--- schema_version: "1.0" project_name: updated_at: "2026-06-05T02:21:09+00:00" node_count: 49 generator: ftl-reasons/0.43.0 --- # Belief Registry ## Claims ### agm-belief-revision [IN] OBSERVATION AGM (Alchourrón, Gärdenfors, Makinson 1985) provides formal theory for rational belief revision. Entrenchment scoring in backtracking is a crude approximation of AGM - Source: repo:beliefs-pi/CLAUDE.md ### atms-de-kleer-1986 [IN] OBSERVATION de Kleer (1986) ATMS uses assumption-based environments and nogoods. TMS beats ATMS for EEM because revision matters more than multiple environments when the problem solver (LLM) produces 13-37% errors - Source: repo:beliefs-pi/CLAUDE.md ### beliefs-cli-vs-reasons-cli [IN] OBSERVATION Two CLIs at different levels: beliefs CLI is a structured markdown KB with provenance and manual maintenance (simple, flat). reasons CLI (ftl-reasons) is a full TMS with automatic propagation, cascades, backtracking, and LLM-driven operations (powerful, dependency-aware). Use beliefs for independent facts, reasons for justified conclusions with dependency chains - Source: repo:beliefs-pi/CLAUDE.md ### challenge-defend [IN] DERIVED Dialectical argumentation: challenging a node makes it go OUT. Defending neutralizes the challenge. Multi-level chains supported — challenge the defense, defend the defense, etc. Preserves the original argument unlike retract - Source: repo:beliefs-pi/CLAUDE.md - Depends on: ftl-reasons-is-tms ### cognitive-budget [IN] OBSERVATION Cognitive budget principle borrowed from graphics frame budgets: decompose work into focused passes (TMS pass, RAG pass, merge pass) each within the model's attention budget. Mixing beliefs and document chunks in a single prompt degrades performance (Opus drops 95.5% to 86%); three focused passes achieve 100% - Source: repo:beliefs-pi/entries/2026/05/04/cognitive-budget-borrows-from-graphics-frame-budget.md ### confidence-unreliable [IN] OBSERVATION LLM self-assessed confidence does not track accuracy. Confirmed across 4 models: Sonnet r=0.198, Opus r=-0.182 (worse than random), Flash r=0.219, Pro r=0.121. Answer and confidence come from the same process — same structural flaw as human overconfidence (Kahneman) - Source: repo:beliefs-pi/CLAUDE.md ### construction-vs-retrieval [IN] OBSERVATION Construction cost dominates: O(chunks) + O(beliefs x rounds). But it amortizes across all queries O(queries). Expensive to build, cheap to query at scale - Source: repo:beliefs-pi/CLAUDE.md ### continuity-human-problem [IN] DERIVED The human cannot track what the LLM currently has in context. EEM solves this via visibility and persistence — the human can always inspect the current belief state. Short compaction cycles are better than large context windows - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-external ### derive-then-review [IN] DERIVED Over-derive, then review catches errors, retraction cascades propagate corrections. Both roles overshoot (derive over-generates, review over-retracts). Working through candidate retractions is where insights hide - Source: repo:beliefs-pi/CLAUDE.md - Depends on: hybrid-tms ### dual-path-architecture [IN] DERIVED Dual-path retrieval: TMS path (pre-computed beliefs) + FTS path (source chunk search), merged by a third pass. This is how EEM is queried at scale. Each path stays within cognitive budget - Source: repo:beliefs-pi/CLAUDE.md - Depends on: cognitive-budget ### eem-compensates-model-size [IN] DERIVED EEM compensates for model size — smaller models with EEM match larger models without it - Source: repo:beliefs-pi/CLAUDE.md - Depends on: evidence-model-compensation, eem-works ### eem-definition [IN] OBSERVATION External Epistemic Memory (EEM) is knowledge that lives outside the model, carries its justifications with it, and lets you understand how the system knows what it knows - Source: repo:beliefs-pi/CLAUDE.md ### eem-epistemic [IN] OBSERVATION Epistemic means not just facts but justified beliefs with truth values (IN/OUT), retraction cascades, contradiction records (nogoods), and derivation depth. This distinguishes EEM from RAG (which is external semantic memory but not epistemic) - Source: repo:beliefs-pi/CLAUDE.md ### eem-external [IN] OBSERVATION External means outside model parameters, in a separate substrate. Survives compaction, model swaps, session boundaries. Inspectable, editable, auditable, shareable across models - Source: repo:beliefs-pi/CLAUDE.md ### eem-memory [IN] OBSERVATION Memory in Tulving's semantic memory category — persistent structured knowledge, not ephemeral context - Source: repo:beliefs-pi/CLAUDE.md ### eem-replaces-confidence [IN] DERIVED EEM replaces 'am I sure?' with 'is this justified?' — shifting from unreliable confidence to auditable justification chains - Source: repo:beliefs-pi/CLAUDE.md - Depends on: confidence-unreliable, eem-epistemic ### eem-three-properties [IN] DERIVED EEM is defined by three load-bearing properties: external (outside parameters), epistemic (justified with truth values), and memory (persistent semantic knowledge) - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-external, eem-epistemic, eem-memory ### eem-vs-context [IN] DERIVED Conversation history and context windows are ephemeral — lost at session boundaries, destroyed by compaction. EEM persists across sessions and model swaps. Context compaction destroys justification networks (quantified across 33 measured compaction events) - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-external ### eem-vs-parametric [IN] DERIVED In-parameter knowledge has no audit trail. EEM makes 'how do you know that?' answerable by justification chain traversal. Of the six externality properties (separable, copyable, shareable, inspectable, editable, auditable), auditability is the most epistemically important - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-external, eem-epistemic ### eem-vs-rag [IN] DERIVED RAG is external semantic memory but not epistemic. It retrieves content by similarity but has no justification chains, truth values, retraction cascades, or contradiction tracking. EEM adds the epistemic layer that RAG lacks - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-epistemic ### eem-works [IN] DERIVED EEM measurably and dramatically improves LLM performance on domain tasks. The core research question is answered: yes - Source: repo:beliefs-pi/CLAUDE.md - Depends on: evidence-expert-vs-baseline, evidence-beliefs-ablation, evidence-dual-path ### evidence-beliefs-ablation [IN] OBSERVATION Beliefs alone outperform beliefs + expert prompt: Opus 100% vs 94.2% (+5.8pp), Sonnet 94.2% vs 91.8% (+2.4pp). Adding expert prompt hurts — agent trusts its 'expertise' instead of consulting the knowledge base - Source: repo:beliefs-pi/entries/2026/03/15/beliefs-ablation-results-structured-knowledge-beats-prompt-engineering-expert-prompts-may-hurt.md ### evidence-depth-ceiling [IN] OBSERVATION Beliefs beyond depth 8 do not survive review. Retraction rate: 0% at depth 0, rising to 100% at depth 9+. The universal TMS is wide rather than deep - Source: repo:beliefs-pi/CLAUDE.md ### evidence-dual-path [IN] OBSERVATION Opus + dual-path architecture achieves 98.5% A/B across 3,853 questions. Zero D/F grades — eliminated the failure tail entirely - Source: repo:beliefs-pi/entries/2026/05/02/full-scale-validation-3853-questions-dual-path-985-percent.md ### evidence-expert-vs-baseline [IN] OBSERVATION Expert-service with EEM scores 88% A-grade vs agents-python 33% on same 50 questions, 15x faster - Source: repo:beliefs-pi/CLAUDE.md ### evidence-model-compensation [IN] OBSERVATION EEM compensates for model size: Sonnet+beliefs approximates Opus without beliefs. Haiku with dual-path achieves 94% A+B, matching Opus at 98% - Source: repo:beliefs-pi/CLAUDE.md ### evidence-retraction-rate [IN] OBSERVATION 13-37% of derived beliefs are retracted per review round across multiple expert KBs. Self-correction works — the system finds and removes its own errors - Source: repo:beliefs-pi/CLAUDE.md ### expert-pipeline [IN] DERIVED Expert pipeline: chunk source material → propose beliefs → human accepts → derive connections → review derivations → export. Value accrues at each stage, with derive producing new knowledge (connections the source doesn't make explicit) - Source: repo:beliefs-pi/entries/2026/05/06/knowledge-as-a-service-kaas.md - Depends on: hybrid-tms, derive-then-review ### expert-prompt-paradox [IN] DERIVED Telling an agent it is an expert reduces belief utilization. The humble generic prompt produces better results because the agent consults the knowledge base instead of trusting its 'expertise' - Source: repo:beliefs-pi/entries/2026/03/15/beliefs-ablation-results-structured-knowledge-beats-prompt-engineering-expert-prompts-may-hurt.md - Depends on: evidence-beliefs-ablation ### frame-problem [IN] OBSERVATION McCarthy & Hayes (1969) frame problem: what persists across state changes. check-stale addresses this by detecting when source files change under beliefs - Source: repo:beliefs-pi/CLAUDE.md ### ftl-reasons-is-tms [IN] DERIVED ftl-reasons implements actual Doyle-style TMS architecture: SL justifications with antecedents and outlists, BFS propagation cascades with restoration, entrenchment-scored dependency-directed backtracking. LLMs fill the problem-solver role Doyle left open - Source: repo:beliefs-pi/CLAUDE.md - Depends on: tms-doyle-1979 ### generate-and-critique [IN] DERIVED LLMs are extraordinary generators but unreliable critics. The belief registry externalizes and persists the critic's judgments, replacing internal self-assessment with external structured tracking - Source: repo:beliefs-pi/entries/2026/05/06/generate-and-critique-llms-are-half-a-mind.md - Depends on: confidence-unreliable, self-critique-harmful ### how-agents-use-eem [IN] DERIVED LLM agents use EEM by: querying beliefs via search/show/explain before answering, citing node IDs for auditability, running derive to generate new beliefs from existing ones, running review-beliefs to self-audit, recording nogoods when contradictions appear. The agent does not need to be told it is an expert — the knowledge base speaks for itself - Source: repo:beliefs-pi/CLAUDE.md - Depends on: expert-prompt-paradox, hybrid-tms ### how-humans-use-eem [IN] DERIVED Humans use EEM by: inspecting beliefs.md for current state, running reasons explain to understand why something is believed, challenging beliefs with reasons challenge, reviewing the network with reasons status, checking staleness with reasons check-stale. The key value is visibility — humans can see and audit what the system knows - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-external, continuity-human-problem ### how-to-start [IN] DERIVED To start using EEM: (1) reasons init — creates reasons.db, (2) add premises from observations with reasons add, (3) add justified conclusions with --sl to link dependencies, (4) use reasons derive to find connections, (5) use reasons review-beliefs to audit, (6) retract when evidence changes and let cascades propagate - Source: repo:beliefs-pi/CLAUDE.md - Depends on: ftl-reasons-is-tms, expert-pipeline ### hybrid-tms [IN] DERIVED ftl-reasons is a hybrid TMS: symbolic TMS handles structure (justifications, propagation, cascades, backtracking, challenge/defend) while LLMs handle semantic operations (derive generates beliefs, review-beliefs critiques them, contradiction detection finds nogoods) - Source: repo:beliefs-pi/CLAUDE.md - Depends on: ftl-reasons-is-tms, llm-as-problem-solver ### llm-as-problem-solver [IN] DERIVED Putting an LLM in the TMS problem-solver slot (generator via derive, critic via review-beliefs and contradiction detection) is what Doyle's architecture prescribes. The open question is whether an LLM is a good problem solver, not whether using one is faithful to the design - Source: repo:beliefs-pi/entries/2026/05/06/paper-expert-doyle-1979-belief-network-analysis.md - Depends on: tms-doyle-1979 ### model-stacking [IN] DERIVED Multi-pass agent pattern: Model A generates candidates → TMS records with provenance → Review critiques (machine + human) → Model B receives validated beliefs → Model B derives new beliefs → Review critiques derivations → Repeat. Each level is a full model pass with fresh context and critique pipeline as quality gate - Source: repo:beliefs-pi/entries/2026/05/06/model-stacking-tms-as-the-layer-between-models.md - Depends on: hybrid-tms, derive-then-review ### multi-agent-beliefs [IN] DERIVED Multi-agent TMS: import-agent imports another agent's beliefs with SL justifications including agent:active as antecedent. Node is IN iff agent is active AND original belief is justified. Doyle-style truth maintenance across agents - Source: repo:beliefs-pi/entries/2026/03/28/import-agent-multi-agent-belief-tracking-for-cross-repo-knowledge.md - Depends on: ftl-reasons-is-tms ### nogood-mechanism [IN] DERIVED A nogood is a set of nodes that cannot all be IN simultaneously. When detected, dependency-directed backtracking traces backward through justification chains and retracts the responsible premise with fewest dependents (minimal disruption) - Source: repo:beliefs-pi/CLAUDE.md - Depends on: ftl-reasons-is-tms, agm-belief-revision ### reasons-for-maintenance-beliefs-for-queries [IN] DERIVED Architecture pattern: use reasons database for all structural operations (add, retract, derive, review). Export to beliefs.md for querying (fast, human-readable, grep-able). Keep both in sync via reasons export-markdown - Source: repo:beliefs-pi/entries/2026/03/24/agents-python-expert-primary-store-migration-complete-and-deep-reasoning-chains.md - Depends on: beliefs-cli-vs-reasons-cli ### restoration [IN] DERIVED When a retracted node comes back IN, dependents are recomputed — no manual rederivation needed. The TMS tracks structure so restoration is automatic - Source: repo:beliefs-pi/CLAUDE.md - Depends on: sl-justification ### retraction-cascade [IN] DERIVED When a node goes OUT, all dependents whose justifications become invalid also go OUT — automatically, transitively. This is the most important operation: retract one belief and the network figures out what else falls - Source: repo:beliefs-pi/CLAUDE.md - Depends on: sl-justification ### scale-evidence [IN] DERIVED EEM scales from small domains (237 beliefs, aap-expert) to large enterprises (12,731 beliefs, redhat-expert). 40+ expert knowledge bases built across code, product, project, and domain-specific experts - Source: repo:beliefs-pi/CLAUDE.md - Depends on: eem-works ### self-critique-harmful [IN] OBSERVATION LLM revision based on self-critique makes answers worse: Sonnet -11pp, Flash -21pp, Pro -56.5pp. Self-critique fails because the same model that made the error evaluates the error - Source: repo:beliefs-pi/entries/2026/05/06/generate-and-critique-llms-are-half-a-mind.md ### self-improvement [IN] DERIVED The system finds problems in itself. Each improvement improves the system's ability to find the next improvement — exponential compounding vs linear improvement in static systems - Source: repo:beliefs-pi/CLAUDE.md - Depends on: derive-then-review, evidence-retraction-rate ### sl-justification [IN] DERIVED SL (Support List) justification: a node is IN when ALL antecedents are IN. Multiple justifications allowed — node stays IN if ANY justification is valid. Enables non-monotonic reasoning via outlist (believe X unless Y) - Source: repo:beliefs-pi/CLAUDE.md - Depends on: ftl-reasons-is-tms ### tms-doyle-1979 [IN] OBSERVATION Doyle (1979) designed Truth Maintenance Systems with SL justifications, propagation, retraction cascades, and an exogenous problem-solver slot. The TMS substrate is content-agnostic by design - Source: repo:beliefs-pi/entries/2026/05/06/paper-expert-doyle-1979-belief-network-analysis.md ### wide-not-deep [IN] DERIVED The universal TMS is wide rather than deep. Depth-8 ceiling is structural. Experiments reset derivation depth by providing new depth-0 observations - Source: repo:beliefs-pi/CLAUDE.md - Depends on: evidence-depth-ceiling