Zero-CKF Mode & Progressive Activation¶
CRP™ works without any pre-ingested Contextual Knowledge Fabric. When the CKF is empty (the "Zero-CKF" condition), the protocol operates in safety-only mode: the DPE still runs, Safety Policy still enforces, audit chain still binds — but quality tiering accepts a reduced floor and some attribution checks degrade gracefully. Fully specified in SPEC-017.
Why This Matters¶
A common objection — "CRP needs domain documents to be useful" — is wrong. With no documents the protocol still delivers:
- Hallucination risk scoring (intrinsic claims only)
- Distortion detection
- Cross-window contradiction detection
- Repetition detection
- Completeness verification
- Flow analysis
- Full Safety Policy enforcement
- Full HMAC audit chain
- Full regulatory tagging
What it cannot deliver in Zero-CKF mode is external attribution — binding a claim to a specific source document — because there are no source documents.
Progressive Activation¶
As you ingest documents, additional DPE capabilities activate automatically:
| CKF state | Capabilities enabled |
|---|---|
| Empty | Safety-only mode (above) |
| > 100 facts | + Entailment scoring against the CKF |
| > 1,000 facts | + External attribution; quality tier A reachable |
| > 10,000 facts | + Cross-document contradiction; tier S possible |
Implications¶
- You can deploy CRP today with zero domain documents and still get governance.
- The protocol degrades gracefully, never silently.
- The site, the spec, and the CRP Gateway all advertise the current mode in
CRP-CKF-Stateso callers can reason about what they will and will not get.