Causality

The causality solver answers questions of the form “was this the cause of that?” over an argumentation theory. It is a separate module, built on top of the core engine, and it adds a small set of Prolog predicates implementing counterfactual and NESS causal tests.

It is available as the Maven module causality-solver. It is also bundled in the API of Arg2P (arg2p, the npm package @tuprolog/arg2p, the Python package arg2p): every solver of the API has it.

From the API

Every solver of the API of Arg2P tests causality on its theory, through its causality, one method for each predicate below, the statements given as Prolog text:

import it.unibo.tuprolog.argumentation.Arg2p

val causality = Arg2p.of("r_0 : exposure => disease.\nf_1 :=> exposure.").causality
causality.ness("exposure", "disease")      // the intervention, [-exposure]: null if it is not a NESS cause
causality.butFor("exposure", "disease")    // true

The same methods are in JavaScript and, in snake case (but_for, ness_original, …), in Python. Each test starts from a reset context, as the others before it left nothing behind.

Getting a causality solver of 2P-Kt

From Kotlin, a solver of 2P-Kt with the causality module is the causality extension on Arg2pSolverFactory:

import it.unibo.tuprolog.argumentation.causality.causality
import it.unibo.tuprolog.argumentation.core.Arg2pSolverFactory

val solver = Arg2pSolverFactory.causality(
    """
    r_0 : exposure => disease.
    f_1 :=> exposure.
    """.trimIndent(),
)

It returns a ready-to-use 2P-Kt MutableSolver with the causality module loaded.

Predicates

All predicates live in the causality module, and are called through the causality:: prefix. A solver of Arg2pSolverFactory.causality() also has them without the prefix: the module is one of its static libraries.

ness_original/3 and ness_original_intervention/2 implement the model of Pisano, Prakken, Sartor and Liepina (ICAIL 2025); ness/3 and ness_intervention/2 implement a revision to appear at ICAIL 2026. See References.

ness/3 and ness_original/3 raise a type error if their first argument is already bound. Always pass a fresh variable and read the intervention back from the solution.

Running a query

A causality query is an ordinary goal: the module evaluates the theory of the solver by itself, so the query needs no reset of the evaluation context first.

import it.unibo.tuprolog.core.Struct

val solution = solver.solve(Struct.parse("ness(X, exposure, disease)")).first()
println(solution.isYes)

With the DSL the same query reads:

import it.unibo.tuprolog.argumentation.core.dsl.arg2pScope

arg2pScope {
    solver.solve("ness"(X, "exposure", "disease")).first()
}

Through the API, the predicates are goals as well: Arg2p.of(theory).solve("causality::ness(X, a, b)").

A worked example

Two independent causes, each sufficient on its own — the classic case where the but-for test fails but NESS succeeds:

f_1 :=> fire_a.
f_2 :=> fire_b.
r_1 : fire_a => house_burns.
r_2 : fire_b => house_burns.

Try it — the playground bundles the causality module, so the query runs there as it does on the JVM.

but_for(fire_a, house_burns) fails: removing fire_a leaves fire_b, and the house burns anyway. ness(X, fire_a, house_burns) succeeds and binds X to the intervention it applied, here [- fire_a]: the intervention negates the candidate cause, and the test reports the set under which that cause turns out to be necessary for the effect.

Notes on behaviour

• The causality solver runs its own inner evaluation with grounded semantics, whatever argumentLabellingMode says. The other flags are those of the outer solver, whether its settings or the theory state them (as graphExtension, prologStrictCompatibility or metaConflicts).
• ness/3 and but_for/2 only name accepted causes: a statement no argument concludes (the absence -d of a fact d never stated, say) or a rule identifier is not a cause for them, though ness_original/3 may report it as one.
• Graphs computed during a causality query are cached, so repeated queries against the same theory and effect are considerably faster than the first one.