Input
What you provide
- Named model candidates and versions
- Locked rubric, thresholds, and measurement version
- Release policy, approvers, and rollback owner
Models / Model Launch Check
Compare the same checks, review blockers and record the evidence for a ship, hold or rollback decision.
Input
Work
Output
The steps from intake to handoff, in order.
01
Confirm candidate versions, owners, and the intended release.
Candidate manifest
02
Lock the rubric and source slice before scores or rank are displayed.
Comparison matrix and source runs
03
Replay retained failures and keep each failed case visible.
Regression outcomes and affected versions
04
Name every requirement that failed or is missing.
Blocker owner and recovery
05
Show who can approve, who has approved, and what remains.
Approval chain and timestamps
06
Record the decision, its evidence, and the rollback plan.
Release decision record
What you receive
| Outcome | Work | What you receive | Program fit |
|---|---|---|---|
| Ship | Every required check passes. Every required signer has approved. A rollback path is in place. | Source-linked results, approvals, and a verified regression replay. | The release record covers the approved candidate, its checks, and the rollback path. |
| Hold | A required check fails, a signer has not approved, or evidence is missing. | The blocking criterion, its owner, and the next action. | The hold names what must be fixed. The candidate goes back through the same checks once it is. |
| Roll back | The decision brief recommends rollback, or a deployed release must be reverted. | The recommendation, its evidence, and the rollback deployment record. | A rollback recommendation cannot be overridden at the gate. The rollback runs as its own recorded deployment. |
Inspectable evidence
Release review with descriptive states, named owners, and next actions.
Interactive walkthrough Illustrative
Select candidates and a launch criterion to see where each release is ready, needs review, or remains blocked.
| Required check | Candidate A | Candidate B |
|---|---|---|
| Review required | Evidence ready | |
| Blocking issue | Evidence ready | |
| Review required | Review required | |
| Evidence ready | Evidence ready |
Criterion inspector
Same checksCandidates share the same rubric and measurement version.
Blocking evidenceA failed regression remains visible and source-linked.
Release effectEvery difference states how it changes the decision.
For teams that buy evaluation data
We supply the evidence: a locked evaluation plan, the cases your model has failed before, and matched comparisons of the candidates you bring. You decide to ship, hold or roll back. We don't train, deploy or control your model.
Which use the source allows.
One task, one ID.
A named, versioned set.
The recorded first attempt.
The candidates you bring, on the same checks.
What regressed, what held.
Ship, hold or roll back.
Training examples are kept out of held-out tests. Sealed tasks are never used to train. Where a scoped release process exists, this record is what Model Launch Check packages.