001 integrity first architecture
ADR-001: Integrity-First Architecture¶
Status: Accepted
Date: 2025-11-09
Deciders: Technical Steering Committee
Tags: architecture, integrity, design-principle
Context¶
Traditional systems architecture follows a "trust then verify" model where integrity is checked post-deployment through audits, monitoring, and incident response. This creates a gap between when issues are introduced and when they are detected.
Mobius Systems requires a fundamental shift to "verify then execute" where integrity gates are embedded at every architectural layer.
Decision¶
We will implement Integrity-Driven Architecture (IDA) where:
- MII ≥ 0.95 is a hard gate for all deployments
- Integrity checks occur at every layer (ledger, API, service, agent)
- No operation proceeds if integrity cannot be verified
- Self-healing mechanisms automatically remediate integrity degradation
Consequences¶
Positive¶
- Proactive Integrity: Issues caught before deployment
- Reduced Incidents: Early detection prevents production failures
- Audit Trail: Every decision is cryptographically attested
- Regulatory Compliance: Meets requirements for auditable systems
Negative¶
- Deployment Latency: Integrity checks add 1-5 seconds to deployment
- Complexity: Additional infrastructure for integrity verification
- False Positives: Legitimate changes may be blocked if MII calculation is conservative
Risks¶
- MII Calculation Errors: If MII algorithm has bugs, system may incorrectly block/allow
- Performance Impact: Integrity checks add computational overhead
- Key Management: Cryptographic signatures require secure key storage
Alternatives Considered¶
Alternative 1: Post-Deployment Auditing¶
Rejected: Does not prevent issues, only detects them after the fact.
Alternative 2: Optional Integrity Checks¶
Rejected: Optional checks are often skipped under time pressure.
Alternative 3: Integrity as Advisory Metric¶
Rejected: Advisory metrics do not provide enforcement guarantees.
Implementation Notes¶
- MII gates implemented in CI/CD pipeline
- Ledger enforces MII threshold before accepting attestations
- Sentinels monitor MII in real-time and trigger safe-stop if threshold breached
Related ADRs: ADR-002 (Model-Agnostic Sovereignty), ADR-003 (Four-Cortex Design)