Initial Commit

experiments/008_catalyst_ecology/hypothesis.md

@@ -0,0 +1,73 @@
+# Experiment 008 Hypothesis — Private Until After Play
+
+## Question
+
+Can the compositional interest found in Experiment 007 survive a strict four-choice budget and changing enemy ecologies, or does it collapse into a universal package or obvious counter-loadouts?
+
+## Why This Experiment
+
+Experiment 007 produced the first complete curiosity chain in the project. Fork, Arc, and Bloom interacted through shared hit/kill interfaces; an inaccurate prediction led to surprising behavior, revised knowledge, and several self-directed build runs. The player also spontaneously distinguished Bloom as useful against small enemies and Arc against large enemies.
+
+That success exposes the original long-term risk: strategy crystallization. The full three-family package may be universally best, all-Bloom may erase every population through runaway scaling, or visible enemy types may turn adaptation into the prescribed part swapping that made Experiment 003 boring.
+
+Catalyst Ecology retains the same movement/shooting substrate and the known Fork/Bloom/Arc vocabulary. It adds three components at narrow existing interfaces:
+
+- **Focus:** repeated primary hits on one body cause a rupture;
+- **Conduit:** secondary-effect hits charge a lance toward the healthiest body;
+- **Resonance:** amplifies secondary-effect damage, but has no output alone.
+
+Three expeditions—Shoal, Bastion, and Brood—use small bodies, large durable bodies, and spawning mixtures respectively. Each run permits only four choices, with duplicates allowed and all six catalysts offered every time. Revision 2 retains the complete build for fields five through eight after the first run showed that a single mature field was insufficient for the system to become observable.
+
+## Competing Interpretations
+
+1. Conditional enemy ecology sustains build questions and makes knowledge transfer without transferring one whole answer.
+2. Ecology produces obvious “Bloom for small, Focus/Arc for large” counter-loadouts with no discovery.
+3. The original three-effect package remains dominant, so the larger option space adds false choices.
+4. All-Bloom still wins through raw exponential scaling and spectacle.
+5. New cross-interface components create multiple viable causal routes: Fork/Bloom feeds Arc/Conduit, Focus starts kills that feed Bloom, and Resonance trades a mechanism slot for stronger products.
+6. Four slots are too restrictive and feel like arbitrary prevention rather than productive composition.
+7. The successful part of 007 was rapid escalation in one repeated field, not adapting between ecologies.
+
+## Evidence Priorities
+
+Strong evidence:
+
+- a build is chosen based on a prediction about an ecology and then revised from its observed result;
+- the same component is valued differently across expeditions for causal reasons;
+- a new interaction creates another self-selected build question;
+- replaying an expedition with a meaningfully different composition;
+- more than one plausible solution is articulated for Brood/mixed populations;
+- a powerful cascade remains legible enough that the player attributes it to selected interfaces.
+
+Ambiguous evidence:
+
+- clearing all expeditions;
+- fastest clear time or largest cascade;
+- taking the visually labeled small/large counter;
+- using all six components across required runs;
+- longer play caused by extra content.
+
+Failure evidence:
+
+- immediate obvious loadouts from the descriptions;
+- one four-choice sequence used everywhere;
+- repeated all-Bloom erasure without further questions;
+- the slot limit makes choices feel like withholding obvious synergy;
+- Brood spawning becomes clutter/attrition rather than an exploitable ecology;
+- new trigger chains are unreadable noise.
+
+## Analysis Guardrails
+
+- Do not nerf a spectacular build merely because it is strong; determine whether strength ends or creates reasoning.
+- Do not infer adaptation from different builds alone; ask whether the difference was predicted, sampled, or prescribed.
+- Compare stop desire and new questions, not expedition duration.
+- Focus and Conduit add some automatic targeting, so reduced aim burden remains a confound.
+- Expeditions are deliberately not numerically matched. They test qualitative population response, not performance balance.
+
+## Feedback Questions
+
+After telemetry inspection:
+
+1. What question or expectation, if any, drove each build and replay?
+2. Did any expedition admit multiple interesting approaches, or did the populations prescribe their counters?
+3. Did a new interaction create another thing they wanted to test, and when were they ready to stop?