spec(queen): claim.t27 - the round's stateOfDispatch as a card, the review valve imported, not restated #8765

Why

Every round, the trios supervisor decides for every queen_dispatch row whether the row still claims its issue: running, accepted, rejected, awaitingReview, failed or cancelled. The round, the board (dispatchRowState) and the files a row holds all read that answer. Today it is stateOfDispatch in trios/agent-server/apps/server/src/api/services/queen-tick.ts (gHashTag/trios queen @ 3654fdd2, lines 791-912), written by hand.

Two specs describe older versions of that rule, and neither agrees with the code that runs:

• specs/queen/dispatch.t27 claim_of: the audit at supervisor pin 879799a8. An escalation is held with no clock, obsolete/cancelled is not a state, and the row's criteria are not read.
• gHashTag/trios rings/T27-02/tick_core.t27 state_of_dispatch: four states, written 2026-09-04, never generated or run.

The review valve (specs/queen/review_valve.t27) already owns the two decisions the claim rule composes (escalation_kind, next_step). The supervisor mirrors them by hand.

What

A card, specs/queen/claim.t27 (module QueenClaim):

• claim_of(finished, verdict, idle_minutes, send_backs, ceiling, releases, criteria_read, criteria, free_attempts, reviewer_misses) -> u8: the rule as it runs on the queen branch, word for word.
• claim_of_step(step, held) and criteria_read_as(criteria_read, criteria).
• escalation_kind, next_step and the valve's kind/step codes and retry floor are imported from review_valve.t27, not restated.
• The verdict word stays at the boundary: the supervisor maps review_state to a V_ code and the CL_ result back to its word.

dispatch.t27 is not edited here: open #6934, #5452 and #7430 touch it.

Acceptance criteria

• t27c test-report specs/queen/claim.t27: every test passes, 0 vacuous, invariants proved.
• Negative controls: each mutant of a branch, floor or ceiling of claim_of either fails a named test or is blocked by an invariant; the table is in the PR.
• The seal .trinity/seals/queen_QueenClaim.json is written by the t27c built from the PR's own tree, with Zig 0.16.
• A dated docs/now/ note.
• The runtime half is gHashTag/trios (separate issue): the supervisor calls claim_of as generated wasm, proven equal to the hand-written function first.

Dmitrii Fedorov (@dmitrii-f-t27)

User Scenarios

• Given the supervisor needs to determine the claim state for a dispatch row, When the claim_of function is called with the row's parameters, Then it returns the correct state code matching the hand-written implementation in queen-tick.ts
• Given the review valve already defines escalation_kind and next_step, When claim.t27 imports these from review_valve.t27, Then the claim rule composes the same decisions without duplicating logic
• Given the verdict word stays at the boundary, When the supervisor maps review_state to V_ codes and CL_ results back to words, Then the spec only deals with internal codes

Requirements

FR-001: The spec specs/queen/claim.t27 MUST define module QueenClaim with function claim_of(finished, verdict, idle_minutes, send_backs, ceiling, releases, criteria_read, criteria, free_attempts, reviewer_misses) -> u8
FR-002: The spec MUST define helper functions claim_of_step(step, held) and criteria_read_as(criteria_read, criteria)
FR-003: The spec MUST import escalation_kind, next_step, valve kind/step codes, and retry floor from specs/queen/review_valve.t27 without restating them
FR-004: The spec MUST keep verdict words at the boundary using V_ codes for input and CL_ codes for output
FR-005: The command t27c test-report specs/queen/claim.t27 MUST pass all tests with 0 vacuous tests and invariants proved
FR-006: Negative controls for each mutant of branch, floor, or ceiling of claim_of MUST either fail a named test or be blocked by an invariant
FR-007: The seal .trinity/seals/queen_QueenClaim.json MUST be written by t27c built from the PR's own tree with Zig 0.16

Boundary

• specs/queen/claim.t27
• specs/queen/review_valve.t27
• .trinity/seals/queen_QueenClaim.json