Repository navigation

Confirmatory runner: lanes, one engine pool per model, admission and the ceiling - #244

Merged

MaxGhenis merged 48 commits into

v2-microcosm-batchesfrom

v2-confirmatory-runner

Oct 31, 2026

Merged

Confirmatory runner: lanes, one engine pool per model, admission and the ceiling#244
MaxGhenis merged 48 commits into
v2-microcosm-batchesfrom
v2-confirmatory-runner

Commits on Oct 25, 2026
- Confirmatory runner: lanes, one engine pool per model, admission, the ceiling
  
  MaxGhenis
  and
  claude
  committed
  6e15d71
- Confirmatory runner: tests
  
  MaxGhenis
  and
  claude
  committed
  c04ffe5
- Confirmatory runner: create both logs at the start
  
  MaxGhenis
  and
  claude
  committed
  ec31e83
- Confirmatory runner: a clock that never goes back, and three test-oracle fixes
  
  MaxGhenis
  and
  claude
  committed
  808ccc8
- Confirmatory runner: the review's eight runner findings, stop causes, chains and anchors
  
  MaxGhenis
  and
  claude
  committed
  fdbd411
- Confirmatory runner: log excerpts from a day before the run's start
  
  MaxGhenis
  and
  claude
  committed
  06abd1f
- Confirmatory runner tests: the interrupt test without a timer
  
  MaxGhenis
  and
  claude
  committed
  f5099db
- Confirmatory runner: round-2 fixes, part 1 (not yet tested)
  
  MaxGhenis
  and
  claude
  committed
  f210b6e
Commits on Oct 26, 2026
- Confirmatory runner: round-2 fixes, part 2
  
  committed
  36d553a
- Confirmatory runner: round-3 fixes
  
  committed
  00d267a
- Confirmatory runner: round-4 fixes
  
  committed
  019f842
Commits on Oct 27, 2026
- Confirmatory runner: round-5 fixes
  
  committed
  9b30a77
- Confirmatory runner: round-6 wording and coverage
  
  committed
  4002f8e
- Run the no-tools stage with recoverable attempts and cumulative spend.
  
  committed
  047c868
- Confirmatory runner: wait for host-log samples in two timing tests
  
  MaxGhenis
  and
  claude
  committed
  060b287
- Confirmatory runner: coordinate no-tools shutdown and freeze inputs
  
  committed
  08d7b18
- Confirmatory runner: require standing no-tools finals before calculation
  
  committed
  5a9735b
- Confirmatory runner: coordinate interruption and worker cleanup
  
  committed
  c60086b
- Confirmatory runner: validate every saved final and baseline inventory
  
  committed
  248e777
- Confirmatory runner: keep exceptional cleanup signal-safe
  
  committed
  4ce7f8e
Commits on Oct 28, 2026
- Confirmatory runner: defer signals and track worker completion
  
  committed
  2f795ea
- Confirmatory runner: restore signal handlers before closing the wake pipe
  
  MaxGhenis
  and
  claude
  committed
  6597adb
- Confirmatory runner: let the forced-exit probe wait for the handler
  
  MaxGhenis
  and
  claude
  committed
  9acc920
- Confirmatory runner: cover watcher startup and pipe teardown
  
  committed
  a1114fe
- Confirmatory runner: qualify signal and cleanup guarantees
  
  committed
  89dd946
- Confirmatory runner: refuse undecodable files with their paths
  
  committed
  4dafb83
- Confirmatory runner: name board and cost refusals
  
  committed
  040d206
- Confirmatory runner: check the CLI with a real board
  
  committed
  6928b06
- Confirmatory runner: guard board rereads and strict saved costs
  MaxGhenis
  committed
  8b6c8f7
- Confirmatory runner: align saved treatment and link checks
  MaxGhenis
  committed
  0300257
Commits on Oct 29, 2026
- Confirmatory runner tests: build preflight finals the treatment checks accept
  
  MaxGhenis
  and
  claude
  committed
  4fb80d9
- Confirmatory runner: reproduce unreadable evidence refusals
  MaxGhenis
  committed
  5e334d9
- Confirmatory runner: bind decoded inputs and name evidence failures
  MaxGhenis
  committed
  d31c1d7
- Confirmatory runner: guard intermediate stage inventories
  MaxGhenis
  committed
  cccc381
- Confirmatory runner: name the new test modules for what they check
  
  MaxGhenis
  and
  claude
  committed
  4d93b40
- Confirmatory runner: refuse listed entries without metadata; one read per record
  
  MaxGhenis
  and
  claude
  committed
  0ae6df5
Commits on Oct 30, 2026
- Confirmatory runner: refuse an archive move before creating its folder
  
  MaxGhenis
  and
  claude
  committed
  c2ca2dc
- Confirmatory runner: refuse unfinished optional histories and special files
  
  MaxGhenis
  and
  claude
  committed
  274e1c7
- Confirmatory runner: format to CI's ruff 0.15.2 and wrap long lines
  
  MaxGhenis
  and
  claude
  committed
  d3673c3
- Confirmatory runner: pin the archive move's descriptor guard
  
  MaxGhenis
  and
  claude
  committed
  0193811
- Confirmatory runner tests: hold the deadline test's second turn for the abandon
  
  MaxGhenis
  and
  claude
  committed
  a8226dd
- Confirmatory runner: hold the preflight to the report's spend and baseline rules
  
  MaxGhenis
  and
  claude
  committed
  4441364
Commits on Oct 31, 2026
- Confirmatory runner: sum saved spend exactly, as the report does
  
  MaxGhenis
  and
  claude
  committed
  ceae93f
- Confirmatory no-tools: pin startup's exact spend at the stage cap
  
  MaxGhenis
  and
  claude
  committed
  3c80377
- No-tools startup: test the sum across models at the cap and the lock's release
  
  MaxGhenis
  and
  claude
  committed
  58ba64b
- No-tools startup: test costs whose sum overflows
  
  MaxGhenis
  and
  claude
  committed
  3846666
- Stub the host facts in the confirmatory tests' World
  
  MaxGhenis
  and
  claude
  committed
  2ee8acf
- Test the host facts that World stubs
  
  MaxGhenis
  and
  claude
  committed
  0ab2b1e

Commits

Commits on Oct 25, 2026

Commits on Oct 26, 2026

Commits on Oct 27, 2026

Commits on Oct 28, 2026

Commits on Oct 29, 2026

Commits on Oct 30, 2026

Commits on Oct 31, 2026