Skip to content

specs/004-memory: three independent edge-strengthening designs coexist with no authoritative statement #6635

Description

@bug-ops

Description

The specs quality audit in PR #6629 found that `specs/004-memory/` documents three independent, unreconciled mechanisms for how a memory graph edge gets stronger with repeated access:

  • A-MEM's `retrieval_count` boost
  • Benna-Fusi's `confidence_fast`/`confidence_slow` dual-rate model
  • HeLa-Mem's Hebbian `weight` update

Each is documented in its own sub-spec as if it were the mechanism, but nothing in `004-memory/spec.md` or elsewhere states which one is authoritative when SYNAPSE (or whichever component actually traverses the graph at recall time) needs to pick a single edge-strength signal.

Expected Behavior

`specs/004-memory/spec.md` (or a dedicated sub-spec) states which edge-strengthening mechanism is authoritative in the current implementation, and how the others relate to it — e.g. one is the shipped mechanism and the others are superseded RFC drafts, or they operate at different layers and combine in a defined way.

Actual Behavior

Three specs each describe their own mechanism with no cross-reference resolving which one governs real graph traversal, which risks divergent implementations or contradictory future changes.

Environment

Logs / Evidence

See PR #6629 description, "Flagged, not executed" section, item 5. Cross-check `crates/zeph-memory/src/` for which mechanism is actually wired into recall/traversal today before writing the resolution.

Metadata

Metadata

Assignees

No one assigned

    Labels

    P2High value, medium complexityarchArchitecture and designdocsDocumentationmemoryzeph-memory crate (SQLite)

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions