EVERY PROOF RUNS THE GAUNTLET

Valid is the floor.
Strength is the score.

A registry tells you a proof exists and has not been changed. It does not tell you how much the proof can carry. ProofGauntlet checks that a proof is real, then rates its strength from 0 to 100 under a published rubric anyone can recompute.

GET EARLY ACCESS READ PP-SPEC-100

Spec published. API in build.

gauntlet-verdict.jsonEXAMPLE

found at: proofregister.com / PR-2026-NNNNN
gate: valid (G1 to G9 pass)
tier: efficacy, claims tested
depth: 8 techniques
independence: reproven once
age: 6 days
rubric: PGR-0.1.0

88STRONG

Two stages. No shortcuts.

STAGE 01 / THE GATE

Is it real?

Nine pass-or-fail checks. Fail one and there is no score, no band, no rank. A check that cannot be run is unverifiable, never a pass.

Registered, with a receipt that verifies, and not revoked
Unchanged, and anchored to a public time beacon
Signed by keys that come from somewhere other than the record
Expected outcome committed before the attack ran
A valid, conforming run

STAGE 02 / THE GAUNTLET

How much can it carry?

Valid proofs are scored 0 to 100 across six published dimensions. Anything the record cannot show scores zero. Thin records score low. Honestly.

weak 0 to 39
moderate 40 to 69
strong 70 to 89
extremely strong 90 to 100

The score is a spec, not an opinion.

No jury. No hidden weights. The rubric is a versioned data file. Same record, same rubric, same time: same answer, for a person or a machine.

DIMENSION	THE QUESTION	WEIGHT
Tier	Was the control shown to work, or only shown to exist?	25
Adversarial Depth	How many techniques, across how many tactics?	20
Environment Fidelity	How close was the test to the real environment?	15
Capture Fidelity	Was the evidence honest when it was captured?	15
Independence	Did anyone with a stake take part? Was it reproven?	15
Freshness	Is it still true today?	10

Ceilings: no proof of efficacy caps at 40. A reproof that diverged caps at 50. Rubric PGR-0.1.0.

Same gate. Very different weight.

EXAMPLES FROM THE CONFORMANCE VECTORS

VALID / STRONGemail security

88/100

TierEfficacy, claims tested

Depth8 techniques

EnvironmentCustomer-like replica

CaptureIndependent at boundary

IndependenceReproven once

Freshness6 days

VALID / WEAKendpoint

31/100

TierEfficacy, no claims

DepthSingle technique

EnvironmentDev environment

CaptureNo coverage map

IndependenceIssuer is witness

Freshness10 months

INVALID / NO SCOREweb application firewall

Registered, intact, anchoredPASS

Signed, schemaPASS

Committed before the runFAIL

The expected outcome was committed four minutes after the attack ran. The goalposts moved. The gauntlet does not score moved goalposts.

Strength is not efficacy. A strong proof can show a control failing. Efficacy and false block rates are always shown beside the score, and rankings rank proofs, not products.

API FIRST

The API is the product.

This site will be a client of it, nothing more. Reads need no key and no account. No bot walls. Agents are welcome. Nothing you send for checking is stored unless you ask. Any proof in proofregister.com can be gated and rated; no one has to submit it, and no one can opt out.

In build. Described in full by the OpenAPI 3.1 file.

curl https://proofgauntlet.com/api/v1/rating/proofregister.com/PR-2026-00001

GET/gate/{registry}/{record} GET/rating/{registry}/{record} POST/gate, /rating (stateless) GET/rankings GET/rubrics/{version}

Valid is the floor.Strength is the score.