Skip to content

FMM_0_CLOSURE_LAB.md

Latest commit

 

History

History
62 lines (42 loc) · 1.59 KB

FMM_0_CLOSURE_LAB.md

File metadata and controls

62 lines (42 loc) · 1.59 KB
Raw

FMM-0 Closure Lab

FlowMemory is not only claiming a memory model.

It is testing the algebra around that model.

The FMM-0 Closure Lab checks whether receipt-bound machine histories remain valid when they are composed, appended, merged, and promoted. It also checks that impossible compositions fail before they can become live memory.

Run:

python tools/fmm0_closure_lab.py demo --pretty

Expected result:

Closure laws checked: 8
Valid closures preserved: 4/4
Invalid closures rejected: 4/4
Escaped: 0

Why This Is Different

Most AI-memory systems benchmark recall.

FlowMemory asks a different systems question:

Can this machine memory history be composed without breaking reality?

Closure Lab makes that question executable.

Valid memory algebra should stay inside FMM-0:

  • a matching receipt can attach to a local draft;
  • a reader-derived FlowPulse can promote only with consistency evidence;
  • monotonic same-rootfield append can preserve a live history;
  • independent rootfields can merge commutatively.

Invalid memory algebra should be rejected:

  • rootfield rollback;
  • split-brain heads;
  • pre-receipt receipt facts;
  • semantic truth overclaims.

This is the category-defining point: FlowMemory treats agent memory like a memory model, not a retrieval cache.

Non-Claims

Closure Lab is not semantic truth verification. It is not model correctness. It is not custody, swap control, production verifier infrastructure, Base mainnet deployment, or GPU acceleration.

It is the local executable proof that FMM-0 has closure behavior around the FlowPulse proof envelope.