Comments only; no behaviour change, build/vet/gofmt clean. `startConsumer`'s doc block said "The task-16/17 seams are still nil ON PURPOSE" and described what each nil degrades to. Engine, Terminator and RemoteDeleter are all constructed and passed 130 lines further down in the same function, so the file contradicted itself and the newer comment beside the wiring said the opposite. The block now records that they are wired, and keeps the degradation contract as the rule that applies IF a seam is ever unwired again rather than as a description of today. Three sites claimed a "noop enqueuer federates nothing" when workers are at zero. No such type exists — the only occurrences of the phrase in the tree were the comments themselves. The truth matters operationally and is what the config doc now says: with the consumer on and OUTBOUND_WORKERS=0 the real enqueuer still persists every intent inside the gate transaction, so state accumulates and only the POSTing stops, and raising workers later drains a backlog rather than starting from empty. This is the fifth instance this session of the same defect — a comment asserting something the code does not do. It has cost a full cycle in a sub-run, produced a false compile-time guarantee that was then repeated in a commit message, and most recently propagated out of the worker into the operator runbook, where it inverted an emergency recommendation. The rule worth keeping: when a comment states a guarantee, find the line that enforces it, and if there is none the comment is the bug. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>