# 0003 — The nursery: boot process as curriculum 2026-08-21. Researcher ruling, verbatim: > "you have to teach children. there is a physical world boot process. > a perturbable thing started with some default shapes needing > perturbation and then blossoms" ## What this settles about the feeding protocol (open question 3) The instrument is not configured; it is *raised*. Day-zero state is a small set of default shapes — the constitution's §79 elementary structures (cycle, tree, feedback, inversion, containment, branching, state transition) standing in for what the physical world hands a child (persistence, containment, support, before/after, part/whole). Taste, trace field, and weighting grow by working through a CURRICULUM of these shapes rendered in multiple representation types and subjected to graded perturbations — before any culturally loaded material is ever ingested. The defaults *need* perturbation: an unperturbed default does not blossom. A curriculum item is therefore not a shape but a triple — shape × renderings × perturbation family — and what the instrument keeps across the perturbations becomes its first taste. Near-misses enter at birth (ADR-0002 ruling 4): the nursery deliberately includes shapes with a piece missing, so the control population grows up alongside the taste rather than being bolted on later. ## The developmental-psychology datum (engineer's addition) Piaget's conservation tasks are literally this project's experiment class run on children: does quantity survive the pouring into a taller glass? — i.e., what survives a transformation of representation? Object permanence is invariance under occlusion; the A-not-B error is a trace field (footprints) overriding current observation. The physical world's boot process has a published experimental literature, and it is a catalogue of invariance tests. Candidate source for curriculum design; NOT to be imported as ontology (§8) — as a menu of perturbation families that a well-studied boot process is known to include. ## Consequences 1. Experiment 1 (§79) is promoted from calibration to **the nursery** — its constructed artifacts are the default shapes; running it IS the instrument's childhood, and its outputs seed the trace field and the day-zero weighting (replacing hand-set GED costs sooner than ADR-0001 assumed). 2. **Curriculum permutation** becomes a first-class deferred experiment: raise two instances of the trace field on the same shapes in different orders; compare their tastes on identical later artifacts. Path dependence (§37) tested on the instrument itself, cheaply, with the local model. If order doesn't matter, good — Markovian taste. If it does, the curriculum is part of the instrument's identity and must be recorded as such (§50: the childhood is instrument metadata). 3. The sapien-logos pattern (seed text booting a being) is explicitly NOT reused here: the nursery seeds with structure, not content. What boots this instrument is shapes-under-perturbation, not a text.