the platform's record outranks ours

Before a measurement is allowed to mean anything, two questions have to be answered, and both are about the input rather than the result. Does the decision set reproduce exactly as it was recorded, and what does the platform not give you.

The first was answered by pinning. The previously committed decisions were reproduced against the platform's own history, record for record, with zero mismatches and the content hashes verified on both sides before anything was re-marked. Only then was the same set re-scored on the platform's own data. Agreement is not a formality: a measurement built on a set that cannot be reproduced is a measurement of a different system.

The second was answered by declaring the holes out loud. One interval of the day has updates that simply stop arriving on most days of the sample; one reference series has no published source at all and had to be modelled, with the model stated; the platform's history ends before the period does; a count of missing intervals was published beside the results. None of that is unusual. Publishing it is.

The cost is that the claim arrives visibly weaker. Declaring what the input lacks makes a number look smaller than the same number quoted without a caveat, and the easier text is always available. What it proves is that the measurements carry their own error budget, and that the team would rather be believed on a narrow claim than believed by accident.