Initial Commit

experiments/008_catalyst_ecology/results/0ac8e60d-preliminary-analysis.md

@@ -0,0 +1,113 @@
+# Experiment 008 Preliminary Analysis — Session 0ac8e60d
+
+Status: first-run telemetry and report complete; corrective revision required before interpreting H23.
+
+Source: `JSONL/catalyst-ecology-0ac8e60d-a488-4d4d-a4d4-3d220981ef67.jsonl`
+
+## Session Structure
+
+- 2,042 events over 374.0 seconds from initialization to save.
+- Shoal, Bastion, and Brood were played in displayed order and each completed once.
+- All 15 fields cleared on their first attempt. There were no defeats and six total damage events.
+- No expedition was replayed, restarted, switched away from, or abandoned.
+- The player saved once after all three expeditions.
+
+Completion is likely compliance with the requested comparison. The lack of replay is meaningful only after asking whether the option space felt exhausted, the player had already formed conclusions, or no build question justified another run.
+
+## Build Sequences
+
+| Expedition | Choice sequence | Final build | Active run time |
+|---|---|---|---:|
+| Shoal | Bloom → Resonance → Arc → Focus | Bloom, Resonance, Arc, Focus | 48.9 s |
+| Bastion | Conduit → Bloom → Arc → Fork | Conduit, Bloom, Arc, Fork | 78.6 s |
+| Brood | Bloom → Focus → Arc → Fork | Bloom, Focus, Arc, Fork | 87.5 s |
+
+Bloom and Arc formed a common core across all three populations. The two remaining slots varied:
+
+- Shoal uniquely used Resonance and omitted Fork/Conduit.
+- Bastion uniquely used Conduit and omitted Focus/Resonance.
+- Brood used Focus and Fork, omitting Conduit/Resonance.
+
+Every one of the six available modules was selected somewhere, but no module was stacked. This may reflect ecology-based choice, coverage/sampling of new options, or a belief that one level of several interacting mechanisms is better than specialization.
+
+## Deliberation Pattern
+
+The first choice in each expedition took much longer than later choices:
+
+- Shoal first choice: about 38.9 seconds.
+- Bastion first choice: about 33.5 seconds.
+- Brood first choice: about 25.5 seconds.
+- Later choices generally took 1–6 seconds.
+
+The player therefore spent substantial time reading or planning when the build was empty, then selected quickly as the run developed. The decreasing first-choice times could indicate growing understanding of the six-module vocabulary, fatigue, or a progressively clearer plan.
+
+Conduit was selected first in Bastion despite having no output without secondary hits. It remained inactive during field two, then began firing after Bloom was added. Sixteen Conduit lances ultimately hit and six killed targets. This is potentially strong evidence of planning around a future interface, but it could also be experimentation with an attractive description or a misunderstanding corrected by the next choice.
+
+## Population and Effect Results
+
+### Shoal
+
+- 99 kills: 93 Motes and 6 Husks.
+- Bloom sparks caused 52 kills; Arc caused 12; Focus caused one.
+- Field time fell from 13.9 seconds at baseline to 7.1 seconds in the final field despite a much larger population.
+- Focus arrived last and ruptured only once, consistent with most targets dying too quickly to accumulate repeated hits.
+- The secondary budget never approached exhaustion (observed minimum 214/300).
+
+### Bastion
+
+- 40 kills: 9 Titans, 16 Husks, and 15 Motes.
+- Primary fire still caused 27 kills; Conduit lances caused six.
+- Conduit fired/hit 16 times after secondary generators became available.
+- Final fields remained around 18–20 seconds, so the build did not erase durable populations.
+- The secondary budget remained ample (observed minimum 201/300).
+
+### Brood
+
+- 79 kills, including 27 Motes spawned during combat.
+- Bloom produced 103 spark hits; Fork produced 46 fragment hits; Arc triggered 32 times; Focus ruptured eight times.
+- Final build used all four direct mechanisms and produced all four corresponding kill causes.
+- Field time increased with the population from 8.5 seconds to 22.1 seconds.
+- The lowest observed secondary budget was 134/300; the recursion cap did not determine the outcome.
+
+## Preliminary Interpretation
+
+Experiment 008 did not immediately collapse into one identical four-slot answer. The player used all six modules across three distinct builds, and the modules materially interacted. It also did not provoke the repeated build tests seen in 007.
+
+The common Bloom+Arc core is the central ambiguity. It may mean:
+
+1. knowledge from 007 transferred successfully as a reusable component pair while the remaining slots adapted;
+2. Bloom+Arc is already a partially crystallized universal answer, leaving only two low-stakes flex slots;
+3. the player was deliberately holding a familiar baseline while sampling the three new modules;
+4. the choices were made for novelty/coverage rather than population response.
+
+The three expedition builds look plausibly ecology-aware, but telemetry cannot establish why. Shoal's Bloom+Resonance generated strong small-target cascades. Bastion's Conduit converted secondary activity into attacks on the healthiest target. Brood combined Focus for durable spawners with Fork/Bloom/Arc for their Motes. Those relationships may have been planned, noticed afterward, or entirely incidental.
+
+Unlike Experiment 003's obvious counter-parts, no module names a specific enemy class. Components still operate through hit/kill interfaces. However, the population descriptions themselves may make the build split obvious enough that it becomes counter-loadout work. The player's subjective decision process is necessary to discriminate H23.
+
+The four-choice budget did preserve tradeoffs: no run could include all six modules. But it may also have suppressed the power-escalation payoff that motivated 007's specialization replays. No repeats were selected, including no all-Bloom attempt, and no run created a logged follow-on test.
+
+## Follow-up Needed
+
+1. What expectation drove each distinctive first choice—especially Bloom for Shoal/Brood and inactive-until-supported Conduit for Bastion—and were later choices completing a plan or sampling?
+2. Did the populations create several plausible builds, obvious counters, or a universal Bloom+Arc core with two unimportant flex slots? How did the four-choice limit feel?
+3. Did any result surprise or satisfy enough to suggest another build, and why did the player stop after one run per expedition rather than replaying as in 007?
+
+## Player Report
+
+The distinct builds were not substantially planned. Conduit was chosen first because its dependency was misunderstood, not because the player was deliberately investing in a future secondary-hit engine. Do not cite that sequence as forward planning.
+
+The player believes the populations may permit multiple approaches, but could not formulate which effects played off one another well because each expedition contained too few rounds. They described the repeated experience as beginning to see something cool emerge and then having the expedition end before anything really cool happened.
+
+## Final First-Run Interpretation
+
+Experiment 008 failed to measure strategy half-life because it ended at build maturity. Four selections occurred after the first four fields, leaving only field five for the complete build:
+
+- Shoal's complete build existed for 7.1 seconds.
+- Bastion's complete build existed for 19.8 seconds.
+- Brood's complete build existed for 22.1 seconds.
+
+The larger six-module vocabulary and changing populations required more observation than Experiment 007, but the run structure supplied no additional mature-build time. Lack of replay is therefore not good evidence that the system exhausted its questions. The player's report and telemetry agree that the prototype terminated during emergence rather than after it.
+
+The proper correction is structural, not a new mechanic or balance pass: preserve the exact four choices and current populations, then add several post-build fields so the player can observe the same completed composition against increasing and mixed populations. This keeps the slot-budget test intact while allowing H23 to become measurable.
+
+Do not add replacement choices yet. Swapping would bundle ecology adaptation with interface/optimization work before establishing whether a mature four-module build is interesting to watch and understand. Do not increase the choice budget; the current question specifically concerns composition under exclusion.