Reproducibility

Evaluation Protocol

A protocol-first framework for comparable, auditable evaluation across time series classification model families.

Overview

Protocol v0.1 preview

The benchmark records dataset-level outputs first and computes aggregate statistics from verified records. Tracks remain distinct to prevent adaptation settings from being conflated.

Standard Evaluation

Use official train/test splits. Each dataset produces one accuracy, and each dataset receives equal weight in aggregate reporting.

Low-shot Evaluation

Use matched class-stratified context sets at fixed label budgets. Report low-shot results separately from Standard scores.

Training-Free / In-Context Learning

Methods infer predictions from labeled context examples without target-specific parameter updates or per-dataset classifier fitting.

Frozen-Probe Evaluation

Keep the pretrained backbone frozen and fit only the standardized lightweight probe on the official training split.

Ranking Rules

  • Each dataset receives equal weight.
  • The primary metric is Average Accuracy.
  • Ranking is descending by Average Accuracy.
  • Only methods completing all required datasets enter the official Overall Leaderboard.
  • Average Rank is reported as a secondary statistic.
  • Low-shot results must not be mixed into the Standard Overall score.
Future Evaluation Workflow

Evaluation pipeline

This visual describes the planned hosted workflow; it is not active in the current static demo.

Model Submission
Validation
Official Evaluation
Dataset Results
Aggregation
Admin Review
Leaderboard

Submission Requirements

Submissions must provide reproducible code, environment requirements, model and checkpoint provenance, a declared license, and a contact for evaluation issues.

Reproducibility

Every future official run will record the code revision, environment image, command, seed, dataset version, dataset-level results, runtime, and peak memory before administrator review.