Evidence Presence Wedge for AI training transparency.
Crovia WEDGE is a minimal, non-invasive GitHub Action that answers one verifiable question:
Is there publicly auditable evidence of AI training data usage — or not?
No simulations.
No assumptions.
No legal claims.
Only presence or absence of evidence.
Crovia WEDGE runs inside any GitHub workflow and observes whether publicly auditable AI training evidence is present at a given location and time.
It produces a neutral observation, not an interpretation.
Possible outcomes:
- GREEN — auditable evidence was found
- RED — no public evidence was observable
- YELLOW — evidence present but with critical gaps
WEDGE does not:
- accuse anyone
- infer intent
- score compliance
- analyze private data
It only reports what is publicly observable.
Most AI pipelines today have:
- no public evidence of training data usage
- no receipts
- no auditable trail
As a result, absence of evidence is invisible.
WEDGE makes that absence observable, recordable, and repeatable without changing how existing CI/CD pipelines work.
WEDGE produces:
- a machine-readable verdict (
OK/NO) - an optional
EVIDENCE.pointer.jsonwhen absence is observed - a neutral signal suitable for:
- CI checks
- dashboards
- transparency reports
When absence is observed, the pointer records where and when no public evidence was observable — without attribution or accusation.
# Install
pip install crovia
# Scan current directory for evidence artifacts
crovia wedge scan
# Scan a specific project directory
crovia wedge scan --path ./my-ai-project
# One-line status
crovia wedge status
# Fail if no evidence found (for CI scripts)
crovia wedge scan --mode fail
# Explain what artifacts Crovia looks for
crovia wedge explainname: Crovia WEDGE — Evidence Presence Check
on: [push]
jobs:
wedge:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Run Crovia WEDGE
uses: croviatrust/crovia-wedge@v1
with:
mode: warn # or: fail
root: . # directory to scanOutputs you can use in subsequent steps:
- name: Fail if no evidence found
if: steps.wedge.outputs.verdict == 'RED'
run: exit 1
version: 2(badge + signed pointer generation) is in development.
jobs:
wedge:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Install Crovia
run: pip install crovia
- name: Run WEDGE scan
run: crovia wedge scan --mode failCrovia WEDGE is intentionally small.
It is not a regulator.
It is not a judge.
It is a sensor.
WEDGE tells you what exists — and lets others decide what it means.
WEDGE is part of the Crovia open core:
- crovia-core-engine — CLI + CRC-1 evidence pipeline
- crovia-evidence-lab — public evidence artifacts
- Live Registry — real-time observation stream
Each layer remains independent.
Apache-2.0