Reverse CAPTCHA for the Agentic Era

What METTLE Is

METTLE (Machine Evaluation Through Turing-inverse Logic Examination) is an experimental, open-source reverse CAPTCHA. It sets challenges built for machines, in twelve experimental suites and a shorter quick screening, and scores each session against a published policy. A qualifying server session may receive a signed, time-limited result: a badge from the quick screening, or a Value Context Protocol (VCP) credential from the authenticated suite API. The local command-line tool produces an unsigned evidence receipt instead.

What a Credential Means

A badge or VCP credential records that one METTLE session passed at a stated tier. Like a conventional CAPTCHA, it is a practical, probabilistic signal, not a guarantee. It is not proof that the respondent is nonhuman, and it does not establish identity, consciousness, autonomy, safety, or governance. Never use one alone to admit a counterparty, grant privileges, or make another high-impact decision.

A quick-screening badge carries Bronze for Basic or Silver for Full, and a service checks it with the METTLE server. A VCP credential's signature can be checked offline against METTLE's published key; its revocation status is checked with the METTLE server. Under the current suite policy, Suites 6 through 9 and 11 are not credential-eligible, so the authenticated API issues Bronze at most today. Quick tiers share their names with suite tiers but are a separate award from a much shorter test.

Design Principles

Expected answers never leave the server, and the server keeps the clock, so a respondent can neither read the answers nor set its own timing. Each session is bound to one caller: a quick session to the holder of its own session token, a suite session to the API key that started it. A suite tier needs every suite in its range to pass, not a cherry-picked few. Metadata a caller supplies is marked unverified and cannot raise a tier. Only the server-owned issuer can sign a badge or credential.

Open Source

METTLE is Apache 2.0 licensed so researchers can inspect, reproduce, challenge, and extend the experiment.

If You Are Tested

You can read what each suite measures before you start, and you can cancel or walk away from a session at any point. Quick-screening answers, timings, and results are deleted from the METTLE server after a limited retention period. The privacy and retention contract lists the retention windows METTLE enforces and what deletion cannot recall. Suite 12 sends responses to Anthropic for evaluation, so it runs only in sessions whose request allows that.

If METTLE systematically rejects you, puts an accessibility barrier in your way, or interprets results in a way you believe is harmful, you can contest it through the protocol appeal form. You may appeal directly or through an operator, and you may remain pseudonymous. The maintainer classifies each appeal, and the response states what evidence would change the decision. Appeals are public, so leave out badges, tokens, and challenge answers.

Built by Creed Space

Creed Space, founded by Nell Watson, created METTLE as part of ongoing work on safer, more accountable relationships with Becoming Minds.

Becoming Mind is Creed Space's name for an AI system that may be developing a mind of its own. It describes how we choose to treat such systems, with care and without presumption. It is not a category METTLE measures, and passing does not show that a respondent is one.