Add closed_world predicates

[ColtonPayne]

Summary

Adds support for closed world predicates to PyReason. Predicates registered as closed-world have their unknown [0,1] bounds treated as [0,0] (false) during rule body satisfaction checks, enabling negation-as-failure style reasoning for a user-specified subset of predicates.

Public API

pr.add_closed_world_predicate('hackerControl')

Any clause in a rule body that references hackerControl will, at satisfaction-check time, treat unknown [0,1] bounds — or absence of the label entirely — as [0,0]. Known bounds are left untouched, so setting hackerControl(A) explicitly still behaves normally. pr.reset() clears the registered set; pr.reset_rules() and pr.reset_settings() leave it intact.

Implementation

• New module-level __closed_world_predicates set in pyreason/pyreason.py, plumbed into each Interpretation* class via Program.reason() as a numba.typed.List[Label].
• is_satisfied_node / is_satisfied_edge in interpretation.py, interpretation_fp.py, and interpretation_parallel.py gained a short-circuit branch: if the clause label is closed-world and the world bound is missing or [0,1], it's coerced to [0,0] before the containment check.
• Fixed a bug where Program.reason() assigned closed_world_predicates to Interpretation and InterpretationFP but not InterpretationParallel, causing parallel mode to silently run with an empty closed-world set. Assignment added in pyreason/scripts/program/program.py.

Tests

• Unit — tests/unit/disable_jit/interpretations/test_minimized_predicates.py: JIT-disabled tests covering is_satisfied_node / is_satisfied_edge directly for both interpretation and interpretation_fp modules. Exercises in-world [0,1], missing label, known bounds bypass, negative-query case, multiple closed-world predicates, and non-closed-world contrast.
• API — tests/api_tests/test_pyreason_state_management.py adds TestMinimizedPredicateApi: registration, idempotence, multi-predicate, and interactions with reset() / reset_rules() / reset_settings().
• Functional — tests/functional/test_circumscription.py: 8 scenarios × 3 modes (regular, fp, parallel). Every test depends on a ~closed_world_pred(Y) clause actually firing, so circumscription must be working for the test to pass (not just present). Covers:
  • Known [1,1] bypass
  • Closed-world vs non-closed-world contrast
  • Multi-timestep cascade (A → B → C, [0,1] at each step)
  • inconsistency_check, persistent, atom_trace settings
  • Two closed-world predicates exercising both branches (explicit [0,1] and missing-from-world) in the same reasoning pass
  • End-to-end example scenario

Test plan

• pytest tests/unit/disable_jit/interpretations/test_minimized_predicates.py -v
• pytest tests/api_tests/test_pyreason_state_management.py -v
• pytest tests/functional/test_circumscription.py -v (all cases green on regular + fp + parallel)
• python examples/circumscription_ex.py runs without error

🤖 Generated with Claude Code

[kmukherji]

Unless there's documentation on another file, this script needs some comments for an user to understand what's going on.

[ColtonPayne]

Updated example with comments