Robotics evidence guide · automated editorial preparation, not expert review
Robotics evidence-package specification
This Maha-authored experimental contract binds one fixed synthetic task plan to complete ordered trials and recomputed metrics. It is deliberately not a general robot-certification format.
Required contents
- schema and domain
- maha-robotics-evidence/0.1; toy-simulation only. Unsupported versions or domains are refused.
- plan and planDigest
- Initial state, grid bounds, units, controller IDs, all six cases, action budgets, assistance flags, scoring and omitted capabilities.
- artifacts
- Digests identify the executing simulator functions and controller implementation. They are not signatures or independent attestations.
- trials
- Twelve ordered trials with IDs, controller, case, outcome, reason, action count and ordered state events. Failed and assisted trials remain in the denominator.
- metrics
- Per-controller autonomous, assisted and failed counts; the autonomous rate is an exact numerator/denominator pair.
- observer, rights, review and limitations
- Same-process observation, synthetic-data provenance and explicit exclusions including hardware safety.
- packageDigest
- SHA-256 of the canonical body excluding packageDigest. Canonicalization sorts object keys, preserves array order and rejects non-finite or non-JSON values.
Verification boundary
The verifier reruns the trusted fixed plan using local code and compares the entire canonical package. Merely rehashing an altered trace cannot make it pass. Unknown fields, omitted trials and altered scoring are refused. Verification proves agreement with this implementation, not its correctness or physical truth. A second independent implementation remains future work.
Function-source fingerprints depend on the execution toolchain. Reproduce with the same Node runtime and direct TypeScript execution; a transpiled build need not have the same fingerprint. No wall-clock timestamp, personal data or external telemetry is included.