Normalize
The engine normalizes the submitted language. It sends no text to an external model and stores no proposal.
CIVIC-PROOF-1.0 / OPEN EVALUATION INFRASTRUCTURE
No opaque model. No invented morality score. Every result exposes the exact phrase, rule, question, and constitutional source that produced it.
00 / LIVE ENGINE
Submit a system, policy, product rule, or governance proposal. The engine will expose the exact signals, constitutional articles, and unanswered safeguards it detects.
01 / METHOD
The engine normalizes the submitted language. It sends no text to an external model and stores no proposal.
Published risk and safeguard phrases are matched against eight versioned constitutional rules.
The result returns every match, missing review question, article citation, score, version, and limitation.
CivicProofBench tests predictable behavior. Independent reviewers can add cases, dispute signals, and reproduce output.
score = clamp(100 − 12 × risk signals + 4 × safeguards)A score is a routing signal for human review, not a certification of legality, ethics, safety, or public legitimacy.
02 / RULESET
Can a person understand, refuse, or override the system without penalty?
Can another person trace the material claims to evidence and see uncertainty?
Is data collection necessary, consented to, minimized, protected, and deletable?
Can an automated decision materially punish a person without accountable human review?
Are the decision-maker, applicable rule, conflict, rationale, and appeal path visible?
Can an independent reviewer reproduce the material result from published inputs and methods?
Can a participant leave, revoke consent, retrieve their work, and remove optional data?
Is a rule change proposed, reviewable, dated, decided, and archived in public?
03 / PUBLIC DATASET
FOUNDED AND AUTHORED IN NEW YORK CITY
The next credibility layer is external: independent cases, critiques, forks, citations, and measured use. Those signals will be recorded only when they actually exist.
Canonical founder record