A population of bit-string genomes evolves under hidden selection. A naive estimator watches only the allele-frequency trajectories and guesses each locus's fitness effect. Slide the knobs, then click the feasibility map to jump between regimes where inference works and where it quietly fails.
Controls
Any change restarts the run.
Regime presets
Reading the panels
beneficial locusdeleteriousneutral
Colors show the hidden truth for you. The estimator never sees them, only the frequency curves.
Population
Each row is a genome (24 loci), sorted fittest first. Lit cell = allele 1.
Allele-frequency trajectories
The only data the estimator gets. Dashed band = frequency range it trusts (3% to 97%).
Inference vs hidden truth
...
Locus rank ρ-
Genotype order ρ-
Sign correct-
Generation-
true effectestimate, 95% bar covers truthcovers missno usable data
Feasibility map (click a cell)
x y
ρ 0 → 0.8+
Each cell = mean locus rank-correlation over 2 fresh simulations, all other knobs held at the current settings. White ring = where you are now.