Glossary

The language of Thoughtware

Concise definitions of the principal concepts. Each term links into the concept web where it lives.

Abstention: A cognitive unit reporting that it does not have enough information to decide. A legitimate result rather than an error, returned rather than thrown.
Agent: A goal-oriented module that composes cognitive units, Skills, deterministic capabilities, memory, and tools through working state, strategy, policies, authority, and stopping conditions. An agent owns a goal. It is not an artificial person.
Agent loop: The repeated movement through planning, action, observation, review, and revision by which an agent pursues a goal through persistent working state.
Authority: The limit within which a system may decide and act. Authority is granted. It is not produced by capability.
Bridging: The deliberate process of filling a named information or capability gap through retrieval, search, tools, another cognitive capability, or a targeted human question.
Closed decision: A decision where competent people, given the same information, would agree, and where the rule they are following can be written down. Closed decisions belong in code.
Cognitive judge: A specialised cognitive unit that evaluates a candidate result against an explicit rubric and returns criterion-level assessments, evidence, confidence, and possible repairs. A judge is evidence, never truth.
Cognitive Posture: The stable stance a Thoughtware system takes toward its role, knowledge, uncertainty, initiative, authority, and relationship with people. Capability determines what a system can do. Posture determines how it conducts itself.
Cognitive unit: A bounded and reusable component that owns one judgment through a named goal, declared inputs and context, structured outputs, a versioned implementation, and an evaluation suite.
Collaboration Contract: The explicit allocation of initiative, approval, challenge, correction, visibility, and control between people and Thoughtware.
Contestability: The ability to inspect, challenge, correct, reverse, or escalate a system’s memory, judgment, action, or working state.
Decision locality: The ability to identify where a meaningful judgment is made, what information influences it, and which component is responsible for its quality.
Determinacy ceiling: The human agreement rate for a class of decision, which bounds the accuracy any implementation can meaningfully claim.
Deterministic code: The exact and auditable layer responsible for arithmetic, schemas, validation, permissions, persistence, transactions, retries, idempotency, audit trails, tool execution, and hard policy gates.
Endorsed knowledge: Information the system is permitted to treat as believed, carrying provenance, scope, validity, and an approval history. Retrieval alone does not create knowledge.
Evaluation memory: Regression cases, scored examples, failure explanations, calibration sets, and behavioural benchmarks retained so that change can be distinguished from improvement.
Experience: Records of what happened during previous runs, including trajectories, corrections, and outcomes. Experience records what happened.
Expertise: Evaluated strategies that tend to work for recognisable classes of cases. Expertise records what tends to work.
Fast path: An implementation or trajectory chosen for speed and cost once its terrain is established, always paired with a guard and a slower alternative.
Functional cognitive work: Bounded work involving interpretation, comparison, abstraction, planning, judgment, criticism, explanation, or revision, where the result cannot be produced through an exact procedure alone. Makes no claim about inner experience.
Instructional Programming: The practice of constructing durable cognitive systems through specifications of judgment, behaviour, memory, authority, evaluation, and learning.
Judgment Chain: The sequence of interpretations, decisions, checks, and actions required to move from a situation to an outcome. Links may belong to code, a cognitive unit, a Skill, an agent, a person, or shared judgment.
Judgment Leadership Curve: A model of how leadership moves between people and Thoughtware as pattern density, knowledge, novelty, consequence, and reversibility change. It is a model of responsibility, not a ranking of intelligence.
Judgment Terrain: The conditions around a decision, including pattern density, knowledge coverage, novelty, consequence, reversibility, and evaluability.
Loop compression: The reduction of a trajectory through better framing, selective memory, expertise, reusable subplans, local repair, deterministic promotion, faster units, and stronger stopping decisions.
Open decision: A decision where competent people would not necessarily agree and where the rule cannot be written down. Open decisions are what cognitive units are for.
Referral: A cognitive unit reporting that the decision is not its to make. Distinct from abstention: the remedy is to find the person with authority.
Retrospection: Structured inspection after a cognitive result, action, state transition, or completed trajectory.
Skill: A named, reusable, bounded procedure that combines Cognitive Units, deterministic code, tools, and sometimes other Skills in order to perform a known capability. A Cognitive Unit owns a decision, a Skill owns a known procedure, an Agent owns progress toward a goal.
Slow path: A deliberate trajectory that constructs strategy, retrieves more information, explores alternatives, and reviews more than once. Novelty and consequence retain the slow path.
Submergence Principle: As a technology becomes normal, its implementation vocabulary moves beneath the surface.
Thoughtware: Software in which meaningful judgments are deliberately handled by cognitive components alongside deterministic code, supported by context, memory, behaviour, evaluation, and mechanisms for improvement.
Thoughtware Map: The shared architectural view of an outcome, its Judgment Chains, agents, cognitive units, deterministic boundaries, memory, behaviour, authority, evaluation, and learning paths.
Trust: An accumulated human judgment about whether a system’s conduct, limits, memory, explanations, and recovery remain dependable across time.
Working state: The persistent and versioned representation of goal, problem understanding, strategy, progress, observations, unresolved gaps, authority, and remaining resources for one running instance.