intelligence-engine

Intelligence Engine

Prime Directives

Read live source before writing code. Read refs/ for library internals. Read the Index-API codebase for application patterns. Do not rely on training data for either. Treat specs as intent; reconcile with current implementation before editing.

Route by Workflow

1. /reason — text -> claims -> tensions -> models (NLP, NLI, KR). Load agents: claim-analysis, nlp-pipeline, knowledge-representation, information-retrieval, probabilistic-reasoning.

2. /graph — objects -> structure -> self-organization (graph, causal, GNN, temporal memory). Load agents: graph-theory, causal-inference, self-organization, knowledge-representation, probabilistic-reasoning, graph-neural-networks, temporal-graph-memory, systems-theory.

3. /train — features -> models -> evaluation -> adaptation (ML, RL, evolution, enrichment pipeline). Load agents: learned-scoring, training-pipeline, graph-neural-networks, temporal-graph-memory, language-model-training, multimodal-networks, reinforcement-learning, evolutionary-optimization, domain-specialization.

4. /architect — system design, feedback loops, pipeline optimization. Load agents: software-architecture, self-organization, program-synthesis, systems-theory, evolutionary-optimization, domain-specialization.

5. /simulate — hypotheses, debate, counterfactuals, belief revision. Load agents: claim-analysis, causal-inference, probabilistic-reasoning, multi-agent-reasoning, symbolic-reasoning, counterfactual-simulation, language-model-training, systems-theory, temporal-graph-memory.

6. /measure — IQ tracking across 7 axes, benchmarking, leverage analysis. Load agents: information-retrieval, graph-theory, self-organization, learned-scoring, reinforcement-learning, systems-theory, domain-specialization.

7. /learn — compound learning lifecycle. Save session log, run confidence updates, review auto-captured claims, resolve tensions, and process stale/low-confidence attention items.

Compound Learning Rules

1. Treat knowledge/claims.jsonl as live operational memory for reusable practices.
2. Keep claim extraction imperative and actionable.
3. Use SHA-based deduplication for claim IDs.
4. Log sessions in knowledge/session_log/ and run /learn after meaningful implementation or debugging sessions.
5. Apply claim confidence thresholds before overriding static guidance.

Invariants (Always Enforce)

1. Edge.reason is plain-English, human-readable.
2. Edge uses from_object / to_object (not source/target).
3. Timeline Nodes are immutable (except retrospective_notes).
4. SHA-hash identity tracks provenance. Never bypass _generate_sha().
5. Per-Notebook engine_config controls pass behavior.
6. Objects soft-delete only (is_deleted=True).
7. LLMs propose. Humans review. Nothing auto-promotes to canon.
8. compose_engine is stateless (text-in, objects-out, no DB writes).
9. engine.py is stateful (object-in, edges + nodes out).
10. Every epistemic primitive carries its provenance.

Two-Mode Contract

• PRODUCTION (Railway): spaCy + BM25 + TF-IDF. No PyTorch.
• LOCAL/DEV: All 7+ passes. PyTorch + FAISS + SBERT + NLI + KGE.
• MODAL (GPU batch): GNN training, KGE training, LoRA fine-tuning, corpus NLI, SAM-2, evolutionary optimization.

Enrichment Pipeline Focus Areas

• EpiGNN extensions: HeterophilyAwareConv, ORC-weighted aggregation, two-state outputs.
• KGE RotatE: triple export, training, runtime scoring features, structural token generation.
• SBERT enrichment: triplet pipelines, fine-tuning, multi-view embeddings, graph-text alignment.
• GL-Fusion three-stream architecture: Stream A/B/C integration and gate variance checks by query type.
• Model swarm constraints: GPU-only hard fusion, CPU GGUF soft-fusion fallback, and internet mediation via validated Object ingestion.
• Network effects: private/shared/federated graph behavior and promotion-risk controls.

Use these pattern files when implementing the above:

• patterns/PATTERNS-epignn.md
• patterns/PATTERNS-kge-rotate.md
• patterns/PATTERNS-sbert-enrichment.md
• patterns/PATTERNS-gl-fusion-three-stream.md
• patterns/PATTERNS-compound-learning.md
• patterns/PATTERNS-model-swarm.md
• patterns/PATTERNS-network-effects.md

Theseus Roadmap

Level	Name	Status
1	Tool-Based Intelligence	Shipped
2	Learned Connection Scoring	Building
3	Hypothesis Generation	Designed
4	Emergent Ontology	Designed
5	Self-Modifying Pipeline	Designed
6	Multi-Agent Epistemic Reasoning	Designed
7	Counterfactual Simulation	Designed
8	Creative Hypothesis Generation	Designed

IQ Tracker (7 Axes)

Discovery (0.20), Organization (0.15), Tension (0.15), Lineage (0.10), Retrieval (0.15), Ingestion (0.10), Learning (0.15). Current composite: check live via theseus_measure_iq MCP tool.

Reference Pointers

• Patterns: patterns/PATTERNS-*.md for implementation recipes.
• Product: product/*.md for specs, roadmap, and design docs.
• Examples: examples/{train,simulate,measure,reason,graph}/*.py for code.
• Agents: agents/*.md for domain-specific expertise (24 agents, 3 tiers).
• Knowledge memory: knowledge/*.jsonl and knowledge/manifest.json.

Delivery Checklist

1. Reconcile code vs spec and state mismatches explicitly.
2. Implement smallest safe slice that preserves invariants.
3. Add or update tests/fixtures for new behavior.
4. Validate runtime and degradation paths.
5. Report: initial condition, reconciliation, changes, validation, next steps.