What they told you is not a measurement

Diligence on a receivables portfolio starts with numbers the seller provides. Historical collection rate, average days to settlement, typical discount achieved. They arrive in a spreadsheet and they get typed into a model, and at that moment they stop being someone’s claim and start looking like data.

They are not data. They are the seller’s estimate of the exact quantity whose value is under negotiation, produced by the party with an interest in it. That does not make them false. It makes them a claim with an author, and the author is material.

So EliAI stores every claimed figure as claimed, with who said it and when, and never as a measurement. Beside it sits the figure derived from the counterparty’s own export — and the derivation names its own choices, because the same underlying book supports very different-looking ratios depending on maturity treatment and denominator. A lifetime ratio over a growing book understates recovery and rewards decline; that is arithmetic, not opinion, and a model that does not say which convention it used has not told you anything.

The case that matters most is the one with no answer. Where the coverage is too thin to rebut a claimed figure, the difference is reported as absent rather than as zero. Zero is a number, and a number in that slot reads as a finding.

This is the same discipline as refusing to price, applied one layer up. Both come from the same rule: the system may not present the absence of evidence in the same format as evidence.