ClosedCode

ML Collaboration Workflow

Design how ML contributors work together on datasets, models and API packaging. Each step is scored and paid live from a draining pool. Reusing someone else's dataset or model sends a share of the points back to its original author.

Read the brief

1 participant · own branch each

Why this challenge exists

ML work comes in independent artifacts (a dataset, a model, its training code, an API packaging). Contributors often build on each other's work. ML challenges have no tasks: contributors submit directly from the ML workspace.

Who hosts this challenge

MyTwin Lab

The brief

Objective

Each step is scored as it is submitted and paid live from a finite pool, in arrival order:

• Dataset: AI-scored against the dataset grid

• Model metric: scored from the reported AUC/F1/accuracy above a baseline, with no AI call

• Model code and API packaging: AI-scored against the code grid

Expected result

• Overtaking the best model earns a lead bonus, never revoked

• Reusing another contributor's dataset or model redirects a share of the points to its original author, while the reuser always keeps a protected minimum share

• Every award or deduction is an append-only ledger row, and the breakdown is visible per contribution

• Awards are clamped to the remaining pool