Mind · Memory
ArchiveServingFrontier
Keep evaluated structures in three layers: an inclusive archive, a selective serving layer and an experimental frontier
Definition
Mechanism
Solved structures, meaning their decompositions, routes, outputs, gate results, provenance and observed outcomes, stay addressable after a task ends, but persistence does not make a carve canonical. Memory is therefore held as a portfolio in three layers. An inclusive archive retains evaluated solution graphs, failed routes included when they carry negative knowledge. A selective serving layer promotes recurrent, validated or niche-leading nodes and subgraphs into active use, and PathwayMemory learns when to route through each. An experimental frontier receives a bounded budget for fresh, convention-resistant carves when similarity is low, routing confidence is poor, frame errors recur, stakes are high or an exploration schedule fires. Exclusion from the serving layer does not discard an archived alternative, so later evidence can promote it. Familiar problems normally use one evidence-backed incumbent rather than every retained carve, and retention stays subject to permissions, expiry and deletion requirements.
Invariants
- Evaluated structures are retained in the archive whether or not they are served, failed routes included when they carry negative knowledge.
- Only recurrent, validated or niche-leading structures are promoted to the serving layer.
- The frontier's budget is bounded.
Risks
Failure modes
Serving read as truth
the incumbent in the serving layer is treated as the only carve, so archived alternatives are never reconsidered.
Archive as attic
alternatives are retained but never promoted, even when evidence favours them.
Unbounded frontier
exploration takes budget from familiar problems that the incumbent already serves well.
Retention without limits
archived structures are kept against permissions, expiry or deletion requirements.
Relationships
Builds on
References
Learned routing memory: problem_class → route → outcome_quality
Used by
References
Finding what a concept is made of: locate it, carve it into the dimensions every case it claims needs, test each, search for what is missing, route real cases through it, say how the parts make the whole, and carve again below
Four tests of a decomposition against a stated claim frame: each candidate dimension must pass Necessity, Independence and Universality, and the accepted set must pass Completeness
Breaking a concept into solver-contract-bound sub-concepts — the move that enables fractal recursion