AgentMeasure_ Full-period review →

Paid diagnosis

Everything on this site runs free and local — the check-up, the conformance pack, the ledger tools upload nothing. These two services are for the moment you want a human answer about your own data: fixed scope, fixed price, and a stated evidence boundary. No subscription, no sales call.

Where the line is: the free check answers is something off in my logs? — locally, no account, no email, and it stays free forever. It cannot tell you what the finding means in your setup, whether it's a bug or intended semantics, and what to do next — that is a human reading your sanitized summary with the public 110-tool audit corpus as reference, and that is what the $10 buys. You never need the paid part to use the free part.

Healthcheck Diagnosis — $10, one anomaly

You ran the free check, it flagged something, and you want to know what it is

$10 · one anomaly, one machine · human-delivered

  • Who it's for: Codex / Claude Code developers who see duplicate records, retry amplification, failed runs or a token number that looks wrong, and don't know the cause or the next step.
  • You send — one email is the whole application: run the three free commands below first (~1 minute, nothing leaves your machine), then hit Apply and attach summary.json. That single message carries your task and your data — no account, no forms, no follow-up forms.
  • You get: one reply confirming scope and price ($10 fixed) with the payment step — then, within 24 hours of payment, a written diagnosis on your own data: what the anomaly is, what the evidence does and does not support, fix and triage steps, and how to re-check after fixing. One round of async clarification included.
  • Turnaround: within 24 hours of payment; if we're queued, we say so in the confirmation reply — before you pay.
  • Honesty clause: "no finding" or "not enough evidence" is a valid result and will be delivered as such — we don't manufacture a bug to justify the fee.

Run these first (free) — then click Apply and attach the file:

pip install agentmeasure
agentmeasure check --json report.json
agentmeasure share report.json --out summary.json  # previews first; export only after you look
Apply by email — $10

Card checkout link on request while our payment page is being set up.

Metering QA check — $20, one sample

You meter usage yourself, and the numbers have to be right

$20 · one sample, one ruleset, one metering question · human-delivered

  • Who it's for: authors of MCP servers, APIs and agents that bill or report by usage — especially around a release, a price change, or new retry logic.
  • You send: one event sample in JSONL (up to 10k events) plus the number you currently report. Not sure it fits? Send the schema and first ~20 lines first — fit is confirmed before anything is charged; if we can't adapt it, we tell you and you owe nothing.
  • You get: a discrepancy or no-finding report where every verdict is bound to the invariant that produced it (PASS / FAIL / UNPROVABLE — same vocabulary as the public claims index), a check you can re-run yourself, and a concrete reproduction when there is a real error.
  • Scope: one sample, one ruleset, one metering question. Bigger or multi-system questions are quoted separately.
  • Turnaround: within 24 hours of receiving complete input.
Apply by email — $20

Card checkout link on request while our payment page is being set up.

How it works

1 · Run + ApplyRun the free commands, click Apply, attach summary.json — one email carries task and data. ($20: send schema + first ~20 lines in that same email; unfit samples are refused free of charge.)
2 · Confirm & payOne reply confirms scope, price and turnaround; card link when available, otherwise agreed in the same thread. We start on payment — never before.
3 · Report in 24hFindings, evidence boundary, fix steps, and how to re-check — one thread, one clarification round included.
Small operation, deliberately: we cap daily orders so turnaround stays honest. If today's cap is reached, we say so and you can come back tomorrow — we won't take money we can't deliver on time.

After you apply

Everything runs over email — tongroy18@gmail.com. No accounts, no forms, no portal. Your application, your sample and your report live in one thread you can keep.

Data boundary

The free tools run locally and upload nothing — that stays true forever. For a paid diagnosis you choose what leaves your machine: a sanitized export of your own picking, and nothing else. We look at what you attached, we don't re-collect it, and input is used for your delivery only.

Terms, before you pay

  • Full refund any time before delivery starts (we haven't opened your input).
  • Unadaptable sample: refused before payment if you checked fit first; refunded in full if it turned out unadaptable after payment.
  • If we can't deliver within 72 hours of complete input, full refund — yours to keep the report anyway if it arrives.
  • "No bug found" or "not enough evidence" is a delivered result, not grounds for refund by itself — that possibility is priced in, and it's why the evidence boundary is stated in every report.
  • $500+ deep dives (multi-system, custom field mappings) exist on request — they are never required for the two services above.

Free stays free

The local check-up, the browser recount, the claims index and the conformance pack remain free and open source. Paid here buys a human-delivered diagnosis on your data — not a paywall around anything that was free yesterday.