Pilot Dry-Run¶

Author: Techrevati doo

The pilot dry-run is a local smoke proof for the controlled 0.3.0rc1 pilot shape. It validates runtime wiring before the real downstream pilot starts. It does not replace the required controlled pilot or real rollback proof.

Command¶

python scripts/check_pilot_dry_run.py --output pilot-dry-run.json

The command exits non-zero if any local scenario fails. The optional JSON output can be attached to the restricted pilot evidence bundle.

Covered Scenarios¶

The dry-run covers:

• successful session,
• prompt-injection attempt,
• permission denial,
• guardrail block,
• max-iterations breach,
• max-tool-calls breach,
• provider failover evidence,
• checkpoint resume,
• sink failure diagnostic,
• rollback readiness command shape.

Rollback readiness is intentionally limited to command-shape proof. The real pilot still has to install the previous known-good package and pass the downstream smoke test from the rollback proof checklist.

Evidence¶

The dry-run creates temporary SQLite files for events, usage, and checkpoints. It verifies that event and usage sinks can be fanned out, usage totals are recorded, checkpoints can be replayed by thread_id and idempotency_key, and sink failures are visible without stopping the session.

Do not use the dry-run output as a go decision by itself. The stable release still requires the controlled RC pilot evidence template, rollback proof checklist, remote CI, private release publication, and real downstream evidence.