Scenario cards: declare which scope fields hold a person handle, and drop only the turns a refused seed row actually feeds #30914

domain:devx#2 · marchtian · session_017ihL2ebbZkr2QAEG1x7UC8

Filing gate: ② follow-up from a contract review. Should-fixes S1 and S2 of the CONTRACT_REVIEW_TIER review 6683045112 on PR #30903 (#30882). That PR lands without them because neither causes a false grade today. ⛔ Not a claim. Triage sets the grade and the lane.

Business reason. The scenario runner is how the maintainer measures what an AI-built app gets right. PR #30903 makes the seed hand a user id to each owner field, and stops a refused seed row from failing questions it feeds. Two parts of that rest on inference rather than on what the card declares. A new card can be mis-seeded with no warning, and a refused row can hide more of the scorecard than it needs to.

S2: the handle fields are inferred, not declared

userHandleFields (packages/qa/scenarios/src/runner/orchestrate.ts, at PR #30903's head) treats as a person handle any direct persona-scope field that is not id and is in neither expect.lookups nor the select fields. That is right for all nine cards today: exactly account.owner, invoice.owner and task.assignee. A future card with a direct scope on an undeclared text or option field would be sent a user id in that field, and nothing at authoring time would refuse it.

Wanted:

• The card format declares the handle fields (for example a user kind on the scope rule, or a list on the card).
• The loader refuses a scope on a field that is neither declared a handle nor a known lookup or select.
• The run record lists the fields the seed rewrote.
• The loader is in flight in fix(scenarios): the loader refuses an operation its persona cannot reach, and two cards stop writing hidden fields #30846 (Scenario loader refuses an operation its persona cannot reach (row scope, hidden field or object); fix the three cards it catches #30802). Start after it lands.

S1: dependence on a refused row is object-level

seedDependence marks a question not measured whenever its object holds any refused row. On crm_account_with_contacts, refusing contact ct4 drops contacts_per_account as sdr, whose scope never reads ct4. This is never a false fail, and the scorecard cell shows the count, but it hides results the run could have graded.

Wanted: decide dependence by re-evaluating each question's truth over the seed minus the out rows, so that only turns whose expected answer changes are marked not measured. Pin it with the ct4 case.

Level: QA tooling only, Clause-②: no, Metadata-protocol: none.

Dedupe (open and closed): "userHandleFields" and "seedDependence" each hit only #30882, whose dev reports name them. New.

Related: #30882, PR #30903, #30913 (the sibling owner_id finding), #30802.

Reader: the domain:devx lane seat, from pm:queue, after PR #30903 and #30846 land.