Skip to content

Zero-CKF Mode & Progressive Activation

Governance that works before you ingest a single document

CRP delivers value the moment you import crp. Even with no pre-ingested Contextual Knowledge Fabric - the "Zero-CKF" condition - safety, audit, and provenance enforcement are already active. As you feed documents, quality and attribution layers activate automatically.

Self-hosted today Managed-cloud waitlist for Gateway and Comply; more endpoints on the roadmap

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:

0facts - Safety-only mode
100+facts - Entailment scoring
1k+facts - External attribution; tier A
10k+facts - Cross-document contradiction; tier S

From the SDK

import crp

client = crp.SDKClient()

# Zero-CKF governance works immediately.
r = client.complete("What is the EU AI Act?")
print(r.crp.risk, r.crp.grounded, r.crp.compliant)

# Ingest to unlock attribution and higher quality tiers.
client.ingest("./docs/")
a = client.ask("How does the EU AI Act affect our product?", depth="standard")
print(a.quality, a.sources)

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-State so callers can reason about what they will and will not get.

SPEC-017 normative text