What Must Remain Possible

Abstract A system may retain an accurate account of earlier experience yet be unable to use it when needed. The supporting warrant may have expired, and obtaining a fresh observation may take longer than the remaining opportunity allows. We investigate this distinction between retained records, present warrant and timely operational readiness in two finite, deterministic campaigns. Developmental Context 01 (DC-P1) integrates experience-dependent consequence learning with clarification, withdrawal and continuous control, but misses its full constructive endpoint and exposes a late-recovery failure. Selective Readiness 01 (SR-P1) prospectively tests the resulting hypothesis: learned contextual trajectories can select capabilities for preparation before future tasks are disclosed, while forecasts remain separate from evidence and authority. In a finite matrix of 72 configurations per route, the Orchard implementation and a conventional implementation using the same learning family and scheduling policy each complete all 24 nominated primary tasks using 48 modelled maintenance units. The nonselective controller completes the same tasks using 156 units, while reactive recovery completes only the eight coarse tasks. Matched histories select different preparations at the same present checkpoint. Information-removal controls distinguish cheaper discrimination from task completion: broader preparation can compensate for erased experience, while removing historical recovery-time knowledge loses otherwise attainable tasks. A separate intervention-uncertainty challenge reverses the comparison: broad readiness completes 16 of 16 tasks and the selective learners complete eight. Main trajectories preserve the supplied protective floors, and unchanged replay and separately expressed audits agree. Substantial copying and resident-instance overhead remain. The results support a bounded integration of learned anticipation, timed preparation and currentness-checked use. They establish neither general policy superiority nor autonomous normative grounding. Keywords: contextual memory; selective readiness; continual computation; evidence standing; anticipatory maintenance; developmental alignment