proact-py 0.2.0

Active logic meets Prolog — three-valued reasoning engine

Proact

pip install proact-py

Prolog like language for decision making and parallel reasoning under uncertainty. Proact adds a continuation state to Prolog's truth values, signalling uncertainty.

Every goal returns a status (3 state value). Conjunction is a sequence. Disjunction is a selector. Backtracking only triggers on failure — cont and done cut the search. Between ticks, queries returning cont are automatically re-evaluated while the database persists, giving you a stateless tick loop for free.

Applications:

• Reasoning under uncertainty. The three values map onto epistemic states: verified, undetermined, refuted. A conjunction of checks naturally implements iterative deepening — work focuses on the earliest unresolved premise. Unary operators give you cognitive control: demote means "don't commit yet" (done→cont); promote means "don't give up yet" (fail→cont). This is relevant to validation layers for LLM reasoning, where forcing every check into true/false is lossy compression of the actual epistemic state.

• Behavior modeling. Reactive agents, planners, robot controllers — anything where logic meets time. Prolog gives you unification, backtracking, and pattern matching. Active logic adds temporal flow. You get behavior trees with the expressiveness of logic programming, without the trees.

Python integration

from proact import Proact

p = Proact()
p.add("mortal(X) :- human(X). human(socrates).")
print(p.query("mortal(socrates)"))  # Result(status='done')

Using Proact in terminal

./proact                          # interactive mode
./proact file.pa                  # load file, then interactive mode
./proact file.pa -e "goal."      # evaluate goal and exit
./proact --max-ticks 100 -e "g." # limit ticks; run until resolved or budget exhausted

Tutorials

• Reasoning under uncertainty — from predicate logic to three-valued epistemics
• Behavior modeling — behavior trees via logic programming

The Transform

Conjunction (,) — Active logic AND / Sequence

         fail  cont  done
  fail   fail  fail  fail
  cont   cont  cont  cont
  done   fail  cont  done

If the left goal is cont, the whole conjunction is cont — don't evaluate the right side yet.

Disjunction (;) — Active logic OR / Selector

         fail  cont  done
  fail   fail  cont  done
  cont   cont  cont  cont
  done   done  done  done

If the left goal is done or cont, don't try alternatives.

Clause resolution

Multiple clauses for the same predicate provide choice points. Backtracking occurs only on fail — if a clause body returns cont or done, stop searching.

Tick loop

Queries returning cont are automatically re-evaluated on the next tick with a fresh environment. The database (asserted facts) persists between ticks.

Built-ins

Predicate	Status	Description
done / true	done	Always succeeds
fail / false	fail	Always fails
cont	cont	Always continues
inv(G) / \+ G	swap done/fail, keep cont	Inverter
condone(G)	fail→done	Forgive failure
promote(G)	fail→cont, cont→done	Optimistic
demote(G)	done→cont, cont→fail	Pessimistic
par_any(A,B)	max(A,B)	Lenient parallel
par_all(A,B)	min(A,B)	Strict parallel
tick(N)	done	Unify N with current tick

Plus standard Prolog: write/1, writeln/1, nl, assert/1, retract/1, is/2, =/2, \=/2, </2, >/2, >=/2, =</2, call/1, consult/1.

Examples

Wait N ticks, then succeed

wait(N) :- tick(T), T >= N.
wait(N) :- tick(T), T < N, cont.

?- wait(3), writeln(hello).
[tick 0] cont
[tick 1] cont
[tick 2] cont
hello
done.

Counter (stateful, using assert/retract)

:- assert(count(0)).

count_to(Target) :- count(N), N >= Target.
count_to(Target) :-
    count(N), N < Target,
    retract(count(N)), N1 is N + 1, assert(count(N1)),
    cont.

Reactive behavior (condition-gated sequence)

:- assert(agent_pos(0)).

move_to(X) :- agent_pos(P), P =:= X.
move_to(X) :- agent_pos(P), P < X,
    retract(agent_pos(P)), P1 is P + 1, assert(agent_pos(P1)), cont.

patrol :-
    agent_pos(P), P < 3, move_to(3), cont.
patrol :-
    agent_pos(P), P >= 3, P < 5, move_to(5), cont.
patrol :-
    agent_pos(P), P >= 5.

Reasoning under uncertainty

% reason.pa — three-valued validation for multi-step arguments

:- assert(known(premise_a)).
:- assert(investigation(premise_b, 2)).
:- assert(investigation(premise_c, 1)).

check(P) :- known(P).
check(P) :- known_false(P), fail.
check(P) :-
    investigation(P, N), N > 0,
    retract(investigation(P, N)), N1 is N - 1,
    (N1 > 0 -> assert(investigation(P, N1)) ; assert(known(P))),
    cont.

argument :- check(premise_a), check(premise_b), check(premise_c).

The conjunction is the search strategy — it focuses work on the earliest unresolved premise and doesn't waste computation on steps whose preconditions haven't been met:

[tick 0] a=done, b=cont          → cont  (c never reached)
[tick 1] a=done, b=done, c=cont  → cont  (work shifts to c)
[tick 2] a=done, b=done, c=done  → done

The epistemic operators give cognitive control over the validation:

cautious(X)       :- demote(check(X)).    % "verified but needs review"  (done→cont)
keep_exploring(X) :- promote(check(X)).   % "refuted but try harder"     (fail→cont)
viable_theory     :- par_any(hyp_a, hyp_b). % first hypothesis verified wins

Relation to active-logic

Proact implements the core active logic operators:

C# (activelogic-cs)	Proact	BT equivalent
status && status	Goal, Goal	Sequence
status || status	Goal ; Goal	Selector
+ (lenient)	par_any(A, B)	Parallel (any)
* (strict)	par_all(A, B)	Parallel (all)
!status	inv(G)	Inverter
~status	condone(G)	Condone
+status (unary)	promote(G)	Promoter
-status (unary)	demote(G)	Demoter
cont	cont	Running