Skip to content

reference-skills-reasoning.md

Latest commit

 

History

History
70 lines (49 loc) · 2.68 KB

reference-skills-reasoning.md

File metadata and controls

70 lines (49 loc) · 2.68 KB
Raw

Reference — Reasoning Skill

Defined in src/skills.metta. Backed by two reasoning engines in lib_nal.metta and lib_pln.metta.

metta

Signature

(metta sexpression)

Purpose

Evaluate an arbitrary MeTTa s-expression in the agent's AtomSpace. Primary use is to invoke NAL (|-) or PLN (|~) inference from within the agent loop.

Parameters

  • sexpression — a MeTTa s-expression. Read by sread, evaluated by eval.

Returns

Whatever the inner expression returns. For NAL/PLN calls, this is a conclusion atom paired with an (stv frequency confidence) truth value.

Examples

NAL — deduction:

(metta (|- ((--> (× sam garfield) friend) (stv 1.0 0.9))
           ((--> garfield animal)         (stv 1.0 0.9))))

NAL — implication with a variable (note $1):

(metta (|- ((==> (--> (× $1 elephant) eat) (--> $1 ([] dangerous))) (stv 1.0 0.9))
           ((--> (× tiger elephant) eat)                            (stv 1.0 0.9))))

NAL — revision (same term, two sources): |- merges the evidence.

PLN — forward chaining:

(metta (|~ ((Implication (Inheritance $1 (IntSet Feathered))
                         (Inheritance $1 Bird)) (stv 1.0 0.9))
           ((Inheritance Pingu (IntSet Feathered)) (stv 1.0 0.9))))

Engine selection, stopping criteria, action thresholds

These are policy decisions, not part of the metta skill's API. See reference-orchestration.md for the full tables and rationale (pattern → engine mapping, halt conditions, ACT / HYPOTHESIZE / IGNORE tiers).

Notes / limits

  • Independent variables are written $1, $2, …
  • Negated knowledge uses (stv 0.0 c).
  • metta evaluates any MeTTa expression, not just reasoning calls. Malformed input reports errors through &error on the next turn.
  • Confidence decays ~10% per deduction hop. Chains past 3 hops usually fall below the ACT threshold — see tutorial-08-reliable-reasoning.md.
  • Premise formulation is the primary failure surface. Verify term order, copula, and granularity before trusting a conclusion. See reference-failure-modes.md.

See also