Module overview

Every public name is importable from the top-level ocpm_engine package (from ocpm_engine import StandaloneEngine, dfg_conformance, ...); the two SQL-compilation modules are imported by module path. Algorithms run in the native Rust extension (ocpm_engine._native) with the GIL released, and the Python layer stays a thin typed facade. Current version: 1.1.0.

Each module page carries the full API and function specs with example usages:

Module	Purpose
ocpm_engine.standalone	File-based analysis: StandaloneEngine over OCEL JSON, XES, SQLite, or DuckDB Parquet, plus serialize_model
ocpm_engine.engine	PostgreSQL compatibility API: OcpmEngine query planning and execution over pg_ocpm
ocpm_engine.analytics	Conformance, prediction, bottleneck, and drift scoring from compact aggregate rows
ocpm_engine.standalone.StandaloneEngine.bottlenecks	Provider-neutral tail latency, synchronization, queue, waiting-cause, drift, spectrum, and blocking analysis
StandaloneEngine.gnn_bottlenecks	Optional built-in CPU graph bottleneck detection plus fit/score artifact APIs
ocpm_engine.gnn	Protocol for separately packaged predictive GNN backends
ocpm_engine.event_batches	Native summarization of factorized pg_ocpm event batches
ocpm_engine.bindings	Native decoding of compact binding-result capsules
ocpm_engine.models	Request contracts, query plans, executions, and capability reports
ocpm_engine.queries	Parameterized SQL for process-mining response shapes
ocpm_engine.dynamic_queries	Compilation of composable dynamic filters into exact DFG queries

A typical standalone flow touches one module; a typical PostgreSQL flow composes four:

from ocpm_engine import (
    EventLogRequest,  # models: request contract
    EventLogWindow,
    OcpmEngine,  # engine: planning and execution
    dfg_conformance,  # analytics: scoring
    TransitionCount,
)

engine = OcpmEngine(dataset_id=42, tenant_id=7)
capabilities = engine.inspect_pg_ocpm(cursor)

request = EventLogRequest(
    object_type="Order",
    windows=(
        EventLogWindow(training_start, training_end),
        EventLogWindow(test_start, test_end),
    ),
)
execution = engine.execute_event_log_summary(cursor, request, capabilities=capabilities)
training, test = execution.summaries  # event_batches: EventLogSummary

test_index = {(e.source, e.target): e.frequency for e in test.dfg}
rows = [
    TransitionCount(
        e.source,
        e.target,
        "directly_follows",
        e.frequency,
        test_index.get((e.source, e.target), 0),
    )
    for e in training.dfg
]
score = dfg_conformance(rows, coverage=0.95)