One verified slow-training rollout with token and reward provenance.
Summary
Types
@type status() :: :available | :claimed | :complete | :failed
@type t() :: %Imp.Training.FastSlow.Rollout{ behavior_logprobs: [number()], behavior_policy_id: String.t(), claim_id: String.t() | nil, cycle: non_neg_integer(), dataset_indices: [non_neg_integer()], failure: Imp.Training.FastSlow.Config.json_value() | nil, generated_at_step: non_neg_integer(), group_id: String.t(), group_size: pos_integer(), id: String.t(), identity_digest: String.t(), input_digest: String.t(), member_index: non_neg_integer(), metrics: Imp.Training.FastSlow.Config.json_value(), output: Imp.Training.FastSlow.Config.json_value() | nil, output_digest: String.t() | nil, problem_id: String.t(), prompt_digest: String.t(), prompt_index: non_neg_integer(), prompt_revision: non_neg_integer(), response_mask: [0 | 1], response_token_ids: [non_neg_integer()], sampling_config_digest: String.t(), score: number() | nil, source: :live | :gepa_cache, status: status(), theta_id: String.t() }