Glossary

This wiki is generated from a justified belief network managed by a Truth Maintenance System (TMS). Every belief on this site is a justified claim with an inspectable “why.” Beliefs that have been abandoned are graves you can visit, not pages that vanish — the record of what was believed and why it was given up is itself knowledge. The following terms explain how to read belief pages.

Premise vs. Derived

• A premise is a direct observation or assertion with no justification chain. It is IN by default and can only go OUT if explicitly retracted.
• A derived belief is supported by one or more justifications that reference other beliefs (see Justifications below). Its truth value is computed automatically from the network.

IN and OUT

Every belief has a truth value: IN or OUT.

• IN means the belief is currently justified — its supporting evidence holds and no active defeater contradicts it. IN does not mean “proven true in all possible worlds”; it means “supported by the current state of the network.”
• OUT means the belief is not currently justified. This is not the same as “false.” A belief goes OUT in two distinct ways:
  • Explicit retraction — someone (a human or an agent) marks the belief OUT with a stated reason. These beliefs show a Reason OUT on their page.
  • Cascade — a belief the network depended on went OUT, and this belief lost its last valid justification as a result. Cascade-OUT beliefs have no Reason OUT label; the cause is visible in their justification chain (one or more antecedents will be OUT).
  OUT beliefs are retained in the network rather than deleted. The record of what was believed and why it was given up is itself knowledge.

Well-Foundedness

IN status must ultimately be grounded in premises — no belief can hold itself up by its own bootstraps. If two beliefs each list the other as an antecedent with no premise anchor, neither can be IN. This well-foundedness constraint prevents circular support and ensures that every IN belief traces back to at least one direct observation or assertion.

Depth

Depth measures how far a derived belief is from the premises it rests on. Depth 0 is a premise. Depth 1 means the belief is derived directly from premises. Depth 2 means it depends on a depth-1 belief, and so on. When a belief has multiple justifications with different chain lengths, depth reflects the longest chain (maximum over all antecedents across all justifications). Higher depth means more reasoning steps between this belief and the observations it ultimately depends on — more inferential distance, but also more points where the chain could break.

Justifications

A justification is a rule that says: “this belief is IN if all its antecedents are IN and none of its unless (outlist) nodes are IN.” This is called an SL (Support List) justification — the term comes from Doyle’s 1979 truth maintenance system.

• Antecedents — beliefs that must all be IN for this justification to hold. If any antecedent goes OUT, the justified belief goes OUT too (unless another justification still supports it).
• Unless (outlist) — beliefs that defeat this justification if they become IN. This is the non-monotonic reasoning mechanism: it allows the network to express “A is true unless B” — default reasoning that can be overridden by new evidence.

A belief can have multiple justifications. It stays IN as long as at least one justification is satisfied.

Challenges and Defenses

A challenge is a belief that contests another belief by adding itself to the outlist of every justification the target has. This means a single challenge defeats all of the target’s justifications at once — the target goes OUT unless a defense neutralizes the challenge. If the target is a premise (no justifications), it is converted to a justified node with the challenge in its outlist.

A defense counters a challenge using the same mechanism in reverse: it places the challenge in its own outlist, creating a dialectical structure. Since the defense is IN by default, the challenge goes OUT, which removes it as a defeater and restores the original belief. Because a defense is itself a belief, it can be challenged in turn — the structure recurses arbitrarily, producing chains of challenge, defense, counter-challenge, and so on.

Nogoods and Dependency-Directed Backtracking

A nogood is a recorded contradiction — a set of beliefs that cannot all be IN simultaneously. When the system detects a nogood, it performs dependency-directed backtracking: it traces the contradiction to its root causes and retracts the least-entrenched premise responsible. This is more targeted than blind retraction — the system uses the justification graph to find the weakest link rather than arbitrarily choosing what to give up. Beliefs retracted by backtracking carry a Reason OUT explaining the contradiction that triggered the retraction.

Retraction and Cascades

When a belief is explicitly retracted, the system propagates the change: every derived belief that depended on the retracted belief is re-evaluated. If a derived belief has no remaining valid justification, it goes OUT too. This cascade continues through the network until all truth values are consistent.

Only explicitly retracted beliefs show a Reason OUT on their wiki page. Cascade-OUT beliefs have no Reason OUT — instead, you can trace the cause by following their justification chain until you find the antecedent that went OUT.

Reading a Belief Page

Each belief page is laid out top to bottom: the status line (IN or OUT, premise or derived with depth, and source provenance), optional dates, the Reason OUT if explicitly retracted, the canonical belief text as a blockquote, a plain-language Summary, then the graph edges — Justifications upward (antecedents and unless nodes, each with their own truth value tags), optional Challenges, and Dependents downward (beliefs that cite this one). Every linked belief is clickable, so you can walk the justification graph in either direction.

Provenance

Belief pages show provenance metadata when available: the source (where the belief was observed or derived from, such as a code exploration entry or review report), an optional source URL for external references, and timestamps for when the belief was created, last reviewed, and last verified. Provenance appears inline on the status line and as dates below it. For code-domain beliefs, the source file is the ground truth. For world-knowledge beliefs, provenance becomes critical metadata for resolving contradictions.

Topics

Beliefs are grouped into topics by an LLM classifier that reads each belief’s text and assigns it to a semantic category. Topics are not part of the TMS data model — they are a navigational layer added by the wiki generator to make large networks browsable.