These case studies make a narrow claim: they document what Grinta did in recorded runs and distinguish runtime completion from broader specification compliance. Exact artifacts are linked wherever they are public; unavailable or private evidence is labeled as such.
| Case study | Result | Evidence |
|---|---|---|
| 4h 33m autonomous session | 16,393 events, 373 tool outcomes, one initial user turn, final state FINISHED |
Sanitized report and committed audit summary |
| Failure recovery | Failure output inspected, targeted repair applied, validation rerun without a follow-up prompt | Demo recording and disclosure notes |
| Issue-tracker build | Flask + React issue tracker produced and validated with 19 passing tests | Run summary; public transcript pending sanitization |
| Raft key-value store | Race-condition failure diagnosed; 39/39 tests passed consistently | Full demo and animated preview |
For the evidence standard used by the engineering narrative, see docs/journey/EVIDENCE.md.