Glossary

Terms appear roughly in alphabetical order. Cross-references in italics link to other entries.

alignment

Whether a candidate action / goal is consistent with the agent's values, beliefs, purpose, and goal stack. Returned as an AlignmentReport with a recommendation.

AGM

Alchourrón, Gärdenfors & Makinson 1985 — postulates of belief revision. Cited by BeliefsEngine.add_evidence.

audit

A full_audit() call: a read-only sweep over every layer, surfacing value conflicts, purpose inconsistencies, stale beliefs, and orphan desires/goals.

BDI

Beliefs, Desires, Intentions — Bratman 1987's classical agent architecture. Three of our six layers map directly.

belief

A statement the agent holds with confidence in [0, 1]. Updated by Evidence; decays if not refreshed.

benchmark

Decorator-registered correctness or scenario check. Tier-1 covers per-module correctness; Tier-2 covers cross-layer scenarios; Tier-3 covers system-level decision quality.

confidence

Two related senses: (1) Belief.confidence in [0, 1]; (2) AlignmentReport.confidence in [0, 1] capturing how strongly the framework supports the verdict.

contextual purpose

A Purpose with scope=CONTEXTUAL keyed by a free-form context_key. Looked up via PurposeEngine.get_for_context.

decay

BeliefsEngine.decay_stale lowers the confidence of beliefs whose last_evaluated is older than threshold_days.

desire

An aspirational driver. Bridges purpose to concrete goals; carries an intensity in [0, 1] and an enum-typed intensity_label.

Evaluator

Pluggable strategy that AlignmentEngine uses to score actions. RuleBasedEvaluator (default) is offline; LLMEvaluator is BYO-client.

evidence

A piece of support or challenge for a belief, with type, supports flag, and strength in [0, 1].

goal

A measurable, time-bound objective. Has metric, target, deadline, status (DRAFT/ACTIVE/IN_PROGRESS/BLOCKED/COMPLETED/ABANDONED), and progress.

Ikigai

Mogi 2017 — purpose at the intersection of love, skill, need, and reward. Cites in PurposeEngine.

integrator

The team or person embedding the framework into an autonomous agent system. Their runtime owns task execution; the framework owns motivation.

orphan

A desire whose source_purpose no longer resolves, or a goal whose parent_desire / parent_goal no longer resolves. Surfaced by full_audit.

primary purpose

The active top-level Purpose. set_primary deactivates any prior primary; only one primary is active at a time.

resolution

The output of AlignmentEngine.resolve_conflict — names the chosen layer, discarded layers, recommendation, and rationale per the active precedence.

Schwartz circumplex

Schwartz 2012's circular arrangement of 10 universal values. Adjacent values are compatible; opposing-pole values are tense. Drives ValuesEngine.detect_structural_conflicts.

SDT

Self-Determination Theory — Deci & Ryan 2000. Frames autonomy/competence/relatedness; informs purpose grounding.

sub-goal

A child Goal created via decompose. parent_goal points at the parent; the parent's sub_goals tracks the children.

Storage Protocol

The five-method async contract every backend implements: get / set / list / delete / count.

structural conflict

Two active values from opposing Schwartz poles aligning with the same action. Surfaced as a layer-internal Conflict in AlignmentReport.

tag match

Action proposals carry tags; values match either by category name or by context_tags. Tag matches are weaker signal than aligned_keywords but useful when actions are pre-categorised.