the gap only the synthesis pass could see

Every lane checks its own work, and that is not review. It is consistency. A step that depends on a property holding across a whole range will pass every check run by a worker who has already assumed the property, because the assumption sits upstream of the check. Self-examination cannot find the assumption it is built on.

The digest recorded exactly this shape of defect: a gap in one lane's argument, found by another lane during a synthesis pass, with the record naming who caught it. The finder was not a supervisor. It was a peer reading the whole for a different reason, and that is why the pass exists as a scheduled step rather than as an appeal to diligence.

The tension is the one that parallel work always creates. Lanes are independent for throughput, and the failure mode of independence is that nobody is reading across them. Synthesis is expensive precisely because its only output is the defects it catches.

The decision was to treat that pass as adversarial rather than editorial. Its product is a list of what does not hold, and the worker running it has to be allowed to report a defect in someone else's result without becoming the owner of the repair. Making the finder responsible for the fix would guarantee that fewer things get found.

Cost: a full worker-cycle spent reading, with a real chance of reporting nothing, plus a coordination rule that must survive the friction of one lane grading another. What it proves is that review here is a role with a name attached and a slot on the board, not a virtue that good workers are assumed to have.