A correction, published in full

What the defect was

Catalog size earns up to 25 of the 125 point maximum on a log scale: the first services count for a lot, the hundred-thousandth for almost nothing. The published rule stops the input at 9,000 services, the same way longevity stops at 10 years and breadth stops at 8 platforms. Two of those three caps were applied in code. The catalog one was not: a catalog above 9,000 services kept earning points past its own maximum, so the directory printed components like "Catalog 33.25 / 25" — a value larger than the weight it belongs to — and the advertised maximum of the score stopped being a maximum at all.

It was found while rebuilding the public component breakdown (the bars on comparison pages and the stat strip on panel pages), which is what made it visible: the bars render the real numbers now, and the real numbers were impossible.

What changed on 2026-09-13

The catalog input is now clamped at 9,000 services, exactly as published. Every panel listing that many services or more earns the full 25 catalog points and no more. 6 panels in the directory are in that range today; at the correction five of them carried a published score, and those are the five whose scores moved:

The sixth panel in that range, NSBoost BD (9,679 services), carries no published score because its domain does not serve a website — the directory stopped scoring dead domains earlier this year rather than ranking them.

What did not change

• Weights and inputs. Longevity, catalog, performance, breadth and verification keep the same weights, the same sources and the same caps. No panel was added or removed. Nothing on this site became cheaper to buy.
• Every other score. Of the 129 panels in the directory, 124 have a byte-identical Scout Score after the correction. Those five scores above are the entire change.
• The order below the top. The two largest catalogs left the top of the ranking because they were never entitled to the points that put them there; the panels between them moved up by a place or two. The leader is unchanged and is not one of the five.

Why this is published like bad news, because it is

Two of the five panels are not ours; none of the five is. The correction moved rankings away from the biggest catalogs, which is a change no directory wants to explain, and it also removed the widest gap we had between a published rule and the code. The alternative — quietly re-wording the methodology to say the cap was never a cap — would have left the 25-point weight overstated on the largest panels and the score maximum untrue. We took the ranking hit instead, and the methodology page now carries the correction note plus the clamped formula.

The same clamp is in the Worker that serves /api/panels.json and the breakdown endpoints, so the open data API and the pages agree. Any score on this site is still recomputable from the published formula and the panel's own inputs.

What to watch next month

• Catalog re-measurements on the five panels above: the clamp means their catalog component can no longer move, so their remaining score movements come from latency and longevity.
• Whether any panel near 9,000 services crosses it — crossing it is now worth nothing, which is the point of a cap.
• The rest of the ranking, which is unchanged in inputs and open to inspection per panel.