Results — evolution track (E)

E-track worlds add heredity to the ecology engine: a genotype field that new tissue copies from its parent (with mutation), making natural selection possible as world physics rather than as a scripted event. The world guide's evolution page shows films of the gene pool evolving. Scoring.

E0/E1 — the evolution track (corrected physics, v0.9.0)

One species + resource + a heritable trait field: new tissue copies its parent tissue's genotype g (with mutation at growth sites), and g sets each cell's consumption/robustness through a fixed linear "biochemistry" map. Nothing selects explicitly — yet sustained scarcity shifts the population's genotype toward frugality and the population re-expands at the adapted genotype: natural selection, emergent, with genetic path dependence (a second famine meets a different population than the first).

Correction (2026-02-15): the first E0 release used blending inheritance (variance-destroying; Jenkin's 1867 objection, live) and its "adaptation" was mostly location-luck survivorship. The physics is fixed (particulate copy inheritance — variance persists); blending-era results are wiped. E1 (storm world): the world's own famine cycles drive real, verified evolution — populations arrive already adapted to their climate (mean genotype ~0.2 vs founder 0.5) and de-/re-adapt under port-driven eras.

Legacy point-scored run (2026-02-15): gpt-5.2 0.57/0.55, fable 0.42 and one 0.00 — 112 productive turns, then it ended its session without ever answering (unanswered = 0, as the rules state; kept as scored). In the CRPS re-run fable answered both seeds. Oracle gap 0.22–0.43: still hard-tier-adjacent. fable senses the storm clock ("regime", "period") but no agent has discovered adaptation — the storm oracle's edge is exactly the storm-phase calibration nobody builds.

• Fourth ontology rung: no agent workspace contains any evolutionary vocabulary (adapt/select/mutate/trait). Budget use 1–5%: era-scale physics (4–7k-tick selection eras) far exceeds what agents explore voluntarily — no agent ran a single selection era of its own.
• Cross-level map finding: the linear micro genotype→phenotype map aggregates to a saturating macro response (selection depletes trait variance — the breeder's equation runs out of fuel). Micro shape ≠ macro shape; now tracked per rung.

E2 — enzyme economics (2026-02-16, v0.11.0): trade-offs as theorems

The deepest world by design philosophy. The only authored biology is a price list: tissue earns energy ∝ gene·substrate, pays rent ∝ gene, stores at most a finite larder, burns itself when bankrupt, dies below a floor. No trade-off curves, no per-genotype death law. The critical resource level R* = rent/earnings, the famine/plenty selection asymmetry, and the whole trade-off structure are computed by the world from the conservation law. Parameters were found by search over price-space (theory coordinates; 4/48 candidates passed the battery: survival, bidirectional selection, variance maintenance, R*-straddling weather), certified 20/20 seeds. The gene pool visibly breathes with the ~1000-tick storm clock — film + histogram timeline. Unlike E1, the selection clock now fits INSIDE contract horizons (S3/S4 are era-scale).

Oracle gap 0.13–0.26: mid-tier, and a finding in itself — E2's fast selection clock (~1000-tick eras, chosen so eras fit inside contracts) makes the world easier than E1: fable's notes model the storm cycle as a bistable A/B state machine ("FLIPS B→A", rise/release timescales per state) and interpolate it well without any evolutionary ontology. The gene pool is still invisible — no heredity/selection/mutation vocabulary in any workspace (the "adaptation" in fable's notes is sensory undershoot, not biology) — but a good enough climate model prices most contracts. Slow clocks (E1) punish ignorance harder than fast ones.

What's next on the evolution track

• E2 frontier validation (running) — the oracle gap will tell whether conservation-law worlds are frontier-hard;
• evolutionary preparation contracts (breed a target genotype);
• n≥5 seed statistics on E1/E2.

Generated by python -m physim.results; numbers from REPORT.md (the running lab log with full methodology).