A control model for bounded search using pressure, causal progress, hypotheses, and explicit control.
18 · Cognition & Control
Natural Search Control Plane
in GenOS.
A control model for bounded search using pressure, causal progress, hypotheses, and explicit control.
DEFINITION AND SCIENCE
What it means
in GenOS.
The control plane combines search monitoring, hypotheses, and explicit stop or continue decisions. This is a software control model; biological metaphors do not show it is adaptive or effective.
Adaptive foraging illustrates choice under limited resources. GenOS borrows bounded search and stopping decisions, not a biological optimizer.
Observe and decide → Execute and checkpoint → Resume under evidence gates
At step t, choose action a_t from eligible actions by a declared score J(a|state), subject to resource, permission, and stopping constraints. Log predictions before observing outcomes to avoid hindsight bias.
ANIMATED DIAGRAM · INTERACTIVE MODEL
01 / 03Explore the mechanism: Natural Search Control Plane
A qualitative teaching model of this concept. It does not run GenOS or validate its implementation.
01Measure search pressureINPUT
02Choose a bounded actionMECHANISM
03Stop or update hypothesisCONDITIONAL OUTCOME
Current transition · Step 1Measure search pressure
The modeled transition may proceed. Change the assumption and move through the stages to inspect the decision path.
PROCESS AND USE
How it works
at a glance.
Observe and decide
Attribute runtime events to their agent, preserve hypothesis identity and provenance, then select a bounded process using pressure, radius and hysteresis.
Execute and checkpoint
Run the selected module and atomically commit its versioned state before reporting success. Serialize agent operations and reject stale writers or corrupt checkpoint formats.
Resume under evidence gates
Reload the committed ledger and seven module states after restart. Cultural delivery is scoped and idempotent; received traits remain candidates, not promoted decisions.
RUNTIME IMPLEMENTATION
What is connected,
and how far.
The published 21-script suite tests SQLite reopen, partial-write crashes, concurrency, failed flushes, provenance and cultural delivery. Recovery restores committed internal search state, not an external workspace or a complete agent history. The 12-task planning harness does not establish a causal production benefit from phases 6-12; cultural validation still requires explicit observed or verified evidence.
Phases 1-12 are connected to the backend runtime. Targeted tests cover atomic checkpoints, seven restored module states, crash recovery, concurrent events, stale writers and scoped cultural delivery. Comparative production gains and generalization remain unqualified.
Reporting success before the checkpoint commits. · Restoring a partial projection instead of the committed state. · Treating a transmitted candidate or self-reported tool result as independently verified progress.
PRIMARY RESEARCH · MECHANISM LEVEL
The mechanism
behind the analogy.
The design of experiments
Ronald A. Fisher · Oliver and Boyd
Original experimental-design treatment; GenOS comparisons need matched inputs and a declared measurement protocol.
Open publication ↗Primary research describes the cited mechanism; specific technologies may instead cite their original technical documentation. These sources provide context and do not validate the GenOS implementation.
EVIDENCE AND LIMITS
Internal recovery is not external replay
The published 21-script suite tests SQLite reopen, partial-write crashes, concurrency, failed flushes, provenance and cultural delivery. Recovery restores committed internal search state, not an external workspace or a complete agent history. The 12-task planning harness does not establish a causal production benefit from phases 6-12; cultural validation still requires explicit observed or verified evidence.