mcdakit

Multi-criteria decision analysis whose rankings hold.

Most MCDA libraries answer what is the ranking? mcdakit also answers would the ranking survive someone disagreeing with my inputs? Seventeen ranking methods, one of them reversal-free by construction, and numpy as the only runtime dependency.

Guides

• Quickstart
  • Describing a decision
  • Ranking
  • Asking how stable it is
  • Comparing methods
  • Weights from pairwise comparison
• Rank reversal and stability
  • The problem
  • Why it happens
  • SPOTIS
  • The guarantee needs bounds
  • Checking any method
• Language-model assistance
  • Connecting a model
  • Proposing criteria
  • Proposing weights, and checking them against the data
  • Pairwise comparison, which can be checked
  • Simulating a panel
  • Writing it up, with the figures checked
  • What the module will not do
  • A complete run
• Extending mcdakit
  • A method in ten lines
  • The one rule: higher must be better
  • What score receives
  • Oriented or raw data
  • Reporting caveats
  • Refusing a problem
  • Accepting options
  • Shipping a method as a package
  • Inspecting the registry
  • Replacing a built-in
  • Checking your method
  • Extending beyond methods
  • Contributing a method upstream

Reference

• API reference
  • Core types
  • Ranking
  • Language-model assistance
  • Explaining a result
  • Group decisions
  • Sensitivity analysis
  • Weighting
  • AHP
  • The extension contract
  • The registry
  • Conformance testing
  • Methods
  • Normalisation
  • Constants
  • Orientation

Project

• Limitations
• Contributing
• Architecture decisions
• Changelog
• Citing mcdakit

Install

pip install mcdakit

At a glance

from mcdakit import Criterion, rank, sensitivity

criteria = [
    Criterion("Price",     weight=0.40, direction="cost",    bounds=(2.00, 4.00)),
    Criterion("Quality",   weight=0.25, direction="benefit", bounds=(0, 10)),
    Criterion("Lead time", weight=0.20, direction="cost",    bounds=(5, 35)),
    Criterion("Support",   weight=0.15, direction="benefit", bounds=(0, 10)),
]
matrix = [
    [2.75, 7.0, 14, 8.0],
    [2.90, 8.5, 16, 8.0],
    [3.40, 9.0, 11, 7.0],
    [2.20, 3.0, 32, 2.0],
]
labels = ["Option 1", "Option 2", "Option 3", "Option 4"]

result = rank(matrix, criteria, method="spotis", labels=labels)
result.winner                 # 'Option 1'
sensitivity(result)["level"]  # 'fragile'

Where to start

If you want to	Read
Rank some alternatives	Quickstart
Know whether the answer is trustworthy	Stability
Have a model help set the problem up	AI assistance
Add your own method	Extending
Look up a function	API reference
Know why it is built this way	Design decisions
Know what it does not do	Limitations
Cite it	Citing

What is here

Seventeen ranking methods — TOPSIS, VIKOR, ELECTRE I, PROMETHEE II with all six preference shapes, SPOTIS, COPRAS, WASPAS, WPM, ARAS, COCOSO, CODAS, EDAS, MABAC, MARCOS, SAW and two scoring baselines — plus full AHP, which ranks from pairwise comparisons with no decision matrix at all.

Answers about the answer. How far each weight can move before the winner changes, solved for rather than sampled. Whether removing a rejected alternative reorders the rest. Which criterion decided a close margin, and whether reversing it alone would reverse the outcome.

Several stakeholders, aggregated two ways, with the disagreement between them reported rather than averaged away.

Optional language-model assistance, where the model proposes and the package checks: suggestions are parsed into validated types, proposed weights are put through the sensitivity analysis, pairwise judgements are tested for self-contradiction, and every number in a generated write-up is compared against the computed figures.

A documented extension contract. A method written in your own package is a first-class citizen — reachable by name, included in cross-method comparison, analysable by the sensitivity machinery — and a conformance suite you run before publishing it.

Runnable scripts are in examples/, including one for the AI workflow that needs no API key.

Indices

• Index

• Module Index