Decision Parsing

Validate and parse unstructured data by asking a decision model, and judge existing columns with semantic checks. See the decision parsing guide and the tutorial for how to use these objects.

Question Types

A decision model answers typed questions; the question type says what shape the answer has. These are the parser objects passed to ParsedField / ParsedColumn.

Choice	A question that selects one of the column's options.
Score	A question that places the state on the column's ordered scale.
Noul	A yes/no question, answered with a calibrated probability.
Confidence	Reports how certain the model was about another column.

Providers

DecisionProvider	Something that can answer a set of typed questions about a state.
MockProvider	Answers deterministically from a hash of the state.
RecordingProvider	Wraps a real provider and records every answer to a cassette.
ReplayProvider	Serves answers from a cassette; never touches the network.

Configuration

set_provider	Set the provider for the current process.
get_provider	Return the active provider, or None if none is configured.
provider	Use a provider for the duration of a block.
register_provider	Make "<prefix>:<model>" resolve to a provider.
capabilities_of	What a provider can be asked; the shared vocabulary if it does not say.
enabled	Whether decision calls are permitted at all.

Semantic Checks

Holds	Check that a statement holds of each row, as judged by a model.
holds	The predicate behind Holds(), for use inside @dataframe_check.

Caching and Inspection

MemoryCache	Keeps answers for the lifetime of the process.
SQLiteCache	Keeps answers in a SQLite file, so they survive the process.
Plan	What a validation would send, and roughly what it would cost.

questions	Compile a schema's questions without contacting a provider.
plan	Report what validating data would send, without sending it.
stats	Statistics for the decision work that filled a validated frame.

Types and Errors

Decision	One answer, normalized across providers.
ProviderCapabilities	What a provider's model can be asked, as opposed to how fast.
ProviderLimits	Throughput limits a provider publishes, used to pace requests.
DecisionsConfigError	Raised when a decision column is validated with no provider set.
ProviderCapabilityError	Raised when a provider's model cannot be asked a schema's question.

Question is a type alias (a union of the three question kinds) rather than an instantiable class; it appears in the signatures of the objects above.