9.1 Fix the same three task rows before comparing support configurations
| User task | Mandatory observation and stop | Configuration evidence |
|---|---|---|
retrieval by TraceReviewer-17 | recover MBE-EC417-B2-1, candidate status, prompt episteme ATP-2, the confidentiality boundary, and the D0 stop | one performed retrieval through SYS-EC417-PLM-1 |
tailoring by SafetyReviewer-17 | use a non-confidential fixture, preserve signed-before-closure, reject a stale edition, and stop on missing permission or authority | one admitted CI-use Work with the observations below; the tailoring input, action and result remain undefined |
branch selection by ReleaseDecider-17 | distinguish pre-entry A, bounded B2, and post-entry R; stop B2 when assignment, permission, authority, confidentiality, or reversibility is absent | one performed selection task through SYS-EC417-PLM-1 |
Published files and manual lookup are candidate configurations with no performed task evidence for these three rows, so those configurations remain untested rather than failed. Adding an interaction with SYS-EC417-AIProvider-1 and feedback Work or a feedback receiving relation also remains untested: no mandatory row needs either, and this support test contains no provider interaction, used AI suggestion, human review of such a suggestion, feedback Work, feedback SpeechAct, or feedback receiving use. The bounded PLM/CI candidate configuration is the only one with one observation for every current row. The described observations support retrieval, branch selection and the tested tailoring actions; the missing tailoring definition and permission/authority-stop observation prevent a pass for the full tailoring row or complete task set. The configuration is not established as globally smallest or superior to every alternative.