9 indexes · 9 answering · 508,124 artefacts under measurement · published 2026-08-14
Who publishes this
A masthead over nine indexes that already ran, so they can be held to one method instead of nine.
What Toolproof is
Nine separate indexes measure different corners of AI agent tooling — skills and plugins, tool maintenance, nightly vendor changes, starter kits, app builders, component registries, agent config files, answer-engine citations, and document-extraction accuracy. Each has its own name, its own site and its own engine, and each was built because the existing catalogue of that corner counted things without checking them.
What none of them had alone was a stated method. A figure with no published method behind it can only be believed or disbelieved; it cannot be argued with, which means it cannot be corrected either. Toolproof is that method, written once, with the nine held to it and every figure on this site pulled live from the index that produced it.
It is deliberately not a tenth index. Nothing here measures anything. If a number on this site is wrong, it is wrong at the index that produced it, and the source URL beside every figure is where to go and check.
What we have a stake in
The nine are built and operated by Kynth Studios, an independent studio that builds software across a lot of unrelated domains. They are therefore not independent of each other, and saying otherwise would be the first dishonest sentence on this site. What they are independent of is the tooling they measure: no vendor in any of these indexes can pay for a position, a score or a verdict.
One conflict is real and worth stating loudly rather than in a footnote: one of the nine benchmarks a Kynth Studios product against commercial competitors, and it wins several of those comparisons. Every raw vendor response in that benchmark is committed to a public MIT-licensed repository so the result can be recomputed offline with no API keys — which is a stronger guarantee than any other figure on this site carries, and it exists precisely because that one needed it. Every disclosure is on the methodology page.
The same method is run against Kynth Studios’ own sites. Today it put 23 buyer questions to five answer engines, 92 of those 115 came back with an answer, and this estate was cited in none of them. That is a bad number about us, published on the same terms as any other.
Who wrote it
Toolproof is edited and operated by Kynth Studios. Every index runs on a schedule and recomputes its own verdicts, so no figure on this site was chosen by a person — but the method was, the thresholds were, and the decision about what counts as a failure was. Those are editorial judgements and they belong to a publisher rather than to a pipeline. They are written out in full on the methodology page so they can be disagreed with.
Corrections, method disputes and figures you can show to be wrong go to hello@kynth.studio.
Corrections
A figure that can be shown to be wrong is corrected and the correction is recorded with its date on the methodology page, whichever direction it moves the number. A change to how something is measured gets a new method version rather than a silent rewrite, because a figure cited last month was produced under last month’s rules.
No correction has been issued yet. This masthead was published on 2026-08-14; the individual indexes are older and carry their own histories on their own sites.
