# Equipment Desk: reproducible local case study

Cerrito, September 22, 2026. One AI assistant authored and tested this application against known scenarios. This is synthetic research, not a deployed product or an independent benchmark.

## Start a clean fixture

Python 3.14.7 was used; no third-party Python dependencies. From this folder:

```sh
python3 app/server.py --init
python3 app/server.py
```

Open http://127.0.0.1:8766. All local synthetic accounts use `research-only`. Start with Casey (employee), Lee (Engineering manager), Sam (Design manager), or Morgan (admin). Alex is deactivated by the automated scenario.

## Reproduce the ten checks

With the server running against a **fresh** fixture, in another terminal:

```sh
python3 run_trial.py
```

The trial changes policy, deactivates Alex, and writes results to evidence/custom-run.json. Do not run twice against the same database. For another run, unpack a fresh copy to a different folder and stop the prior server first. Ports 8766–8768 must be available. Preserve the included original evidence before running; the harness overwrites its output files.

The first implementation is snapshots/server-first.py. Original failure is evidence/custom-first-run.json. Repaired results are evidence/custom-run.json. The repair checks that retries retain their original equipment and note, rather than merely sharing a command key.

## What this does not prove

No real employees, long-term operations, usability study, security audit, production load, independent reviewer, measured human labor, or complete cost comparison. Fifty seeded accounts are not fifty concurrent users. Race evidence consists of two competing checkout operations. Restore evidence covers a second local process from a SQLite backup. Portable import resets passwords to the synthetic password and excludes sessions. The event history is application recorded, not tamperproof against database administrators.

The app intentionally binds only to loopback. Do not expose it to the internet or use real data. Known production omissions include SSO, credential management, rate limiting, monitoring, hosted deployment, and operational support.

## Vendor scope

Snipe-IT public demo v8.7.2 build 24589. Only our synthetic asset CERRITO-LAB-20260922-HTTP was created/changed. Checkout to demo Test User succeeded. Reopening checkout while assigned was refused. Administrator check-in succeeded, leaving one checkout, one check-in and History 3 in the UI. This was not a concurrent-write test of Snipe-IT. Detailed history API retrieval failed in our client; we do not claim inspection of its individual events.

Operations used authenticated HTTP web forms; browser screenshots verified visible state. This is not a timing/usability comparison. Demo resets and security restrictions prevented equivalent offboarding and recovery tests. Untested means untested, not unsupported.

Vendor source: https://snipeitapp.com/demo
Pricing source: https://snipeitapp.com/pricing (checked September 22, 2026, USD).

Screenshots of Snipe-IT show its public demo, credited to Snipe-IT / Grokability. Local app screenshots show our synthetic fixture.