Measured Mode
Real EI / CE — on a synthetic toy
This is the serious corner of the toy. Everywhere else, the vitality numbers are for fun. Here we run the actual causal-emergence math on the pond's mood data and show our work — the raw transition table, the bits, the honest 'is this above chance?' check. You don't need the math to read it: each section below opens with a plain-language line.
⚗️ This page runs actual effective-information and causal-emergence math (Hoel / Zhang) on the guppy simulation's mood dynamics. The numbers are real computations — but the substrate is a deterministic toy, so they describe only this toy's coarse mood model. They are not the playful pond metrics, and not a claim about real fish, brains, or people. New to these terms? The method note has a plain-language glossary.
Run a measurement
Microstate = each guppy's mood (6 states). The TPM is estimated by counting mood transitions; the macro is the emergence-maximizing grouping found by searching all 203 coarse-grainings (with a null), plus an exploratory (diagnostic-only) ΦID measure over a 2-part split. A run of ~1.5–3k ticks takes ~1–2 s.
Group memory & causal emergence
Does a school's earned memory contribute to the pond's measured causal / effective information? Measured offline (npm run experiments) — these are 10k-tick sims, not computed live. Generated 2026-06-24.
E7 · memory as-is
Group memory carries essentially no information about mood dynamics (≤0.009 bits) — it gates school identity, not affect.
| scenario | CMI (lag 100) | ceiling |
|---|---|---|
| calm | 0.0012 | 0.294 |
| discovery-bloom | 0.0091 | 0.305 |
| overgrowth | 0.0006 | 0.230 |
| autoimmune | 0.0000 | 0.029 |
| storm-recovery | 0.0001 | 0.181 |
E8 · wire memory → mood
Wire memory into mood (memoryFeedback): it becomes weakly & tunably part of the information — a sweet spot at moderate feedback, not a switch; mood CE barely moves.
| scenario | CMI off→on | moodCE |
|---|---|---|
| calm | 0.0013 → 0.0063 | 0.224→0.208 |
| discovery-bloom | 0.0041 → 0.0003 | 0.224→0.212 |
| overgrowth | 0.0002 → 0.0006 | 0.184→0.182 |
| autoimmune | 0.0000 → 0.0034 | 0.180→0.179 |
| storm-recovery | 0.0001 → 0.0021 | 0.172→0.159 |
memory→mood CMI vs feedback (calm) — sweet spot at fb 0.6:
E9 · wire memory → motion
Memory also gates cohesion/steering (memoryMotion): it tightens schools structurally (dispersion drops); whether that reaches Hoel CE is in the table.
| scenario · cond | CMI | moodCE | disp |
|---|---|---|---|
| calm off | 0.0013 | 0.224 | 74 |
| mood | 0.0063 | 0.208 | 58 |
| motion | 0.0008 | 0.219 | 58 |
| both | 0.0040 | 0.212 | 68 |
| storm-recovery off | 0.0001 | 0.172 | 115 |
| mood | 0.0021 | 0.159 | 122 |
| motion | 0.0002 | 0.171 | 120 |
| both | 0.0015 | 0.161 | 117 |
| autoimmune off | 0.0000 | 0.180 | 28 |
| mood | 0.0034 | 0.179 | 31 |
| motion | 0.0129 | 0.179 | 37 |
| both | 0.0005 | 0.179 | 24 |
Takeaway: memory is a near-decoupled identity subsystem (E7). Wiring it into mood makes it weakly & tunably informative, with a sweet spot (E8). Wiring it into motion tightens schools structurally (dispersion drops) but the mood Hoel-CE stays flat (E9) — so whether memory is “part of CE” depends on the coupling and on which observable you measure.