Public API reference

This page is generated from decishift.__all__ by python scripts/generate_api_reference.py. It is an inventory of the stable top-level import surface, not a replacement for the conceptual guides.

Symbol	Kind	Defined in
`ComponentIdentity`	class	`decishift.core.identity`
`DecisionPipeline`	class	`decishift.core.pipeline`
`PipelineTrace`	class	`decishift.core.pipeline`
`ComparisonResult`	class	`decishift.diff.compare`
`compare_pipelines`	function	`decishift.diff.compare`
`compare_predictions`	function	`decishift.diff.compare`
`DecisionFlow`	class	`decishift.flow.flow`
`DecisionNode`	class	`decishift.flow.node`
`DecisionOutput`	class	`decishift.flow.trace`
`FlowComparisonResult`	class	`decishift.flow.compare`
`FlowTrace`	class	`decishift.flow.trace`
`compare_flows`	function	`decishift.flow.compare`
`__version__`	value	`decishift.version`

Entry points

Use DecisionFlow, DecisionNode, and compare_flows for structured row-aligned decision DAGs and categorical actions.
Use DecisionPipeline and compare_pipelines for the backwards-compatible fixed binary pipeline path.
See Decision flows, Flow attribution, and Failure modes for semantics and trust boundaries.

Minimal examples

These examples use local, deterministic components and the CPU-only pandas and numpy dependencies.

Compare fixed binary decisions

import pandas as pd

from decishift import DecisionPipeline, compare_pipelines

records = pd.DataFrame({"record_id": ["a", "b", "c"], "score": [0.2, 0.6, 0.9]})

def score(frame):
    return frame["score"].to_numpy()

baseline = DecisionPipeline(model=score, threshold=0.5, versions={"model": "score-v1"})
candidate = DecisionPipeline(model=score, threshold=0.7, versions={"model": "score-v1"})
comparison = compare_pipelines(baseline, candidate, records, id_column="record_id")

print(comparison.summary()["changed_decisions"])  # 1

Compare categorical decision flows

import numpy as np
import pandas as pd

from decishift import DecisionFlow, DecisionNode, compare_flows

records = pd.DataFrame({"record_id": ["a", "b", "c"], "score": [0.2, 0.6, 0.9]})

def score(frame):
    return frame["score"].to_numpy()

def route_at(threshold):
    def route(_frame, inputs):
        return np.where(inputs["score"] >= threshold, "approve", "review")
    return route

baseline = DecisionFlow(
    nodes=(
        DecisionNode("score", score, version="score-v1"),
        DecisionNode("route", route_at(0.5), depends_on=("score",), version="route-v1"),
    ),
    final_node="route",
)
candidate = DecisionFlow(
    nodes=(
        DecisionNode("score", score, version="score-v1"),
        DecisionNode("route", route_at(0.7), depends_on=("score",), version="route-v2"),
    ),
    final_node="route",
)
comparison = compare_flows(baseline, candidate, records, id_column="record_id")

print(comparison.summary()["changed_actions"])  # 1