Initial Commit

experiments/003_rig_trials/hypothesis.md

@@ -0,0 +1,47 @@
+# Experiment 003 Hypothesis
+
+## Primary Comparison
+
+One persistent rig is exposed to three qualitative requirements:
+
+- **Haul:** tool exposure, cargo mass, and navigation;
+- **Furnace:** drive heat, ambient heat, ward orientation, and cooling layout;
+- **Gale:** changing wind, ballast anchoring, directional thrust, and maneuvering.
+
+The comparison is not which trial is best. It is whether the player changes a physical construction for reasons learned by piloting it, and whether knowledge about parts transfers without one whole layout solving everything.
+
+## Competing Explanations
+
+1. Construction is valuable when it becomes the player's directly controlled capability.
+2. Changing requirements create useful adaptation even if any single trial is shallow.
+3. Direct piloting adds surface action but construction remains obvious loadout selection.
+4. Physical placement is too weak or too opaque, so only part counts matter.
+5. A universal rig crystallizes immediately.
+6. Cheap revision still feels like repetitive implementation rather than reasoning.
+
+## Evidence Priorities
+
+Strong positive evidence:
+
+- rebuild after a concrete observed failure;
+- preserve a useful subassembly while changing other structure;
+- deliberately exploit exposure, adjacency, mass, or directional behavior;
+- voluntary trial or layout experiment not needed for completion;
+- desire to improve or test the rig after a pass.
+
+Weak evidence:
+
+- completing all trials;
+- spending a long time driving;
+- repeatedly retrying the same build;
+- placing the part named by a trial description.
+
+## Confounds
+
+- poor driving feel;
+- an impossible or trivial tuning threshold;
+- unclear placement/rotation controls;
+- trial geometry requiring dexterity rather than construction;
+- the starter rig being too competent;
+- checklist-like descriptions prescribing the solution.
+