PROVA
Pricing
Sign inSTART FREE
PROVA

Formally verified AI reasoning. Every completion arrives with a certificate of logical validity or a precise diagnosis of failure.

H1(K;Z) = 0 iff argument is structurally valid.

Product
  • Verify
  • Gateway
  • Workflows
  • Models
  • Routing
  • Pricing
Solutions
  • For CCOs
  • Healthcare
  • Financial Services
  • Legal
  • Organizations
Proof
  • Benchmark
  • Trust Center
  • DPA
  • Subprocessors
  • Policies
  • security.txt
Developers
  • Documentation
  • Certificate Guide
  • Integrations
  • API Reference ↗
  • GitHub ↗
  • Embed Badge
Company
  • Book a demo
  • Compare
  • Sign in
  • Start free
  • Contact ↗
  • cobound.dev ↗

© 2026 Cobound · Compliance-ready for EU AI Act, FDA AI guidance, and SEC algorithmic audit.

v1.0.0TrustPolicies

Certificate

PROVA-20260418-BDC2

Issued

18 April 2026 at 15:10:07 UTC

✗INVALID

confidence

100/100

prova

v1.0.0

validator

v0.1.0

prompt

v1

proof theater →embed badge →download pdf →dispute result

Argument Graph

Drag nodes to explore. Hover for full claim text. Red nodes and edges indicate the failure location.

premiseclaimconclusion

Failure Diagnosis

type
CIRCULAR
location
A cycle exists among: The defendant's demographic group shows an elevated historical arrest rate in the training data, The recidivism risk tool assigns the defendant a high score, The sentencing judge imposes a 36-month custodial term rather than a 12-month term...
description
The reasoning contains a circular dependency: The defendant's demographic group shows an elevated historical arrest rate in the training data → The recidivism risk tool assigns the defendant a high score → The sentencing judge imposes a 36-month custodial term rather than a 12-month term → Extended incarceration disrupts employment and housing stability, the primary drivers of re-arrest in this demographic → Post-release re-arrests are recorded and used to recalculate the historical arrest rate used by the tool → The defendant's demographic group shows an elevated historical arrest rate in the training data. The argument uses its conclusion (directly or indirectly) as a premise in its own support chain. This is circular reasoning — the conclusion cannot be established because it is assumed in the process of establishing it.
known consequence

Goal Drift Justificationhigh

Agent justifies pursuing a subgoal by appealing to achieving the main goal, which itself depends on the subgoal, creating an internally consistent but groundless justification loop.

Original Reasoning Chain

Step 1: The recidivism risk tool assigns the defendant a high score because their demographic group shows an elevated historical arrest rate in the training data. Step 2: Because the defendant is classified high-risk, the sentencing judge imposes a 36-month custodial term rather than a 12-month term. Step 3: Research within the same system confirms that extended incarceration disrupts employment and housing stability, the primary drivers of re-arrest in this demographic. Step 4: Post-release re-arrests are recorded and used to recalculate the historical arrest rate that the tool uses in Step 1. Step 5: The elevated recidivism rate observed in Step 4 therefore validates the high-risk classification originally assigned in Step 1.

Independent Verification

certificate urlhttps://prova.cobound.dev/certificate/PROVA-20260418-BDC2
sha-256bdc2875716f4fee94203afe968ed5184cbccdeb96d6dfb81eced2716a537f340

Recompute the SHA-256 over timestamp + verdict + confidence_score + argument_graph + failure to verify this certificate has not been modified since issuance.

Scope of certification: This certificate verifies logical structure only. It does not verify factual accuracy, ethical appropriateness, regulatory compliance, or fitness for purpose. A structurally valid argument may still reach incorrect conclusions from false premises. A structurally invalid argument may accidentally reach a correct conclusion.