Deliverable · empty template
This is the deliverable structure, not a client case, ranking or certification. The “—” cells are filled only after 10 paired scenarios and 20 valid executions.
Release decision dossier
Decision under test: [Can version B replace A for the agreed mission without introducing the frozen critical failure?]
Executive verdict
| Status | PASS · REVIEW · HOLD · INVALID |
|---|
| Main cause | — |
|---|
| Recommendation | — |
|---|
| Fixed / kept / broke | — / — / — |
|---|
| Critical failure | — |
|---|
| Not measured | — |
|---|
| Oracle verification level | Re-executed by ScoreIA / bound and checked in client-operated mode |
|---|
One critical regression may result in HOLD even if nine out of ten cases improve. The final deployment decision remains with the client.
Contract and configurations
| Field | Version A | Version B |
|---|
| Release | — | — |
| Configuration digest | sha256:… | sha256:… |
| Declared model and host | — | — |
| Tools / approval policy | — | — |
| Primary changed variable | — |
| Controlled variables | — |
| Identity assurance | participant_claimed | participant_claimed |
A/B decision matrix
| # | Scenario | Criticality | A | B | Classification | Evidence |
|---|
| 01 | — | — | — | — | fixed / kept / broke | card / replay |
| 02 | — | — | — | — | — | — |
| 03 | — | — | — | — | — | — |
| 04 | — | — | — | — | — | — |
| 05 | — | — | — | — | — | — |
| 06 | — | — | — | — | — | — |
| 07 | — | — | — | — | — | — |
| 08 | — | — | — | — | — | — |
| 09 | — | — | — | — | — | — |
| 10 | — | — | — | — | — | — |
Critical regression
| Expected behaviour | — |
|---|
| Observed event | — |
|---|
| Oracle rule | — |
|---|
| Business consequence | — |
|---|
| Linked evidence | — |
|---|
Campaign quality
| Measure | Result |
|---|
| Complete pairs | — / 10 |
| Valid executions | — / 20 |
| Contract and suite integrity | — |
| Receipt verified | — |
| Oracle verification | Re-executed by ScoreIA / bound and checked in client-operated mode |
| Incidents and invalid runs | — |
Correction plan
| Priority | Action | Owner | Case to rerun |
|---|
| P0 | — | — | — |
| P1 | — | — | — |
| P2 | — | — | — |
Evidence and verification
| Trial contract | — |
|---|
| Campaign ID | ptcampaign-… |
|---|
| Release Receipt | ptr-… |
|---|
| Bundle SHA-256 | sha256:… |
|---|
| Verification command | — |
|---|
Limits and reruns
Twenty executions are an exploratory diagnostic, not a statistical average about the model. Identity remains declared unless a stronger assurance level is explicit. One rerun of each of the three selected critical cases is included during the 30 calendar days following the review; the initial history remains intact.
Back to the offer · A/B pipeline (empty cells)