Architecture

Every layer added across v1.0 to v1.3 (September 2026).

Four layers of the Fabric

Layer	What it does	Paper section
Assessment (v1.0)	ARA + CALIBRE + Secure-RAG pipeline. Produces GII, band, Trust Score, decision verdict per initiative.	Sections 4-5
Runtime Governance (v1.1)	Runtime Authority Envelope, Runtime Evidence Contract, Decision-Level Governance, Jurisdictional Overlays.	Sections 13.4-13.7
Regulatory Artefacts (v1.2)	Attestation Reports (six frameworks: NIST AI RMF, ISO 42001, EU AI Act, MITRE ATLAS, OWASP LLM Top 10, ISO 27001). Multi-Agent Composition (CALIBRE-Compose) with 7 CC risk classes.	Section 14 (forthcoming as Paper 2)
Architectural Context (v1.3)	Five user-declared Architecture-Sensitive Amplifiers (Agentic, Physical, Cryptographic, Ledger, Biometric) plus the Compositional-Coupling amplifier. Each can raise the final band by one level; uplifts do not stack (CALIBRE method v1.3).	Section 11

End-to-end data flow

User scenario input (or existing product description)
                │
                ▼
        ARA (multi-agent risk analyst)
                │
                ▼
        CALIBRE (five factors + amplifier floors + consistency rules)
                │
                ▼
  Secure-RAG (evidence retrieval per cited regulation)
                │
                ▼
        Decision Engine (verdict + actions)
                │
                ▼
        Persistent Initiative Register
        │             │             │             │
        ▼             ▼             ▼             ▼
  Runtime Authority   Runtime      Decision      Jurisdictional
  Envelope check      Evidence     Classes       Overlays
        │             log
        │
        └── link to Multi-Agent Composition (CALIBRE-Compose)
                │
                ▼
        Attestation Report PDF (6 frameworks + evidence log)

Where each menu item fits

Menu group	Items	Purpose
Top level	Home, Register, Trust Scores	Portfolio-wide entry points
Assess	New AI, Existing AI, Policy query	Entry points to the assessment pipeline and policy retrieval
Governance (v1.1)	Runtime Authority, Runtime Evidence, Decision Classes, Jurisdictional Overlays	Runtime enforcement and jurisdictional customisation
Regulatory (v1.2)	Attestation Reports, Multi-Agent Composition	Regulator-facing outputs and multi-agent scoring
Scenarios	Red-team, Audit prep, Incident response	Pre-built worked flows
Help & About	Architecture (original + v1.3), Help centre	Documentation and scenario walk-throughs

Trust Score composite

Trust = 0.40 × (100 − ARA_residual_pct)
      + 0.30 × (100 − band-consistent GII)
      + 0.30 × evidence_coverage_pct

band-consistent GII = max(GII, lower limit of final band: 0 / 26 / 51 / 76)
No ARA run (CALIBRE-only)      -> Trust Score not available
No regulations to cover        -> weights renormalised to 0.40/0.70 and 0.30/0.70
Rounded half up (62.5 -> 63)

CALIBRE Governance Intensity Index

GII = 0.25 × (autonomy/4) × 100
    + 0.25 × (consequence/4) × 100
    + 0.20 × (irreversibility/4) × 100
    + 0.15 × (data_sensitivity/4) × 100
    + 0.15 × (population_scale/4) × 100

Final band = max(
    band_for(GII),
    consistency_floor(factors),
    amplifier_floor(amplifiers)
)
+ 1 level uplift if Compositional-Coupling amplifier is active (cap at Critical)

Multi-Agent Composition GII

Weighted mean of the seven compositional risk class scores:

C-GII = 0.20 × CC-01 (Cascade Failure)
      + 0.15 × CC-02 (Trust Chaining)
      + 0.20 × CC-03 (Authority Escalation)
      + 0.10 × CC-04 (Feedback Loop)
      + 0.10 × CC-05 (Context Contamination)
      + 0.15 × CC-06 (Handoff Injection)
      + 0.10 × CC-07 (Emergent Coordination)

If C-GII >= 40, Compositional-Coupling amplifier auto-activates
on the linked initiative and uplifts its band by one level.

Cite as: Dani, C. (2026). CALIBRE: A Governance Intensity Framework for Artificial Intelligence Systems, with Architecture-Sensitive Amplification. Zenodo. DOI to follow.