feat(context, repo): consistency of repeated runs; pass^k and path stability in repo compare by napetrov · Pull Request #155 · intel/dokimi

feat(context): consistency — result, path and contract stability of repeated runs

napetrov

committed

1626b1f

feat(context): consistency reads shell steps and blocked Stop hooks

napetrov

committed

dcf2b46

fix(context): consistency blames a failed chain on all its commands, distrusts piped checks

napetrov

committed

a23848b

feat(repo): compare pass^k and path stability of repeated cells

napetrov

committed

9f83a05

chore(context): CI fixes for consistency: header, nav, max_docs, fragment

napetrov

committed

8b8bd09

Say repeated cells come from repo run --reps in compare docs

antonwolfy

committed

5017d01

Merge intel/main into determinism-profile

napetrov

committed

a5ea88b

fix(benchmark): raise golden_manifest max_docs to 32

napetrov

committed

6148101

fix(context): consistency parses shell syntax, matches shell edits, refuses ambiguous outcomes; compare skips empty traces

napetrov

committed

911a507

Navigation Menu

Repository navigation

feat(context, repo): consistency of repeated runs; pass^k and path stability in repo compare - #155

feat(context, repo): consistency of repeated runs; pass^k and path stability in repo compare#155
napetrov wants to merge 17 commits into
intel:mainfrom
napetrov:determinism-profile

Commits on Oct 22, 2026

feat(context): consistency — result, path and contract stability of repeated runs

feat(context): consistency reads shell steps and blocked Stop hooks

fix(context): consistency blames a failed chain on all its commands, distrusts piped checks

feat(repo): compare pass^k and path stability of repeated cells

chore(context): CI fixes for consistency: header, nav, max_docs, fragment

Say repeated cells come from repo run --reps in compare docs

Merge intel/main into determinism-profile

fix(benchmark): raise golden_manifest max_docs to 32

fix(context): consistency parses shell syntax, matches shell edits, refuses ambiguous outcomes; compare skips empty traces

Commits on Oct 23, 2026

fix(context): a saved test log is not a shell edit

fix(context): count a trace given twice to one arm once

test(context): pin outcome keys, verdicts and the 911a507 shell guards

Merge intel/main into determinism-profile

fix(context): read shell lines for the commands they run; failed runs count; contract errors are clean

fix(repo): compare lists too-few-cell tasks once and skips score tasks for pass^K

test(repo): the score task has no trace, so it counts toward path_distance's too-few-cells line

fix(context): type the message, content and usage of a stream record for mypy