Skip to main content

Crate fp_runtime

Crate fp_runtime 

Source
Expand description

Runtime policy + decision-recording layer for frankenpandas.

Pandas operations frequently hit “do we accept this input or fail closed?” decisions: a dtype that doesn’t quite match, a frequency that’s almost-but-not-quite regular, an alignment that produces NaNs the user maybe didn’t expect. fp-runtime gives the rest of the workspace a single place to record those decisions, score their compatibility, and persist a verifiable evidence trail so pipelines can audit “why did the IO layer / groupby / merge make this choice on this input?” after the fact.

§Decision recording

§Conformal prediction guards

  • ConformalGuard: rolling-window nonconformity calibration used to gate uncertain decisions inside hot-path code (e.g. “this dtype inference is too unsure — fail closed”).
  • ConformalPredictionSet: the calibrated prediction set (inclusion / exclusion of candidate labels) the guard produces.

§RaptorQ envelopes

§Error reporting

§Cargo features

  • asupersync (off by default): enables the asupersync submodule and the outcome_to_action helper for converting an asupersync::Outcome into a DecisionAction. Pulls in the asupersync crate as an optional dep. (Items are gated behind the feature so they don’t appear in the default docs.rs render.)

Modules§

asupersync

Structs§

CompatibilityIssue
ConformalGuard
Calibration window for conformal guard.
ConformalPredictionSet
Conformal prediction set: which actions are admissible at significance level alpha.
DecisionMetrics
DecisionRecord
DecodeProof
EvidenceLedger
EvidenceTerm
GalaxyBrainCard
LossMatrix
RaptorQEnvelope
RaptorQMetadata
RuntimePolicy
ScrubStatus
SemanticFingerprintBuilder
SemanticIndexIdentity
SemanticWitnessRecord

Enums§

DecisionAction
IssueKind
RuntimeError
RuntimeMode

Constants§

DEFAULT_RAPTORQ_SYMBOL_BYTES
MAX_DECODE_PROOFS

Functions§

decision_to_card
outcome_to_action
semantic_fingerprint_bytes