Passports and evidence
Every managed run comes back with a signed Evidence Passport: a record of what was run, what was checked, and what the checks found. It’s what you give someone who needs to judge the result without rerunning it.
A local run produces no Passport. See Running: local or managed.
It isn’t a score
A Passport is not a single validity number, and REVIEW is a real answer
rather than a failure. Read the caveats and prohibited claims as written.
A Passport runs a named set of checks and reports each one as PASS, WARN, FAIL or NOT_APPLICABLE. The overall decision follows from those:
| Any check | Decision |
|---|---|
Any FAIL | BLOCKED |
Any WARN, none failed | REVIEW |
All PASS | READY |
READY means every declared check passed. No automated check can tell you whether the science is right.
Quality profiles
The quality profile a run selects decides which checks run. Issued profiles are immutable, so a change in what a check means becomes a new version and Passports already issued keep their original meaning.
| Profile | Adds |
|---|---|
research-ready@1 | Provenance, requirements, schema, runtime, source, test, caveat and artifact checks |
research-ready@2 | Per-connection compatibility — Passport schema 1.1 |
research-ready@3 | Declared output acceptance, binding the recorded spec hash and expected check IDs to the immutable Lab manifest — Passport schema 1.2 |
ligand-ranking-early-prioritization@1 | Completeness and provenance for small-set Boltz-2 early-prioritization runs |
What the research-ready checks look at
| Check | Looks for |
|---|---|
immutable_subject | The run names an exact, immutable release |
manifest | A manifest is present |
intended_use | The model declares what it is for |
requirements | An MRS is present |
technical_specification | An MTS is present |
evidence | Evidence material is present |
tests | The package ships tests |
caveats | Caveats are declared — their absence is a FAIL, not a warning |
run_completion | The run finished |
artifacts | The declared artifacts exist |
Missing requirements, specification, evidence or tests each raise a WARN, which is what drives a run to REVIEW. This is the qualification track showing up per run. A model with no MRS and no MTS still runs, and its Passport says so every time.
Signing
Passports are signed with Ed25519 and carry the signing key ID, so anyone can verify one later without trusting whoever handed it to them.
Reading one
From an agent, fetch it with passport_get once the run finishes; see Agent tool reference. In Studio it sits on the run. Read the caveats and prohibited claims before you quote any number from it.
A finished run means the software ran. It says nothing about efficacy, safety or clinical guidance, whatever the Passport decision says.
Next steps
- How compatibility works: the run record behind the compatibility check
- Check scientific outputs: declaring the acceptance checks
research-ready@3binds - Qualify a model: MRS, MTS and the Hub qualification states