Artifact · Gate transcript
A gate transcript, annotated
A gate is an automated check that work must pass before it moves forward in the factory, and the transcript is the logged record of every check it met on the way. This is one work item's transcript: a single generated journey spec for the Bugzilla port (a bug update notifies its watchers), from catalog projection to a green suite run. The margin notes are ours; everything in mono is the machine's.
Lightly redacted: internal run and station identifiers are masked
(████).
Gate names, verdicts, details, and timings are verbatim. The source citations point
into Bugzilla's public code, so they need no masking and you can check them yourself.
11 gate entries · 2 fail
The run
Each row carries a kind label: det marks a deterministic gate (a grep, a compiler run) whose verdict is mechanical, while llm marks a gate judged against a written rubric, where anything under 90 fails. Ten of these eleven entries are deterministic. The two red ones stay in the record; a transcript with no red entries reads as theater.
- scenario-scoped-ids det ✓ pass 1.4 sAnnotation: The scenario is projected deterministically from the behavior catalog; this gate checks that every generated test still carries its
@catalog:<id>tag back to the behavior it covers. - mapfromevent-namespace det ✕ fail 380 msBugFieldUpdated handler subscribes to a short event name; expected bug.Events.BugFieldUpdatedAnnotation: The read-model gate rejects short event names. A handler that subscribes to
BugFieldUpdatedinstead ofbug.Events.BugFieldUpdatedwould silently never catch up, so it's a hard failure. - mapfromevent-namespace det ✓ pass 410 msre-run after namespacing; the handler resolves to bug.Events.BugFieldUpdatedAnnotation: A failed gate sends the work back to the station that produced it, and the re-run is logged beside the failure rather than over it.
- subscription-events-exist det ✓ pass 520 msAnnotation: Every event a handler subscribes to must be one some aggregate actually emits. A subscription to a non-existent event never reaches a test run.
- handler-casing det ✓ pass 95 ms
- assertion-floor det ✕ fail 240 msscenario asserts the command succeeded but not the notification it emits; below the floorAnnotation: Rejects a test that only checks the command returned success. A body that compiles but asserts nothing is the most expensive kind of green.
- assertion-floor det ✓ pass 230 msre-run after asserting the notification event reaches the watchersAnnotation: A failed gate sends the work back; the re-run is logged beside the failure, not over it, so the rework stays on the record.
- scenarios-typecheck det ✓ pass 6.2 sAnnotation: The full TypeScript check, emitting nothing. Type errors stop the line here instead of surfacing in review.
- result-type-shape det ✓ pass 180 msAnnotation: Failures must throw a typed
DomainErrorrather than encode success-or-error in the result shape, so callers can't read a failure as a success. - judge-score llm ✓ pass 9.4 srubric 94/100Annotation: The one LLM-judged gate in this slice. It scores the filled test against the journey spec's rubric, and it only grades; it never edits the work.
- scenarios-pass det ✓ pass 41.7 sAnnotation: The finished suite runs. The work item doesn't leave the station until it's green.
What this transcript does and doesn't show
Ask any vendor for their gate logs.