Post-competition Analysis

This section provides an overview of the two winning solutions and the baselines for both the RL and non-RL tracks.

Competition Paper

Real-World Baselines Challenge: Dynamic Train Rescheduling under Stochastic Perturbations (under review; will be published in the ECML PKDD conference journal)

Leaderboard

Track	Name	Score	Submitted by	Repo
non-RL	v4-sipp-locks	20.8429	darshanmakwana412	ecml2026
non-RL	flatland_v3	12.4925	ge48cel	v5-dispatcher
non-RL	Baseline: flatland-baselines-deadlock-avoidance-heuristic	10.7320	chenkins	flatland-association/flatland-baselines
RL	Baseline: Organizer: adaptive-completion-7d07d93	9.4231	roman.liessner	dynamik1703/ecml2026-starterkit

Evaluation

See eval.md for rewards and normalized rewards definitions.

Sucess Rates

Normalized Rewards

Submission Comparison per Level

Submission Comparison per Scenario

See eval.md for the definition of the reward categories normalized score.

Detailed Plots per Submission

Detailed Overview Plot