discipline-distillation

name

discipline-distillation

description

Universal methodology for distilling any mathematically rigorous discipline into a reasoning-constrained skill graph consumable by AI agents. Three-track extraction (reasoning mining, knowledge grounding, dependency mining), systematic rumination with blind-derivation validation, and cross-volume bridge construction. Complete worked example: Landau-Lifshitz 10-volume Course of Theoretical Physics (96 nodes, 188 edges).

Discipline Distillation — From Textbook to Reasoning Graph

What This Is

A complete, battle-tested methodology for turning a textbook (or any mathematically rigorous body of knowledge) into a graph of reasoning templates and knowledge nodes that AI agents can use for autonomous physics reasoning — without access to the original text.

Worked example: Landau-Lifshitz 10-volume Course of Theoretical Physics → 96 reasoning + knowledge nodes, 188 typed edges, validated by two independent AI agents in blind-derivation experiments.

The Big Picture

TEXTBOOK(S)
    ↓ Track A: Extract reusable reasoning patterns
    ↓ Track B: Ground each pattern in concrete knowledge
    ↓ Track C: Map conceptual dependencies
REASONING GRAPH (nodes + typed edges)
    ↓ Rumination: re-read original, find hidden premises, fix
    ↓ Cross-volume: build bridges between volumes
    ↓ Validation: blind-derivation experiment
PRODUCTION GRAPH (agent-consumable)

Three Phases

Phase 1: Extraction

Three-track extraction per volume/section. Full details: references/extraction-workflow.md

• Track A — Reasoning Mining: Find reusable PHYSICAL THINKING PATTERNS (not chapter summaries). Scale hierarchy, symmetry→conservation, effective theory, variational logic, perturbation, causality constraints, dimensional analysis, adiabatic approximations.
• Track B — Knowledge Grounding: For each reasoning pattern, attach concrete physics: equations, physical pictures, validity conditions, edge cases, misconceptions.
• Track C — Dependency Mining: What MUST be known before what? Conceptual order, not textbook order. Build prerequisite DAG.

Phase 2: Rumination

After extraction, systematically re-read the original text with the graph loaded. Find hidden premises, missing edge cases, unstated assumptions. Full details: references/rumination-workflow.md

Critical lesson from Landau distillation: fast passes that find "no issues" are diagnostically wrong. Every volume (10/10) had issues when properly re-read. Common patterns: references/rumination-check-patterns.md

Phase 3: Validation

The acid test: give the reasoning graph (WITHOUT the original text) to independent AI agents. Ask them to derive the discipline's core results. Every point where they get stuck is a gap in the graph.

Full details: references/blind-derivation-validation.md

Worked Example: Landau-Lifshitz 10 Volumes

See examples/landau-10-volumes/ for the complete distillation record.

Aspect	Result
Source	10 volumes, ~7000 pages, Russian original
Output	96 nodes (37 reasoning + 59 knowledge), 188 edges
Rumination	All 10 volumes ruminated, 80+ fixes applied
Hidden parents	3 discovered (symmetry_drives_physics, timescale_hierarchy, causality_as_constraint)
Validation	2 independent AI agents, 60+ confusions reported and resolved
Iterations	3 rounds of blind-derivation → fix → re-validate

Skills Consolidated Here

This skill supersedes and replaces:

• landau-distillation (methodology)
• landau-graph-workflow (extraction pipeline)
• landau-rumination (rumination workflow + check patterns)

The landau-graph skill remains as the PRODUCTION GRAPH (96 nodes, 188 edges) — it is the OUTPUT of this methodology applied to Landau.