How we score — and why money never touches a score or a rank
Every verdict we publish is ranked by an independent 9-dimension rubric, not by who pays us. That's an easy thing to claim. This page is how you check it.
Scored independently by David Kooi, Skylark Creations.
The firewall: money can't touch the score or the rank
A vendor's payment never reaches the score or the rank. The agenticness evaluator and the search ranking never read whether a listing pays us — and we don't rely on good intentions to keep it that way: a test fails CI if one of a named set of monetization fields (an affiliate link, a commission, a sponsored flag) is ever added to the scoring or search-ranking inputs. On a category page, a paid Sponsored listing gets its own labeled card, and the ranked list stays exactly as the scores order it. Affiliate links, where they exist, attach only at the final click-through — after the ranking is decided — so they cannot reorder anything.
The rubric: 9 dimensions, out of 36
We judge — G2 crowdsources. Every tool is scored on the same public Agenticness rubric across nine dimensions of real autonomy, for a total out of 36. Capability, pricing, and licensing are pulled from structured data; adoption is real outbound-click activity. Read the full rubric and every dimension →
Reliability: evidence, not marketing
Reliability is graded conservatively on two sub-signals — a harness-linked benchmark and real-world incident/workflow history — and only against evidence we can trace to a primary source. We anchor on contamination-resistant benchmarks and documented incidents, never a vendor's own marketing. A “not yet evidenced” is a sourcing stance, never a verdict that a tool is unreliable.
Questions about our independence
- Do vendors pay to rank higher?
- No. A vendor's payment never touches the score or the rank. The agenticness evaluator and the search ranking never read whether a listing pays us — and a test fails CI if one of a named set of monetization fields (an affiliate link, a commission, a sponsored flag) is ever added to the scoring or search-ranking inputs. On a category page, a paid Sponsored listing gets its own labeled card, and the ranked list stays exactly as the scores order it.
- Do you earn money from these tools at all?
- Sometimes, via affiliate links in a few categories. But an affiliate link only attaches at the final click-through redirect — after the ranking is already decided — so it cannot reorder anything. Analytics stay keyed to the tool's real URL. Coding-agent tools, our most-read category, pay us nothing.
- What about the "Sponsored" card on category pages?
- A category can carry one self-serve Sponsored card, clearly labeled. Sponsoring leaves the ranked list below exactly as the scores order it, so no tool's position changes. Payment buys the card, never a better score or a higher rank. A category's Featured slot and its Sponsored slot are separate — one doesn't bump the other.
- What if the best tool for me is one you can't profit from?
- Then the report says so. Recommending the genuinely best option — even when that's not a paying tool, and even when it's not us — is the whole asset. The moment we put a thumb on the scale, the verdict is worthless.
- How do you score a tool?
- Every tool is judged on a public 9-dimension Agenticness rubric, scored out of 36. Capability, pricing, and licensing come from structured data; reliability is graded only against evidence we can trace to a primary source; adoption is real outbound-click activity from the directory.
Our decision reports are reviewed against this public rubric before they're published. If you ever spot a verdict that looks like it favors a paying tool over a better one, tell us — that's the one bug we treat as critical.