Create a stable experiment identity from seed, dataset IDs, and configuration so a result can be reconstructed later.
Required APIdef reproducibility_key(config, dataset_ids, seed):
return key_stringBehaviorThink through the mechanism first if you want the extra reasoning step. It never blocks the editor.
reproducibility_key({'lr': 0.1, 'width': 4}, ['b', 'a'], 7)equivalent evidence has one identity
reproducibility_key({'lr': 0.2}, ['a'], 7)one changed hyperparameter creates a new run
2 hidden edge tests run after the visible contract passes.
Run Tests to see the contract verdicts here.