measurement scripts; CI has been red since 25ae4b4 Marius CiepluchandClaude Fable 5.1 2026-10-14 02:49:44 +02:00
• fbbc03f3ea measure: access-control feature table and separation screen (952 rows, 74 positive rows, 5 repositories); first ablation manifest registered Marius CiepluchandClaude Fable 5.1 2026-10-14 02:45:35 +02:00
• dee23c23f5 measure: both arms over the access-control training pool (2 of 5 pushes, 4 of 48 locations each); runner books real cost Marius CiepluchandClaude Fable 5.1 2026-10-14 01:14:46 +02:00
• fcb43fdd47 fix(hunt): an empty product file no longer fails the source producer on every row of a change Marius CiepluchandClaude Fable 5.1 2026-10-14 00:53:37 +02:00
• c01e61c4dd fix(bench): the pilot hunter arm undoes an inherited sparse checkout before it reads a hotspot Marius CiepluchandClaude Fable 5.1 2026-10-13 23:18:44 +02:00
• 537b61e8d9 fix(hunt): diff producer records where a push removed lines; worktree checkout is proven complete Marius CiepluchandClaude Fable 5.1 2026-10-13 23:17:15 +02:00
• 60cbefec9b data: access-control batch verified, awaiting spot-check: 57 of 78 cloned (6 GB disk guard), 148 catalog sites, 69 pass all three checks in 27 changes Marius CiepluchandClaude Fable 5.1 2026-10-13 22:07:39 +02:00
• 6a9e4093fe data: 78 access-control advisories sourced from the OSV dumps (maven 48, npm 12, pip 11, composer 7), 237 candidate repositories after exclusions Marius CiepluchandClaude Fable 5.1 2026-10-13 21:53:00 +02:00
• 3d08d692ff bench(train): class-targeted sourcing, --family, from the public OSV dumps Marius CiepluchandClaude Fable 5.1 2026-10-13 21:46:45 +02:00
• 499f6c4474 fix(hunt): label every verified target of a change; undo an inherited sparse-checkout in the worktree Marius CiepluchandClaude Fable 5.1 2026-10-13 21:32:43 +02:00
• 30921a12b9 data: access-control training pool frozen (53f9f364): 5 fix changes, 48 accepted locations, 20 ordinary pushes per fix, floor 20, history limit 1500 Marius CiepluchandClaude Fable 5.1 2026-10-13 21:30:04 +02:00
• 1d1c87f63b bench(valid): a training freeze may cap the history walk at the newest N first-parent commits Marius CiepluchandClaude Fable 5.1 2026-10-13 20:47:23 +02:00
• 6f9613329e feat(hunt): candidates scored on folds through the runner; the pilot gain, once per accepted program (6.1) Marius CiepluchandClaude Fable 5.1 2026-10-13 20:44:18 +02:00
• bacd7a1e3f feat(hunt): proposals and the acceptance rule (5.1, 5.2) Marius CiepluchandClaude Fable 5.1 2026-10-13 20:22:22 +02:00
• 47d0f0fac1 bench(unseen): per-commit history cache for ordinary_history; assemble.freeze --history-cache Marius CiepluchandClaude Fable 5.1 2026-10-13 19:40:34 +02:00
• dbaaa86d5b feat(hunt): contamination check (4.5): runs that used a reported repository are marked and kept out of any gain Marius CiepluchandClaude Fable 5.1 2026-10-13 19:03:23 +02:00
• 6e5d060de6 feat(hunt): registered paired ablation, repository folds, leak audit, importance table and the keep/drop rule (4.2-4.4) Marius CiepluchandClaude Fable 5.1 2026-10-13 18:49:58 +02:00
• bbbb0f397e bench(valid): a training freeze may lower the ordinary floor to the ratio; the draw takes a size Marius CiepluchandClaude Fable 5.1 2026-10-13 18:33:10 +02:00
• f29b3a3df2 data: maintainer spot-check, 51 access control locations accepted (7 changes) Marius CiepluchandClaude Opus 5.5 2026-10-13 18:09:10 +02:00
• caa56387a1 spec(hunter-learning-loop): tick groups 1-3, implemented and measured Marius CiepluchandClaude Opus 5.5 2026-10-13 17:42:02 +02:00
• 2b65680d5d data: hunt training-set candidates (task 4.1), not frozen: 99 verified locations in 28 changes, awaiting spot-check Marius CiepluchandClaude Opus 5.5 2026-10-13 16:48:41 +02:00
• 25ae4b489c fix(p1,p0): restore the pre-P1 range contract for two callers; docs link and maintainer paths Marius CiepluchandClaude Opus 5.5 2026-10-13 16:15:32 +02:00
• a99579b547 measure: final push-hook arm on all 252 pilot changes (0/12, 2.1 false alarms per 100) and the combined whole-tool score Marius CiepluchandClaude Opus 5.5 2026-10-13 15:50:29 +02:00
• ec40dc1c93 measure: loop parse fix recovers 2 of 7 lost hunts; DSPy baseline 20/60 with the fix Marius CiepluchandClaude Opus 5.5 2026-10-13 15:45:56 +02:00
• 455637bea5 fix(hunt): a malformed tool-loop step ends the loop and the answer is still extracted Marius CiepluchandClaude Opus 5.5 2026-10-13 15:30:13 +02:00
• e07ab2ac93 measure: 5 paired repeats, DSPy hunt baseline 30% vs pilot hunter 52% recall on the 12 positives Marius CiepluchandClaude Opus 5.5 2026-10-13 15:29:51 +02:00
• 26720ed0dc measure: pre-registered analysis for 5 paired repeats of the hunt baseline against the pilot hunter Marius CiepluchandClaude Opus 5.5 2026-10-13 13:21:15 +02:00
• 3692c85d76 measure: hunt cache replay verified (zero tokens on rerun) Marius CiepluchandClaude Opus 5.5 2026-10-13 13:16:31 +02:00
• c2bfcef3df measure: hunt baseline on the 12 pilot positives Marius CiepluchandClaude Opus 5.5 2026-10-13 13:15:22 +02:00
• 602ec23e03 fix(hunt): lines read through a tool count as shown; an unparseable answer is an empty answer Marius CiepluchandClaude Opus 5.5 2026-10-13 13:10:40 +02:00
• c5bf52f47d measure: Semgrep arm on the valid pilot, our rules and facts converted Marius CiepluchandClaude Opus 5.5 2026-10-13 13:10:19 +02:00
• d08dea19b8 spec(hunter-learning-loop): a pattern engine's miss is never evidence Marius CiepluchandClaude Opus 5.5 2026-10-13 13:01:32 +02:00
• c95b9241d8 feat(stats): one statistics module on scipy, statsmodels and scikit-learn; engineered features tied to the annotations Marius CiepluchandClaude Fable 5.1 2026-10-13 12:43:02 +02:00
• d9d6335ab3 feat(hunt): feature table, producers, labels, metric, artifact, the DSPy hunt program, runner and CLI Marius CiepluchandClaude Fable 5.1 2026-10-13 12:33:58 +02:00
• e55e44d981 research: Semgrep CE and Opengrep as the quick tier, measured Marius CiepluchandClaude Fable 5.1 2026-10-13 12:21:02 +02:00
• 108dc2e5f3 measure: lizard and tree-sitter-language-pack against our function-span engines Marius CiepluchandClaude Fable 5.1 2026-10-13 12:18:57 +02:00
• 5d1d6f61d0 spec(hunter-learning-loop): implementation plan Marius CiepluchandClaude Fable 5.1 2026-10-13 12:15:45 +02:00
• dc2da3d83a spec(hunter-learning-loop): gap analysis and design Marius CiepluchandClaude Fable 5.1 2026-10-13 12:14:05 +02:00
• 062a7b780e refactor(p1): one source-facts module owns function ranges, callers, declaration tables and language aliases Marius CiepluchandClaude Fable 5.1 2026-10-13 12:00:37 +02:00
• f63703a604 chore(p0): drop never-enabled hunter tools, test-only functions and receiver delivery Marius CiepluchandClaude Fable 5.1 2026-10-13 11:06:09 +02:00
• f5a904b7f2 chore(p0): retire regress, the skill router, the module audit tool and vocabulary.sc Marius CiepluchandClaude Fable 5.1 2026-10-13 10:50:01 +02:00
• 94e96c6885 chore(p0): remove the improve lever and the plane improvement loop Marius CiepluchandClaude Fable 5.1 2026-10-13 10:34:57 +02:00
• d2010d0693 docs: Pathfinder map of src, duplication report, unified proposal, and the breadth ledger Marius CiepluchandClaude Fable 5.1 2026-10-13 09:55:21 +02:00
• e979fb0d95 chore: remove retired specs and dead code, state the 2026-10-13 focus Marius CiepluchandClaude Fable 5.1 2026-10-13 09:12:37 +02:00
• 3b8327ee51 spec(hunter-learning-loop): requirements Marius CiepluchandClaude Fable 5.1 2026-10-13 08:47:51 +02:00
• ad20405fc2 docs(benchmark): where the hunter's false alarms come from Marius CiepluchandClaude Fable 5.1 2026-10-13 08:35:04 +02:00
• 23aeeada10 feat(benchmark): combine --keep-verdicts drops hunter suspicions the classifier rejected Marius CiepluchandClaude Fable 5.1 2026-10-13 08:21:59 +02:00
• af7d3e49b1 docs(benchmark): pilot numbers, hunter arm final and push hook interim Marius CiepluchandClaude Fable 5.1 2026-10-13 08:13:02 +02:00
• abd82c6e33 fix(benchmark): meter the provider client before building the classifier Caller Marius CiepluchandClaude Fable 5.1 2026-10-12 20:47:12 +02:00
• d37d9634c6 fix(benchmark): classify hunter findings into the taxonomy before scoring arm (b) Marius CiepluchandClaude Fable 5.1 2026-10-12 20:42:08 +02:00
• 9f7ca132ff feat(benchmark): hunter arm runner and combiner (group 6) Marius CiepluchandClaude Fable 5.1 2026-10-12 20:29:03 +02:00
• bd7b7db25d docs(benchmark): drop two em-dash headings in the brief Marius CiepluchandClaude Fable 5.1 2026-10-12 20:18:05 +02:00
• 98825db94a docs(benchmark): R6 holds the push path to the LLM tier; measure hook as shipped and hook + hunter Marius CiepluchandClaude Fable 5.1 2026-10-12 20:17:44 +02:00
• 7a636d60fb feat(benchmark): freeze the valid-benchmark pilot, 12 verified positives + 240 negatives Marius CiepluchandClaude Fable 5.1 2026-10-12 19:42:45 +02:00
• 43ef254134 feat(benchmark): valid detection benchmark spec + offline modules (groups 1-4) Marius CiepluchandClaude Opus 4.8 2026-10-12 18:55:42 +02:00
• 3855945311 docs: rename benchmark metrics to standard security terms Marius CiepluchandClaude Opus 4.8 2026-10-12 18:24:13 +02:00
• 39a1873e5b docs: plain-English pass over the active specs Marius CiepluchandClaude Opus 4.8 2026-10-12 18:16:41 +02:00
• b08f372022 docs: retire the SZZ-pool metric, add the valid-benchmark brief, plain-English pass Marius CiepluchandClaude Opus 4.8 2026-10-12 16:38:18 +02:00
• c3f610f9f4 test(infra): drop the Dockerfile-text assertions from the engine-queue test Marius CiepluchandClaude Opus 4.8 2026-10-12 16:03:14 +02:00
• a36f99e15e test(infra): delete the change-detector Dockerfile tests, keep the real policy check Marius CiepluchandClaude Opus 4.8 2026-10-12 16:02:47 +02:00
• e49c150316 test(infra): fix the 3 stale/flaky infra tests to the current contracts Marius CiepluchandClaude Opus 4.8 2026-10-12 15:58:11 +02:00
• 485aacca02 wip(archive): fix-mechanisms arm — parked WIP preserved before worktree prune archive/fix-mechanisms Marius CiepluchandClaude Opus 4.8 2026-10-12 15:47:24 +02:00
• 6c789eac40 docs: correct G0 status (record 07 done on docker) + ax-lane is a G1 target Marius CiepluchandClaude Opus 4.8 2026-10-12 15:30:11 +02:00
• 9a3250f6c5 fix(corpus): advisory-pair repo-overlap -- remove real pilot leaks, skip vendored trees Marius CiepluchandClaude Opus 4.8 2026-10-12 15:22:55 +02:00
• 60ffff0168 fix(docs): add the two orphaned pages to mkdocs nav Marius CiepluchandClaude Opus 4.8 2026-10-12 14:52:42 +02:00
• 5ca1db7b7c docs(bug): close the blast-radius question for the taint single-query skew Marius CiepluchandClaude Opus 4.8 2026-10-12 14:27:31 +02:00
• a024656da0 fix(cpg): single-query path drops params the script does not declare Marius CiepluchandClaude Opus 4.8 2026-10-12 14:16:36 +02:00
• 8f7fc4ae83 measure(g2): banks fixed-search MVP -- "no wrong targets, only wrong searching" demonstrated Marius CiepluchandClaude Opus 4.8 2026-10-12 14:02:41 +02:00
• 7ab5c2ab2d measure(g2): proof-arm on the 3 fair hits -- push-time lift 0/3 (structural), proof is async Marius CiepluchandClaude Opus 4.8 2026-10-12 11:46:26 +02:00
• 04667c815e measure(g0): fair-target deep scan -- detection 3/5 on app-shaped+sound vs ~0 library-heavy Marius CiepluchandClaude Opus 4.8 2026-10-12 11:16:35 +02:00
• e15ed680a9 spec(g0): strategic pivot -- core problem is unprovable reachability; fix the eval target first Marius CiepluchandClaude Opus 4.8 2026-10-12 10:49:56 +02:00
• 27a7c12ca1 measure(g2): model hunter on sound targets -- first LLM catch the static tiers missed (1/2) Marius CiepluchandClaude Opus 4.8 2026-10-12 10:37:10 +02:00
• a43e6293f5 measure(g0/g2): LLM corpus-label audit -- only 47% of public pool labels are sound Marius CiepluchandClaude Opus 4.8 2026-10-12 10:09:32 +02:00
• 53b1d31d82 measure(g2): banks LLM sizing -- '2 guided by 1' validated; corpus label mislocated Marius CiepluchandClaude Opus 4.8 2026-10-12 10:02:18 +02:00
• 90e0098a94 spec(g2): root-cause engine 0/5 (library params not sources) + the '#2 guided by #1' workstream Marius CiepluchandClaude Opus 4.8 2026-10-12 09:24:06 +02:00