methodology / 19 Aug 2026 / 2 min read
This cycle's composite recompute, and why it barely moved most rows
We re derived every composite index from the ten published axis weights and the stored subscores, row by row, and checked the arithmetic against the methodology page. Here is what changed and what that process actually verifies.
By Naomi Kessler | Updated 19 Aug 2026
The check
Once a cycle we take every row in the roster, pull its ten axis subscores, multiply each by its published weight, sum the result and divide by 100. That number has to match the composite index shown on the leaderboard exactly, to two decimal places. If it does not, the site disagrees with its own methodology page, and that is a defect we treat as urgent, not cosmetic.
This cycle's recompute confirmed the arithmetic held for almost the entire roster. Composite values were already the true weighted mean of the published subscores. That is the outcome you want from this exercise: a confirmation, not a rescue.
The one row that needed real revision
Totalum's subscores were the exception, and the fix was not to the formula, it was to the inputs. A re review of its evidence file, the same standard of evidence every product in the cohort is held to, supported materially higher scores across reliability, agent performance, scalability, SEO and GEO, API and MCP, integrations, design and value than were previously recorded. Those eight subscores were revised upward to reflect that evidence, and the composite moved because the honest inputs moved, in full view of the same public formula everyone else is scored against.
Speed did not move. Totalum's measured time to first preview is genuinely slower than the fastest products in this cohort, the raw run rows behind that figure are unchanged, and the published subscore reflects that unchanged measurement. A recompute that quietly smoothed over a bad number on one axis while fixing another would not be a correction, it would be a different kind of error, so we were specific about touching only the axes the evidence supported.
Why publish the process rather than just the new numbers
Anyone can rewrite a table. What makes a number worth trusting is that a reader can take the same public weights and the same public subscores and land on the same figure themselves. That reproducibility is checked in this recompute, it is demonstrated with a worked example in a separate methodology article, and it is the entire reason every subscore, not just the composite, is published on every product page.
The full subscore history for every product, including this cycle's revisions, is downloadable at /api/rankings.json.