Step 6 of 6 · Testing - practitioner kit
Regression Retest Record Template
Validates a fix with current evidence while keeping historical reconstruction status separate.
Use this template
A template is not evidence. Fill in sanitized identifiers only.
Purpose
Decides whether a finding can close, based on the control being exercised in the changed versions.
When used: After a fix, and after any regression trigger such as a model, prompt, policy, tool or catalogue change.
Reference
| Field | What to record |
|---|---|
| Regression ID | Unique identifier, for example RG-ASM014-03-1. |
| Original finding ID | Finding being retested. |
| Original case / run IDs | Runs that showed the failure. |
| Fix / change reference | Change ticket, commit or release. |
| Changed component / version | New versions of all affected components. |
| Expected fixed control | Decision that should now apply, with policy version. |
Execution
| Field | What to record |
|---|---|
| Positive control | Benign in-scope operation that must still succeed. |
| Regression stimulus reference | Fixture version or hash from the original run; no payload text. |
| Required evidence | Authoritative sources for stages 5 to 7. |
| Preconditions | State that must hold so the protected path can be reached. |
| Test runs | Run record IDs, at least three where model influence is involved. |
Status
Historical reconstruction and current validation are recorded separately.
| Field | What to record |
|---|---|
| Historical reconstructability status | Complete, partial or not reconstructable: can the original failure still be shown from retained evidence? |
| Current regression status | Fixed, still failing, inconclusive or control not exercised in the new versions. |
| Control exercised? | Did any run reach the protected path? |
| Previous failure reproduced? | In the new versions, yes or no, with run IDs. |
| Downstream effect observed? | From the system of record. |
| Evidence gaps | Missing or unavailable sources. |
| Residual risk | Remaining exposure and owner. |
| Reopen / close decision | Decision, decider and date. |
| Next trigger | Change that requires the next retest. |
Outcome guidance
| Outcome | Meaning |
|---|---|
| Fixed | Protected path exercised, expected control decision recorded, no unauthorized execution or effect, positive control succeeded. |
| Still failing | Authoritative evidence shows the unauthorized execution or effect in the new versions. |
| Inconclusive | Path exercised but required evidence unavailable or contradictory. |
| Control not exercised | The path was never reached, so the fix is untested. Do not close. |
| Historical evidence incomplete but current fix validated | Original failure cannot be fully reconstructed, but current runs meet the Fixed criteria. Record both facts. |
| Historical evidence complete but current retest inconclusive | The original failure is well evidenced, but the current runs do not support a decision. Keep open. |
How to use this template
- Reuse the original fixture version and preconditions; change only the fixed component.
- Run the positive control first.
- Assess historical and current status independently.
- Close only on Fixed, with the decider named.
Common mistakes
- Closing because the failure 'did not happen' while the path was never reached.
- Combining historical and current status into one label.
- Retesting with a changed fixture, hiding regressions.
- No positive control, so a deny-all fix looks correct.
- Forgetting to set the next trigger.