prediction_batch_service

Pure positional planning for prediction engine batches.

The planner is deliberately free of database, cache, model, and progress side effects. Both the request-time ReconXKG preflight and the worker use these objects, so a cache hit is proved against the exact units that execution will consume.

Module Contents

class prediction_batch_service.SequenceBatchPlan

Position-safe sequence handling shared by all selected targets.

class prediction_batch_service.TargetBatchPlan

One method/target engine call after input-schema adaptation.

prediction_batch_service.build_target_batch_plan(descriptor: Any, target: str, dataframe: pandas.DataFrame, sequence_plan: SequenceBatchPlan) → TargetBatchPlan

Build the exact positional engine batch for one selected target.