CVS

informative

Competitive Viability Selection

2026-03-05 L1 level paper2 100 runs

new

premortem

mock

real

metrics

analyze

review

krit

done

In plain language

We tried "natural selection" for text fragments — those that copy better survive. It turned out the only working mechanism is literal word matching (template matching). Neither text length nor "complexity" helps survival. Only exact repetition.

What we found

Eigen-style viability selection does not significantly amplify copy rate. CVS enrichment: 14% vs 10% baseline (p=0.257,

Eigen-style viability selection does not significantly amplify copy rate. CVS enrichment: 14% vs 10% baseline (p=0.257, not significant). Copy fidelity Q ≈ 0.000004 — far below Eigen error threshold. Prompt wording has 3× more effect on copy rate than fragment selection. The substrate's "physics" is dominated by the LLM interpreter, not by fragment properties.

Predictions we made before running

2/4 confirmed · 2 refuted

▶ Technical details

Research hypothesis

Selection for high-template-score fragments increases copy rate in the resulting population. The text substrate supports cumulative viability selection: enriching for "copyable" fragments amplifies the fraction of copy-producing reactions.

Experimental setup

Type: selection_test

Condition	Parameters
A1_full	label: Full pool (control), n_fragments: 40, n_reactions: 100
A2_cvs	label: CVS-selected (top 20 by template score), n_fragments: 20, n_reactions: 100
A3_anti	label: Anti-selected (bottom 20), n_fragments: 20, n_reactions: 100
A4_random	label: Random 20 (3 draws), n_fragments: 20, n_reactions: 300

Parameters

pool_size: 40
n_runs_per_condition: 100
model: gemini-2.0-flash
temperature: 1

chemistry selection copy-rate eigen