Buyer-sourced sensitivity analysis on every scorecard

From 2026-Q2 every supplier scorecard reports two purity distributions: all records, and the buyer-sourced subset alone. The change is small in presentation and consequential for a handful of suppliers.

Where it matters

For suppliers whose supplier-supplied fraction is under about 15 %, the two views are nearly identical and the distinction is bookkeeping. Above 25 % it is not: at that share, an effect of +0.31 pp on a quarter of the records moves the aggregate median measurably, and the buyer-sourced subset is both more pessimistic and more reliable.

Three suppliers in the index exceed 25 %. Each now carries a note saying so on its scorecard, positioned above the aggregate rather than below it, because a caveat that appears after the number has already been read is not a caveat.

The wider point about selection

The provenance effect is a measurable instance of a general problem the index cannot measure: who chooses to submit, and why. Bad results get submitted when someone is annoyed. Good results get submitted by the small number of people who submit everything. Neither is a random sample of the material on the market, and no amount of statistical care fixes a sample that was never random.

The aggregate caveat that appears on every rollup page is not boilerplate. It is the most important sentence on those pages: read every figure here as a description of this dataset rather than as an estimate of a population.

Coverage

Coverage remains highly uneven and the index is explicit about it rather than smoothing it over. Three compounds carry a large share of all records. Several suppliers have fewer than fifty. One has fewer than thirty, and its page says in the first paragraph that no median on it should be read as an estimate.