Conformance & Evaluation
"Can I prove what the agents did, why they did it, and whether the system stayed within approved lifecycle guarantees?"
MPLP's answer is: evidence-first lifecycle governance, not "trust the prompt" or "trust best practices".
2. What This Section Is NOT
This section does not provide:
- Certification, endorsement, or compliance badges
- Legal advice or regulatory determinations
- A mandatory runtime, SDK, or framework adapter
- "One true" implementation pattern for agents
MPLP is vendor-neutral by design. Conformance is judged by evidence, not by brand names or frameworks.
3. How to Use This Section
| Your Goal | Start Here |
|---|---|
| Understand conformance model | Conformance Model |
| Understand evidence structure | Evidence Model |
| See evaluation axes | Evaluation Dimensions |
| Understand result semantics | Results & Status |
| See future plans | Roadmap |
| Practical evaluation steps | Conformance Guide |
| Validate with Golden Flows | Golden Flows |
4. Core Principles
4.1 Evidence-First
Conformance is judged by exported evidence, not runtime inspection:
| Evidence Type | Purpose |
|---|---|
| Plan | Intent declaration |
| Confirm | Governance gate record |
| Trace | Execution history |
| Snapshot | State checkpoint |
| Manifest | Configuration declaration |
4.2 Vendor Neutral
Evidence does not depend on proprietary runtime features to be interpretable. Any conformant runtime MUST produce evidence that:
- Uses MPLP JSON Schemas
- Is self-describing (contains protocol version)
- Is replayable (trace can reconstruct timeline)
4.3 Non-Certification
MPLP does not provide:
- Certification programs
- Compliance badges
- Runtime endorsements
Conformance is a binary outcome (conformant / non-conformant / incomplete-evidence), not a score or certification level.
5. Minimal Conformance Checklist
Use this as a quick sanity check before deeper evaluation:
| # | Check | Evidence Source |
|---|---|---|
| 1 | Protocol version declared | meta.protocolVersion in artifacts |
| 2 | Plan → Confirm → Trace chain exists | Linked IDs across objects |
| 3 | High-risk actions gated | Confirm objects for gated steps |
| 4 | Replayability | Trace segments with timestamps |
| 5 | Bounded failure | Recovery events or safe-stop records |
6. Conformance Documents
| Document | Purpose |
|---|---|
| Conformance Model | Conformance classes (L1/L2/L3) and outcomes |
| Evidence Model | What constitutes valid evidence |
| Evaluation Dimensions | 6 axes for judging conformance |
| Results & Status | Outcome semantics and reporting |
| Roadmap | Future plans and boundaries |
| Conformance Guide | Practical evaluation workflow |
| Conformance Checklist | Vendor self-verification template |
7. Related Documentation
- Golden Flows — Lifecycle invariant validation
- Versioning Policy — Protocol version semantics
- Schema Mapping Standard — Evidence schema definitions
Document Status: Informative (Navigation Entry Point)
Scope: Evidence-based conformance evaluation
Exclusions: Certification, legal compliance, vendor endorsement