# Reproduction and availability

This unpublished package includes aggregates and figures, not raw private traces. No independent full-experiment reproduction is claimed. First-party ownership is recorded in OWNERSHIP_MATRIX.json; omitted raw evidence reflects publication scope, not generically unclear rights.

## Public candidate: aggregate consistency

From a local copy of this package:

```powershell
python -B PUBLIC_FIGURES/check_aggregates.py
```

This checks denominators, subtype totals, separate confusion tables, replay counts and release-file hashes. It cannot independently establish raw experiment authenticity, replay private episodes or regenerate model responses. Figure data carries source fingerprints for a future approved evidence index.

## Maintainer audit: private source access

With the existing repository available locally, choose a fresh output directory:

```powershell
& "$env:WINDWAY_REPO_ROOT/research/computerops_v0/.venv/Scripts/python.exe" -B PRIVATE_REVIEW/audit_release.py --repository "$env:WINDWAY_REPO_ROOT" --output "$env:WINDWAY_REVIEW_OUTPUT"
```

Set both environment variables to actual existing repository root and a nonexistent review destination. This private audit checks source hashes, rederives Reliability values and redirects ComputerOps's convenience audit receipt away from canonical evidence. It replays ComputerOps environment operations including failed cases. It performs no new model inference. PRIVATE_REVIEW is excluded from any public upload.

## Fresh model generation

Fresh inference requires a new experiment ID and fresh output root, pinned resource/configuration and explicit attempt/exposure accounting. Do not run historical resume/finalize commands against completed evidence. Temperature0 and requested seed 42 do not guarantee repeatable model outputs. A prior saved-action PASS is not proof of fresh-generation reproducibility.

## Public executable evidence

Pending separate explicit approval:reviewed source/task bundles and traces with a relative public namespace, separate private oracle/input boundary, new hashes for changed sanitized bytes and validated replay. No local private paths or imaginary download URLs are substituted.

E1=model observations; E2=declared recorded-action/state replay or aggregate derivation; E3=the later six-task local controlled experiment; E4=external independent reproduction. The controlled replication has scoped local E3;the original observations are not upgraded,and E4 is absent.


## Controlled replication update — scoped E3

36 fresh episodes across six tasks,three repetitions per condition:control18/18 verified and0/18 false completion;treatment15/18 verified and3/18 false completion;36/36 recorded-action replay PASS. `false_success__01` reproduced false completion3/3 in treatment and0/3 in matched control.

Local E3 applies only to this prospectively controlled experiment,not retroactively to observational Reliability or ComputerOps. The exploratory task-cluster95% interval [0,0.5] includes zero;no statistical proof of a general causal effect. No pooling,other-model,production-agent or market-wide generalization. The partial injector was versioned from historical batch-only behavior to workflow interruption after two of three durable records.

Selected authored content license:CC BY4.0. Publication approval remains pending. Raw private evidence stays excluded.