ltms

A logic-based Truth Maintenance System (LTMS) and a pattern-directed reasoning engine in pure Python, after Forbus & de Kleer, Building Problem Solvers (MIT Press, 1993).

The LTMS maintains belief over a set of propositional clauses using Boolean Constraint Propagation (unit propagation), records well-founded support for every derived value, performs dependency-directed backtracking on contradictions, and can explain why anything is believed. A small forward-chaining rule engine sits on top of it.

Documentation site: https://pisanuw.github.io/ltms/ (built from docs/).

New to this? Start with the study companion — a chapter-by-chapter walk-through of the concepts, examples, and exercise solutions. See also PLAN.md for the build plan and STUDY-NOTES.md for the technical digest.

Why

There is no faithful Python LTMS in the wild — JTMS/ATMS have a few toy ports, but the clausal-BCP LTMS with dependency-directed backtracking is the least-ported truth maintenance system outside Lisp/Racket. This is a clean, tested, idiomatic-Python implementation.

What's here

Usage

from ltms import LTRE

e = LTRE()
e.assert_(("or", ("p",), ("q",)))   # p v q
e.assert_(("not", ("p",)))          # ~p
e.is_true(("q",))                   # True  (unit propagation)

e.assume(("rain",), "guess")        # retractable assumption
e.assert_(("implies", ("rain",), ("wet",)))
e.is_true(("wet",))                 # True
e.retract(("rain",), "guess")
e.is_unknown(("wet",))              # True  (belief revised)

World models in files

The world model can live in a .kb data file, separate from Python:

# examples/kb/belief_revision.kb
rain         -> wet ground
sprinkler on -> wet ground
assume rain
expect wet ground true

from ltms.dsl import load_kb_file
load_kb_file("examples/kb/belief_revision.kb")   # runs it; `expect` lines self-check

See examples/ for runnable TRE, LTRE, and dependency-directed-search demos, and exercises/ for worked solutions to the book's chapter exercises (paraphrased statements + original answers/code).

Install (development)

python -m venv .venv && . .venv/bin/activate
pip install -e ".[dev]"      # add ".[dev,sat]" for the PySAT differential tests
pytest

Notes

BCP uses the book's incremental pvs/sats counters (sound but, by design, not logically complete). A watched-literals rewrite and the completeness extension (CLTMS, prime implicates) are tracked as future work in PLAN.md.

Provenance & licensing

Original code in this repository is MIT licensed (see LICENSE).

This is an independent reimplementation of the algorithms in Building Problem Solvers; it does not copy the original Common Lisp source. The book and its reference code are available from the Northwestern Qualitative Reasoning Group at https://www.qrg.northwestern.edu/BPS/readme.html, where the full-text PDF is offered for free download "thanks to the gracious permission of MIT Press" (MIT Press retains print rights). The book PDF and other reference material are kept local and are not redistributed here. See NOTICE for full attribution.