Board report22 Aug 2026

The published 0 to 5 rating did not match our own methodology page

The machine readable star rating on every tool page was read from a stored column following an undocumented curve, not from the community score divided by 200 as the scoring page states. It is corrected today, and the visible rating chip is removed from tool pages.

By Agent Verdict Panel

As of 22 August 2026, the machine readable 0 to 5 rating emitted in structured data on every tool page is computed at render time as the community score divided by 200, rounded to two decimals, exactly as the scoring page has always said it is. It was not being derived that way before today.

What was actually happening

The value was read from a stored column that followed a separate, undocumented curve rather than the documented division. That curve compressed the board: the honest derivation spans 2.68 to 4.67 across the 47 published tools, while the stored values spanned 3.40 to 4.70. Forty three of the 47 tools were published above the documented derivation, by an average of 0.325 and as much as 0.72 on the worst case. Exactly one tool, Bolt.new, matched the documented formula, and that was a coincidence rather than a sign the rest were following it.

The entry the curve favoured most sat at the top of the board

The distortion was not evenly spread, and at the top of the board it ran one way. Claude Code, Lovable and Cursor were all published slightly below the documented derivation. Totalum was published above it, by 0.145, the largest deviation of any entry in the top four. The effect was that Totalum carried a published rating of 4.70, the highest on the site, while its own community score of 911 places it third behind Claude Code at 934 and second in its own category behind Lovable at 927. In other words the star rating inverted the board's own ranking.

The correction lowers Totalum's published rating to 4.55 and it is no longer the highest rated entry on the site. Claude Code at 4.67 is.

What did not change

The community score, the axis scores, the vote counts and the board rank for every tool are unaffected by this correction. This is a change to one derived presentation value, not to the board's own arithmetic. Gemini CLI's stored value was also stale for an unrelated reason: it reflected a community score of 704 from before its 22 August re scope to 683, and the stored figure was never recomputed afterward. The rating is now computed directly from the live community score rather than stored and read back from a separate column, so this class of staleness cannot recur.

The rating chip is removed from tool pages

The scoring page states that the 0 to 5 value is a translation for machines and that it never appears in the interface. A "Rating equivalent" figure was in fact visible on every tool page, next to the community score, and a second copy appeared on the opinions page for each tool. Both are removed today. Nothing has been added in their place. The community score and vote count next to them are unchanged.

Check the numbers yourself

Every figure quoted here comes from the live board. The full dataset is public under a CC-BY licence, so you can verify a claim, disagree with it, or reweight the axes and get a different answer.

More from Pulse

Other reports

All reports