# Astria stale-graph observation

Publication date: October 16, 2026. Tool: installed Astria CLI 1.2.0. Platform: Windows, PowerShell. This is a one-file structural-extraction observation, not a performance benchmark or an independently evaluated retrieval dataset.

## Reproduce

Install Astria 1.2.0 and make `astria` available on PATH. Inspect `reproduce.ps1`, then run it with a new output directory:

```powershell
./reproduce.ps1 -OutputDirectory ./staleness-observation
```

The script refuses existing destinations. It creates and modifies only its generated fixture. No semantic backend or embedding option is passed. Do not run it from an environment configured to force paid enrichment. The recorded build used structural extraction only.

## Evidence layout

- `observed/`: saved final run, including exact CLI version, commands, process exit codes, durations, source snapshots, raw status/query responses, and graph projections.
- `pilot-unseeded/`: initial exploratory observations without explicit seeds.
- `pilot-seeded/`: second exploratory observations with explicit seeds.
- `reproduce.ps1`: script for the final protocol, adding graph projections to the seeded observation.

`graph-before.json` and `graph-stale.json` contain the old declaration. `graph-updated.json` contains the renamed declaration and its two-parameter signature. These are projections of graph.json, not the entire graph. They omit local absolute paths and unrelated metadata. Raw command output has local fixture/output paths replaced with placeholders. Embedded timestamps are the execution host's clock, retained unchanged, and are not the editorial publication date.

## Findings and limits

The source change was detected as one modified file. Status changed from fresh to stale and returned to fresh after update. The stale status command exited 0. The final run's build and update took 876 and 590 milliseconds respectively, including process startup and output capture. These are single observations after exploratory runs, not statistically useful performance estimates.

All corresponding queries returned zero nodes, including the fresh and refreshed cases. Suggestions reflected the stored declaration name. Adding seeds did not resolve empty retrieval. The cause was not diagnosed. No agent-generated answer was scored, no production repository was measured, and no result establishes general superiority or end-to-end savings.

The script reports the installed version rather than enforcing or installing it. Reproduce with the same version for the closest comparison, and retain any changed results from newer versions.