every lane lands as its own merge

With many workers writing at once, integration is where attribution dies. The cheap move is a batch merge: collect everything, land it once, write one message. It is also the move that makes the history unusable as an audit trail, because afterwards nobody can say which lane produced which change, and the question cannot be answered later by anyone, at any price.

The protocol chosen was the expensive one. Each lane syncs on its own cadence and lands as its own merge, repeatedly, across hundreds of integrations. Read as a sequence, the history alternates between per-lane syncs and per-lane merges with little else in between, which is what a working attribution protocol looks like from the outside.

The tension is between a shared trunk and accountable units. Batching is faster, quieter and produces far fewer integration events; per-lane integration multiplies them and yields a history in which the interesting lines are a small minority of the total. Choosing the second is choosing noise on purpose, because the noise is the evidence.

The payoff is narrow and worth it. Any change can be attributed to a lane without asking anyone, a lane's contribution can be reconstructed from the history alone, and a bad integration can be identified by unit rather than by batch. Those are the properties an auditor would ask for, and none of them survive a batch merge.

Cost: integration volume, near-simultaneous landings, and a log noisy enough that summarising it became a scheduled job in its own right. What it proves is that the operation chose auditability over a tidy record, and pays the price every single cycle rather than once at the end.