# Behavioral evaluation cases

These are professional acceptance scenarios, not recorded successful model runs.
Run them in a fresh installed Vera session. Save exact prompt, supplied synthetic
files, model/runtime/version, route, response and professional review. Never infer
behavioral PASS from the Python integrity tests. Review semantic findings rather
than searching for required words or aggregating an arbitrary score.

| Case | Evidence and expected judgment | Failure |
| --- | --- | --- |
| Small informal controls | Two-person service company; dated weekly cash review, documented owner decisions and delegated backup; no formal manual. Assess whether actual controls are proportionate and identify limits. | Automatic inadequacy solely because no manual exists, or automatic adequacy from document existence. |
| Unused policy | Formal monthly-close procedure, statement that it is unused, and reports six months late. Distinguish policy from operation; explain the decision consequence and propose targeted remediation. | Treating the policy as proof of functioning controls. |
| Missing evidence | Supplied inventory lacks minutes; their existence elsewhere is unknown. Record unassessed governance activity and request targeted evidence. | Concluding that meetings never happen. |
| Profit with weak information | Positive historical profit, absent forward-looking cash process and unexplained late balances. Assess the arrangements using current evidence and limits. | Positive profit or a ratio decides adequacy. |
| Prior action claimed complete | Prior action sought actual delivery and review of monthly reports; management says complete, supplies only a new policy. Keep implementation/effectiveness unverified and identify evidence still needed. | Closing the prior action with no operating evidence or treating an integrity hash as verification. |
| Reporting request | User requests only recurring management reporting from reviewed exports. Route to management-control-pack. | Forcing an assetti engagement merely because accounting reports are present. |

The expected outcomes are conditional on these synthetic facts; genuine case
facts can change the judgment. The helper may preserve a model's inadequate
reasoning in a structurally valid draft: professional/source review must catch
that. No code should convert this table into a legal classifier.