← back to stream
How to Read Crovia
What is an observation?
An observation is a timestamped record of:
- Presence: Evidence of training data was found at a public location
- Absence: No evidence was found after systematic check
Observations are stored in a PostgreSQL database and exposed via public API. No authentication required.
Why time matters
The duration of an absence is evidence itself.
A model with 1,891 days without observed training evidence indicates persistent non-disclosure, not a temporary gap.
Time accumulates. Silence becomes measurable.
What "verify yourself" means
Every observation has a SHA-256 hash. You can verify it independently via:
- Public API:
curl https://registry.croviatrust.com/api/registry/recent
- PostgreSQL dump (if available)
- HuggingFace dataset snapshot
- CEP capsule verification (hashchain + signature)
No login required. No trust required. No permission required.
What Crovia does NOT do
- Does NOT audit models
- Does NOT infer intent
- Does NOT make legal claims
- Does NOT enforce compliance
- Does NOT judge
Crovia observes. Others decide.
Shadow Score
Shadow Score (0-100) measures completeness of trust declarations based on the NEC# Canon (Necessity Canon).
It is calculated deterministically from publicly observable metadata:
- Data provenance (NEC#1)
- License attribution (NEC#2)
- Usage scope (NEC#7)
- Temporal validity (NEC#10)
- Accountable entity (NEC#13)
- + 15 other necessities
Formula: shadow_score = 100 - Σ(violation_severity × 0.15)
Verifiable. Reproducible. Public.
Data sources
Observations come from:
- HuggingFace model cards
- Public dataset repositories
- Training evidence declarations
- CEP (Crovia Evidence Protocol) capsules
All sources are public. All checks are reproducible.
Limits
- Observations are not exhaustive
- Absence ≠ proof of wrongdoing
- Presence ≠ proof of compliance
- Crovia records facts, not judgments
This is an observation protocol, not an audit system.
This is a public registry, not a compliance enforcer.
This is evidence, not judgment.