Grades that can’t be bought

Who pays, and for what

Reading Proofground is free. Buyers never pay for grades, and we take no affiliate commission from the vendors we grade — a ranking funded by referral fees isn’t a ranking.

Money reaches us from two directions, and neither one reaches a grade. Buyers pay to run their own private test calls against their own shortlist; those results belong to the buyer, stay private, and never become a public grade. Vendors can buy the four things listed below. Paying changes when and how often we test — never how you are scored.

Everything a vendor can buy — the complete list

• Disputes — free, always. Not a product; the first item on this list because it is the one that matters. Any verified vendor can challenge any result at no cost, on the free provider plan, forever. How disputes work.
• Off-schedule retesting. A fresh assessment sooner than our own calendar would reach you, against the demo line you publish. What it changes: when the calls happen. What it does not change: which scenarios run, the rubric, the judges, the failure panel, the second opinion, or whether the result publishes. It publishes whatever it finds — including a worse grade than the one it replaces.
• Recurring monitoring. The same assessment on a repeating schedule, so a prompt change that breaks rescheduling shows up in your results rather than in your customers’ calls. It buys frequency. It buys nothing else.
• The remediation brief. A private written fix-list built from an assessment of yours that has already published — which capability failed, on what evidence, and what an engineer would look at first. Payment gates the packaged, downloadable brief and nothing upstream of it: what the brief finds, how the calls were judged, and which issues qualify are all decided before money is involved, and the brief is private — it never appears on your public page and never changes it.

“You take my money and you grade me. Why would I believe you?”

It is the right question, and the answer is not a promise — it is what the system can and cannot do:

• The judges cannot favour you, because they do not know who you are. Calls are scored blind. There is no field carrying a vendor name, let alone a billing status, into a scoring decision.
• Paying does not change a single input. Same scenario suite, same rubric, same pass thresholds, same failure panel of three reviewers at three AI labs, same second opinion from a judge at a different AI lab, and the same rule that a disagreement goes to a person before anything can publish.
• A retest you paid for publishes whatever it finds. There is no approval step, no embargo, and no way to buy the old grade back. Vendors have paid for a retest and ended up with a worse letter. That is the deal, and it is the point.
• Disputes are free precisely so money is never the route to a correction. If getting an error fixed cost money, then money would buy outcomes. It doesn’t, so it can’t. The correction path is evidence, and it is open to every vendor on the free plan.
• One line, so the numbers stay comparable. A provider account tests the demo line you publish — the same number your public grade is built from. A different number is a buyer account, and buyer results are private and never public grades.

The structural safeguards

• Blind judges. The people and models scoring calls never see vendor names.
• Published criteria. The scoring criteria are published; a grade can be checked against them by anyone.
• Evidence behind every score. Every score is backed by a recorded call, and disputes are resolved against that recording rather than against anyone’s memory. Verified vendors read the transcripts and per-check reasoning behind their own results.
• Dated assessments. Grades are snapshots with dates, never quietly edited. Corrections are badged as corrections, beside the number they changed.
• Free disputes. Any verified vendor can challenge any score at no cost. Resolutions — either way — are recorded.