ScienceReference Model
Paper 1: AIJIM Reference Model
Canonical invariants, RFC status, and paper citation anchors after the Core-first integration sequence.
Canonical citation anchor
For formal scope statements in papers, cite
services/aijim-core/CORE_SCOPE.mdrather than summarizing architecture pages from memory.Reference-Model Framing
The reference model remains the paper-level statement of what an evidence-first decision system must guarantee: evidence-bound outputs, artifact-mandatory runs, audit-safe traceability, oversight-preserving control, and governance separation.
What changed after the LINA/Core absorption is not the model itself, but the enforcement locus: the paper-facing claims now map more cleanly to the Core-owned runtime rather than a split TypeScript/Python write surface.
Canonical Invariants
| Invariant | Canonical Definition | Spec / Source Anchor | Enforcement Locus | Error Codes |
|---|---|---|---|---|
| I1 | Evidence-bound outputs | services/aijim-core/src/aijim_core/spec/ and RFC-001 / RFC-009 | evidence canonicalization, bundle fingerprinting, immutable evidence paths | — |
| I2 | Artifact-mandatory runs | services/aijim-core/src/aijim_core/spec/run_contracts.py and RFC-003 | run contracts, publication cross-link requirements, audit artifact surfaces | P0002 |
| I3 | Audit-safe traceability | RFC-009 and audit replay/publication cross-reference rules | replay head checks, self-hash uniqueness, cross-reference enforcement | P0001, P0008 |
| I4 | Oversight-preserving control | RFC-006 and review-gate / publication gate semantics | review gates, publication user-confirmed enforcement, HITL write boundary | P0007 |
| I5 | Governance separation | RFC-006 and current canonical mapping in `CORE_SCOPE.md` | membership checks, SECURITY DEFINER layers, signer operator allowlists | P0003–P0006 |
RFC Status Summary
| RFC | Topic | Status | Paper Relevance |
|---|---|---|---|
| RFC-001 | Evidence Object and Store | ACCEPTED | Foundational paper/runtime contract |
| RFC-002 | Story and Claims | ACCEPTED | Reference-model entity layer |
| RFC-003 | Artifacts and Runs | ACCEPTED | Experimental reproducibility anchor |
| RFC-004 | Scientific Conformance | ACCEPTED | Methodology and conformance bridge |
| RFC-005 | Runner Registry and Model Connectors | ACCEPTED | Execution and determinism contract now enforced in runtime |
| RFC-006 | Policy and Governance Layer | ACCEPTED | Oversight and governance substrate |
| RFC-007 | Story Rendering and Narrative Projection | ACCEPTED | Projection layer, not the audit spine |
| RFC-008 | EU AI Act Readiness | ACCEPTED | Compliance framing around the runtime |
| RFC-009 | Audit Bundle Portable Verification | ACCEPTED | Paper-critical audit verification contract |
| RFC-010 | Federated Evidence Sharing | TARGET | Future protocol scope, no current runtime baseline |
| RFC-011 | LINA / Core Integration Boundary | ACCEPTED | Formal owner split for post-E.3 audit runtime |
| RFC-012 | API Surface Repositioning | ACCEPTED | Formal namespace truth for the Core-first surface |
Paper Citation Anchors
- Invariant definitions and scope boundary:
services/aijim-core/CORE_SCOPE.md - Spec-level run contracts:
services/aijim-core/src/aijim_core/spec/run_contracts.py - Audit bundle implementation anchor:
services/aijim-core/src/api/routers/audit_bundle.py - Evidence canonicalization:
services/aijim-core/src/aijim_core/evidence_canon.py - Conformance enforcement:
services/aijim-core/src/aijim_core/infrastructure/conformance/