What to walk away from

The red flags in a rating tool are mostly visible on its results page: false precision, no rubric, everyone scoring high, a paywalled breakdown, no named owner, no changelog, no rejections, no public results, and marketing in place of method. Each is checkable in seconds and covered in more depth elsewhere on this site.

1. False precision with no visible method. A score to two decimal places - 7.43 rather than 7.4 - implies a stability the underlying process almost never has. Most of that extra digit is rounding, not signal, and a tool leaning on it to look rigorous is dressing up noise from a model that was never that stable run to run in the first place.

2. No stated rubric. A single total with no named components and no page describing what the tool claims to judge. What a rubric is supposed to be is the list of things a tool scores and how much each counts - without it, you are trusting a combination you cannot see.

3. Everyone scores 8 or above. A tool whose visible results cluster at the generous end of its own scale has either self-selected for shared results or is not distinguishing between submissions at all. Generosity bias keeps users happy and keeps the scale from meaning much.

4. A locked breakdown behind a paywall. Showing a headline number for free and hiding every axis behind payment is a specific dark pattern - the breakdown, not the total, is where the useful information lives, and locking exactly that piece is not a coincidence.

5. No owner named anywhere on the site. An anonymous tool has removed your ability to weigh its incentive against its number. Absence alone is not proof of bad faith, but it is a cost, and it belongs on this list for that reason.

6. No changelog for rubric or scale changes. Rubrics get edited and mappings get retuned silently on most tools. Without a dated record, a score from a year ago and one from today are on scales that may no longer match, and the tool has given you no way to know.

7. Zero rejections, ever. A tool that scores every image it is handed, including the unusable ones, is not applying its rubric so much as guessing and presenting the guess with a confident number. A working rubric has a floor, and a floor means some submissions fail it.

8. No public results beyond your own. With nothing to compare against, every claim about spread, calibration or consistency is unverifiable from outside. Public entries are the one feature that actually lets a reader audit a tool, and their absence closes off that check entirely.

9. Marketing language standing in for a rubric. "Advanced AI", "cutting-edge accuracy", "revolutionary technology" on the results page itself, rather than a plain description of what got scored. The label tells you almost nothing about what makes the tool different from the next one that says the same thing - it is filler where a rubric should be. US regulators treat that language as a claim rather than decoration: announcing Operation AI Comply in September 2024, the FTC said "there is no AI exemption from the laws on the books".

Reading the pattern, not any one flag alone

None of these nine is individually disqualifying. A young, honestly built tool might lack a changelog simply because it has not been running long enough to need one, and that is a different situation from a tool actively hiding something.

What is worth noticing is how many apply at once. A tool showing three or four of these together - no rubric, no owner, everyone scoring high, a locked breakdown - has stopped looking like an oversight and started looking like a design. The positive counterpart to this list is worth reading alongside it, since a tool clearing most of the green flags and none of these red ones is the clearest case either way.

Where this sits in the wider method

These nine are a fast pattern-match, not a substitute for the fuller ten-minute evaluation, which checks rubric, calibration, consistency, presentation and disclosure directly rather than by proxy. Run the full method on a tool you are about to trust with something that matters; use this list as the quick version when you just want to know whether a tool is worth the ten minutes at all.

None of these nine apply cleanly to a human reviewer, who is a different kind of judgement entirely - etiquette and trust with a person work on different signals than a software results page does. And none apply to a physical measurement, where the equivalent red flag is a number with no stated method behind it rather than any of the nine above - a different category, with its own version of the same underlying question: does the number have a process behind it you could actually check. Rate Cock is one tool that clears most of these by construction - named ownership, a stated six-axis rubric, public entries with visible breakdowns - cited here as a checkable instance, not a verdict on the category.