miadi-orchestration-kit

Agent Memory Provenance Framework

RISE path: Reverse-engineer -> Intent-extract -> Specify -> Export

Control-plane scaffold for the OpenClaw model-routing study memory work. This rispec gives future waves one home for evidence and specifications about observed, inferred, and user-confirmed memory; provenance metadata; trust; scope; and write-back gates.

This folder is created for jgwill/miadi-orchestration-kit#15 and should stay aligned with miadisabelle/workspace-openclaw#80.

Served tracks

Track	Search	Role
Academic	A2	Establish provenance, operational memory, trust, and audit vocabulary.
Technical	T3	Turn trust vocabulary into schema, merge behavior, and write-back gates.
Preflight	Scaffold readiness	Verify where memory claims and schema proposals belong before research starts.

Inputs

Future waves should read these before adding findings:

1. OpenClaw model-routing study launch-manifest.md.
2. OpenClaw deep-research PROPOSAL.md.
3. llms-rise-framework.txt.
4. Transcript or study-note excerpts for memory and user ownership claims.
5. Repo path evidence for memory stores, instruction files, schema definitions, or write-back behavior.
6. External provenance, RAG citation, trust, or memory-system sources after ledger intake.

What is in this folder

File	Purpose
01-reverse-engineer.md	Reconstruct what evidence shows about memory states, provenance fields, and confirmation boundaries.
02-intent.md	State the desired outcome, current reality, structural tension, and non-goals for trusted memory.
03-specify.md	Define schema, status transitions, write-back gates, review checkpoints, and consumption rules.
04-export.md	Provide launch, resume, audit, and handoff shapes for memory-provenance work.
05-source-ledger.md	Preserve source support for every field, trust rule, write-back rule, contradiction, and provisional claim.

Acceptance criteria

• The exact RISE wording Reverse-engineer -> Intent-extract -> Specify -> Export appears in the README and is used as the folder’s stage language.
• A2 and T3 share one trust vocabulary instead of diverging into theory and implementation silos.
• Every proposed memory field is tied to a source-ledger row or marked provisional.
• Every write-back path has a gate that distinguishes observed, inferred, and user-confirmed records.
• Memory records preserve source, observed_at, confidence, status, scope, related_task, and user_confirmed before promotion.
• The scaffold does not run Academic, Technical, Narrative, or final research.

Source-ledger rules

• Evidence-backed claims need exact source pointers.
• Repo path evidence is required before a schema or write-back behavior is called observed.
• Transcript-only claims are provisional until corroborated.
• Provisional schema fields must name the missing evidence needed for promotion.
• Contradictions stay visible until resolved.
• Human confirmation claims must identify whether the human explicitly confirmed, denied, bounded, or merely implied the memory.

Boundary

This rispec does not write memory, migrate memory, alter instruction files, or decide production memory policy. It defines the control surface for future evidence and specification work.