Mind · Strategy
MarginalValueRule
Deepen only while expected improvement plus an exploration credit, divided by incremental cost, exceeds a domain threshold
Definition
Mechanism
The economic governor of depth for a RecursionDive. At any node, it deepens if and only if the expected improvement from going deeper, plus an exploration credit U(n), divided by the incremental cost, exceeds a domain-appropriate threshold θ and budget remains. Improvement and cost come from an Estimate grounded where possible in PathwayMemory records of what deepening similar problems has yielded and cost; without such history the estimate falls back to a calibrated prior, and without that to a fixed exploration budget spent on shallow probes. Incremental cost includes the interface load of an added boundary, such as translation, serialization, routing, synthesis, monitoring and semantic loss, and the value term credits any error or risk reduction that makes that load worthwhile. U(n) is inversely proportional to how much Pathway Memory exists at the node, so rarely visited nodes receive a temporary subsidy and the rule trades exploration against exploitation. θ is not a fixed depth limit: another problem or later evidence may justify more detail. Across sibling branches, the remaining Budget goes to the branch with the highest ratio until no branch clears θ or the budget runs out, so problems get precision in proportion to their stakes.
Invariants
- Deepens if and only if (expected improvement + U(n)) / incremental cost exceeds θ and budget remains.
- Incremental cost includes the interface load of the added boundary.
- θ is set for the domain, not by depth: it does not rise level by level.
- Never approves a dive that exceeds the remaining Budget.
- Compares future value only; prior investment does not count (SunkCostIgnore).
Risks
Failure modes
Overestimation
Expected value is too optimistic, wasting budget on low-value dives.
Underestimation
Expected value is too pessimistic, stopping too early on valuable problems.
Sunk Cost Fallacy
Continuing to invest because of prior investment, not future value.
Exploration starved
U(n) is left out, so rarely visited nodes are never deepened and their Pathway Memory never forms.
Interface load ignored
the cost of an added boundary is left out, so the rule deepens where the boundary costs more than it adds.
Relationships
Builds on
References
Infrastructure
Resource allocation limit
Predictive resource costing
Learned routing memory: problem_class → route → outcome_quality
Vertical traversal of the solution tree
Rationality over historical investment
Used by
References
Low-cost prototype generation
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
Before decomposing a concrete problem, locate it by climbing specialization links to the root, then retrace the route
Breaking a concept into solver-contract-bound sub-concepts — the move that enables fractal recursion
Composes with
Decompose quality-constrained production into substance, structure, generation, evaluation, and refinement — five intended low-coupling dimensions a producer must address
Expansion of cognitive capability through recursive decomposition of concepts into contract-bounded sub-concepts
Decompose empathetic understanding into situation, emotion, intent, response, and boundary — variable-depth recursion tracks the stakes of the interaction
Solvers that operate on the solver tree itself — monitoring topology, detecting redundancy and drift, arbitrating across decomposition paths
Layered epistemic verification (cache → coherence → decomposition → specialists → empirical), with verified claims persisting as content-addressed records — the accretion loop
Reference
Parameters
thetaRatio, (0.0, inf)- Domain-appropriate threshold that the value-to-cost ratio must exceed; not a fixed depth limit.
Signature
Budget(RecursionDive)