Prototype 03
Ralph Loop8do-mcp
Verdict-gated task orchestration. Evidence-based verification with ACCEPT/REVISE/REJECT gates and circuit breaker safety. Philosophy: "Success = Imperfect AI x Persistent Loop x External Verification".
Key Capabilities
Evidence over opinion, safety over speed
Verdict gates
ACCEPT/REVISE/REJECT based on evidence, not opinion
Circuit breaker
Configurable failure threshold halts runaway loops
Priority queue
Tasks processed in priority order with lane separation
Retry with feedback
REVISE loops carry specific failure context for targeted fixes
Verdict Gates
Three possible outcomes per task cycle
ACCEPT
All evidence checks pass. Task moves to complete. Output is trusted.
REVISE
Some checks fail. Task loops back for another attempt with specific feedback.
REJECT
Fundamental failure or circuit breaker tripped. Task halted with reasoning.
Evidence Checks
External verification sources driving verdict decisions
Lint
Static analysis — code style, import ordering, unused variables
Typecheck
Type verification — mypy, tsc, or equivalent for the project language
Test
Automated test suite — unit, integration, and snapshot tests
Build
Compilation and bundling — ensures the artifact is deployable
Integration Points
Part of the Adaptivearts.ai interpretive loop infrastructure
SPINE OODA
OODA provides the reasoning cycle; 8do-mcp provides the verdict gate.
evaluation-mcp (Evalla)
Rubric-based quality scoring feeds into verdict decisions.
Additional Materials
Study materials coming soon.