Back to Architecture
Core Framework

RIGOR Framework

Research, Inspect, Generate, Optimize, Review - The 5-phase execution framework governing all autonomous agent operations.

The Five Phases

R
I
G
O
R

Phase 1: Research

R

Gather context, validate assumptions, and identify constraints before taking action.

Activities

  • Analyze codebase structure and patterns
  • Retrieve relevant knowledge from graph
  • Identify dependencies and constraints
  • Validate initial assumptions
  • Build execution context

Outputs

Context analysisConstraint mapKnowledge retrievalAssumption validation

Phase 2: Inspect

I

Audit safety, verify constraints, and run pre-execution quality gates.

Activities

  • Run security audit checks
  • Verify preconditions met
  • Execute multi-dimensional quality gates
  • Check reversibility requirements
  • Validate bounded autonomy

Outputs

Security reportGate resultsRisk assessmentGo/no-go decision

Phase 3: Generate

G

Execute the task, produce artifacts with confidence scoring and decision documentation.

Activities

  • Execute primary operation
  • Generate artifacts (code, config, etc.)
  • Calculate confidence scores
  • Document decisions and reasoning
  • Select appropriate patterns

Outputs

Generated artifactConfidence scoreDecision logPattern selection

Phase 4: Optimize

O

Tune performance, minimize costs, and trigger learning from the execution.

Activities

  • Performance optimization
  • Cost minimization analysis
  • Feedback integration
  • Pattern effectiveness update
  • Learning trigger

Outputs

Optimized resultCost analysisLearning observationsFeedback loop

Phase 5: Review

R

Validate outcomes, run A/B tests, prepare rollback, and record success metrics.

Activities

  • Validate execution outcome
  • Run hypothesis testing
  • Prepare rollback plan
  • Record success metrics
  • Update knowledge graph

Outputs

Validation reportTest resultsRollback planSuccess metrics

Quality Gates

Every change passes through a set of quality gates before it is applied. The gates fall into two groups: operational quality gates that check the change itself, and governance gates that check how it was produced. A gate that fails stops the execution; it cannot be bypassed for high-risk operations.

Operational Quality

Confidence

The agent must be confident enough in its own result before anything is applied

Security Scan

Vulnerability detection on the generated change

Test Coverage

A minimum level of test coverage on the code the change touches

Code Review

Automated review checks must pass

ACMF Validation

The operation must be permitted at the agent's current maturity mode

Performance Check

No material regression against the performance baseline

Compliance Check

Regulatory and policy rules are validated

Dependency Audit

No critical vulnerabilities in dependencies

Enterprise Governance

Runtime Confidence

Confidence is monitored while the task runs, not only at the end

Context Quality

Detects polluted or stale context before it shapes a decision

Audit Trail Present

Every execution carries a complete decision record (EU AI Act readiness)

Model Drift

Output distribution is compared against its baseline

Cost Anomaly

Spend that spikes above the expected baseline is stopped

Rollback Capability

A rollback path must exist before a change is applied

Human Approval

High-risk operations wait for an explicit human decision

Supply Chain Integrity

Dependencies are verified before they are trusted

Cascade Protection

Failure isolation is confirmed so one fault cannot spread

ACMF Integration

RIGOR works in conjunction with the ACMF (Agent Capability Maturity Framework) to control execution flow based on agent maturity level.

Execution Flow with ACMF

RESEARCHRINSPECTIGENERATEGOPTIMIZEOREVIEWRACMFCheckContinueOBSERVE / LEARNINGStatus: awaiting_approvalASSISTEDCritical ops: await approvalAUTONOMOUSContinue executionPause until approvedNormal ops: continueAudit trail onlyAll phases produce audit records for EU AI Act compliance

Runtime Governance (ACMF)

  • • Controls execution during RIGOR phases
  • • Based on agent maturity level
  • • Automatic progression/regression
  • • Per-agent state persistence

Pre-Deploy Governance (Gates)

  • • Validates before deployment starts
  • • Based on operation risk level
  • • Cannot bypass for high-risk ops
  • • Pending approvals expire rather than wait indefinitely

Continue Exploring

Learn more about related architectural components

ACMF Maturity ModelSafety ArchitectureArchitecture Overview