# Portfolio III defense — learner and reviewer template

Status: NOT READY. Blank fields and supplied examples are not observed evidence.
Reviewer/self-review context, date, tested source revision and environment:

## Evidence index

For every row link to a local artifact/test transcript and state its scope.
Keep private source, quarantine, traceback detail and credentials local.

| Claim | Actual evidence | Assistance | Status/gap |
| --- | --- | --- | --- |
| Specification predicts normal/refused behavior | | | |
| Own integration uses returned IDs/revisions and read-back | | | |
| Actual quarantine inspection and explicit metadata basis | | | |
| Strict input and actor/state/stale refusal | | | |
| Actual restart preserves finding and ordered audit | | | |
| Actual not-ready/timeout/programming fault stays honest | | | |
| Fresh kit/environment/tool setup works as documented | | | |
| NEW output refuses collision; partial effects disclosed | | | |
| Architecture/known limitations match actual code | | | |

## Live transfer, not a rehearsed example

Ask the learner to predict first, then demonstrate and explain:

1. Change asset/metadata/criterion/evidence. Retain a machine/human disagreement
   without inventing a review verdict. Identify supplied versus measured fields.
2. Reject a submitted finding using a distinct fixture reviewer; demonstrate
   refused self-review and stale revision without extra successful audit events.
3. Make an owned dependency unavailable, observe not-ready, repair only that
   fixture and observe recovery. Distinguish an owned child bug from timeout.
4. Stop/restart the service and read current finding/audit; explain why metrics
   reset while SQLite persists. Do not reinitialize an existing database.
5. Add one requirement with a new authored test, explain the smallest change,
   show the initial failure and repaired run without delegating to a supplied
   completed function. Name what stays untested.

## Rubric — keep evidence status honest

Evaluate correctness, failure handling, explanation, fresh transfer and assistance
disclosure separately: missing / attempted / guided / independently demonstrated.
No average converts a missing critical criterion into a pass. A reviewer must
actually observe the live transfer and record what was seen; a self-recorded
walkthrough is useful self-report, not independent reviewer verification.

Automated tests verify declared behaviors, not authorship, learning efficacy,
production security or real human judgment. Quiz answers are knowledge evidence.
Do not label the course/portfolio independently defended without actual artifacts
and observation. Record unresolved gaps, next action and any reviewer decision.