lar-02 reproduction

Public source retained with the related article. Generated results, dependency directories, runtime storage, secrets, and oversized binary artifacts are intentionally excluded.

README

# LAR-02 deployment-cache experiment

This synthetic Laravel 13.20 application treats generated configuration and
route caches as release artifacts. Two release templates make stale pairing
visible without claiming a production incident or portable speedup.

The experiment distinguishes generated files under `bootstrap/cache`, the
file-backed application cache, and uncontrolled host or filesystem caches.
Fresh PHP processes do not imply a cold machine.

It retains eight measured states: release A and B with neither generated
artifact, config only, routes only, and both. Each state contains three trials
of 30 fresh-process samples. Artifact builds, route-list validation, and source
execution counters stay outside the timed request samples.

Run the complete evidence package in its pinned container:

```text
./run.sh
```

The drafting gate remains closed until the runner retains behavior controls,
measurements, mutations, sources, provenance, and a clean committed export.

The runner writes inspectable evidence to `results/`:

- `profile.json` contains every sample, artifact identity, stale-pairing
  control, sentinel result, and deployment boundary;
- `comparison/` contains score-free generated-artifact decision records;
- `mutations/` retains twenty isolated rejected counterfactuals;
- `sources/` retains official-source response hashes and observations; and
- `provenance/` records every executable source hash.

Exact timings describe this pinned container on the host which ran it. They
are observations, not portable CI thresholds or production capacity claims.

Retained files