---
license: cc-by-4.0
tags:
- synthetic
- optimization
- benchmark
---

# Douvras Evolution Lab Constructions v0.1

Small, synthetic benchmark for the loop **candidate → verifier → score**.
Each problem is an eight-node cycle independent-set toy problem. The verifier
checks node range, uniqueness and absence of selected edges. The published
`best_known_score` is 3; a valid score of 4 is marked `IMPROVED` for this toy
family.

The dataset contains 54 records (36 train, 9 validation, 9 frozen test) across
six relabeled problem instances. Splitting is by `problem_id`, so candidates
from one instance cannot cross train/validation/test. All records are synthetic
and generated locally; no upstream AlphaEvolve notebook or data is included.

This release demonstrates evaluator plumbing, not a new mathematical result.
The graph family is intentionally solved by exhaustive reasoning and the
construction is not evidence of novelty outside this toy benchmark.