Crate proofframe 

Native ProofFrame engine for Arrow-backed data contracts.

The crate exposes a Rust-native API by default. The Python extension module is available behind the python feature and is enabled by the PyPI build configuration. Core invariants are stable enough for the stable 0.5 line: versioned canonical dataset fingerprints, compiled relational and dataset contracts, disk-backed exact state, deterministic partitions, and signed evidence.

Modules

evidence

Strict, versioned evidence envelopes binding data, execution and result claims.

receipt

Ed25519-signed proof receipts over canonical JSON reports.

Structs

AssertionAst

CancellationToken

Cooperative cancellation shared across cloned operation handles.

ChangedRow

A row present in both datasets whose values changed.

ColumnContract

Rule set applied to a single column during validation.

ColumnPlan

One schema-resolved column in the execution intermediate representation.

ColumnProfile

Per-column summary produced while profiling a dataset.

CompareAst

ComparePlan

CompiledContract

Contract execution IR compiled in Arrow schema order.

CompiledRules

Semantically checked rules ready for typed execution.

CompositeUniqueAst

CompositeUniquePlan

Contract

Validation contract: per-column rules plus a bound on emitted findings.

ContractAst

Strict, versioned syntax tree parsed from a ProofFrame contract document.

ContractAstV2

CountPlan

CountRangeAst

DatasetPlan

DatasetRulesAst

DiffMetrics

DiffOptions

DiffReport

Exact keyed diff between two datasets.

DuplicateSample

ExactMetrics

ExactState

ExactSummary

ExecutionMetrics

Accounted engine state for a compiled validation execution.

ExecutionOptions

Execution hints that do not alter contract semantics.

FastValidationReport

Rules-only validation result that skips profiling and fingerprinting.

Finding

A single rule violation with bounded, row-level evidence.

Fingerprint

Typed canonical digest with its algorithm version and row count.

FingerprintOptions

Inputs that affect canonical fingerprint semantics.

LeakageOptions

LeakageReport

Train/test overlap result that exposes only hashed sample identifiers.

MemoryReservation

RAII charge for engine-owned memory.

PiiFinding

A single PII detection that never carries the matched value.

PiiFingerprintOptions

Secret-bearing scan configuration. The key is deliberately never serializable.

PiiReport

Aggregated PII scan result with bounded per-cell findings.

Profile

Deterministic dataset profile with a canonical content fingerprint.

RatioPlan

RatioRangeAst

ResourceAccount

A node in a resource hierarchy. Reservations charge this node and every ancestor.

ResourceLimits

Hard limits shared by validation, distinct, diff and evidence operations.

RowPlan

RowRuleAst

RuleAst

Syntax-level rule set for one named column.

RuleAstV2

TempReservation

RAII charge for operation-owned temporary storage.

ValidationReport

Full validation result: findings plus the dataset profile and fingerprint.

Enums

BoundAst

Exact source literal for a numeric, decimal, or timestamp bound.

CompareOpAst

CompositeNullPolicyAst

ContractDocument

ContractVersion

Version of the serialized contract language.

DiffOutput

DistinctMode

Exact distinct counting can dominate large profiles, so callers can opt out.

ErrorCode

Stable machine-readable category for every ProofFrame failure.

FingerprintVersion

Canonical fingerprint algorithm selected for an operation.

KernelKind

Arrow-specialized kernel selected once during contract compilation.

NaNPolicy

Runtime NaN behavior compiled for floating-point kernels.

NaNPolicyAst

Source-level floating-point NaN policy.

NullPolicyAst

OperandAst

OperandPlan

ParameterizedTypeAst

PiiFingerprintMode

Linkability policy for redacted PII value fingerprints.

PrimitiveTypeAst

ProofFrameError

Errors returned by ProofFrame profiling, validation, diffing, and receipts.

RowPlanKind

ScalarValuePlan

SpillPolicy

Policy for exact-state overflow during a keyed diff.

TimeUnitAst

TypeAst

TypedBound

Exact semantic value used by compiled range checks.

ValueKind

ValueRef

Functions

check_partition_readers

Validate ordered Arrow partitions with bounded native workers.

check_partition_readers_with_evidence

Validate ordered partitions and bind each partition fingerprint into a manifest.

detect_leakage_readers

Detect train/test row or key leakage between two Arrow readers.

detect_leakage_with_options

diff_readers

Compute an exact keyed diff between two Arrow readers.

diff_readers_with_options

execute_reader

Execute a compiled contract without reparsing or performing rule-map lookups.

execute_reader_with_fingerprint

Execute validation and compute the V2 dataset fingerprint from the same batch traversal.

fingerprint_reader

Return only the canonical dataset fingerprint, without profiling or exact distinct state.

fingerprint_reader_with_options

Return a typed canonical fingerprint using an explicitly selected version.

profile_reader

Profile Arrow record batches and return typed Rust metadata.

profile_reader_with_distinct

Profile Arrow record batches with configurable exact distinct counting.

profile_reader_with_resources

Profile with a hard resource budget for optional exact distinct state.

profile_reader_with_resources_and_hint

Profile with hard resource budgets and an exact, non-consuming row-count hint.

scan_pii_reader

Scan Arrow record batches for high-signal PII patterns.

scan_pii_reader_with_options

Scan for PII using an explicit linkability and key-management policy.

validate_fast_reader

Validate Arrow record batches with the rules-only fast path.

validate_reader

Validate Arrow record batches with the full profiling path.

Type Aliases

PartitionReader

One independently consumable Arrow stream in a logical dataset partition.