The irreducible epistemic atoms underlying the curriculum. 4,828 atoms across 8 types and 2 molecules
When writing stalls on a supposedly understood topic, treat the stall point as a specific learning target rather than a writing problem.
Match capture modality to information structure: use text for sequential verbal content, voice when hands are occupied, and photographs for spatial or visual information.
When a note contains multiple ideas connected by 'and' or 'also,' create separate notes—one per idea—with explicit links between them, rather than allowing compound ideas to remain fused in a single container.
Assign a unique identifier to every note before writing any content, treating the addressing decision as the first step that enables all subsequent linking and referencing.
Before attempting decomposition of any complex idea, map it as a whole with your current understanding externalized, then decompose systematically until you encounter steps you cannot explain clearly—those uncertainty points are your actual knowledge gaps.
Apply the 'link test' by checking whether all links from a note feel relevant to the entire note—if some links connect only to parts, the note contains multiple units requiring separation.
Before committing resources to a plan, extract every assumption it depends on and classify each by importance (would the plan fail if this is wrong?) and vulnerability (how likely is this to be wrong?), then test assumptions marked both high-importance and high-vulnerability first.
When someone challenges one part of your compound plan and you defend the whole thing, treat this as a diagnostic signal that you're still operating on fused ideas rather than independent assumptions.
When presenting compound statements to AI systems, explicitly ask for assumption enumeration rather than direct answers, then critically verify the decomposition's completeness since the AI may introduce its own hidden assumptions.
When unpacking task estimates, decompose complex tasks into component steps before estimating duration—unpacking improves accuracy by forcing visibility of dependencies and transitions that holistic estimation skips.
When encountering difficulty naming a concept precisely, treat that difficulty as a diagnostic signal revealing incomplete understanding requiring further processing rather than a labeling problem.
Store evidence with full methodological metadata (sample size, control conditions, limitations) as independent nodes rather than as decorative citations on claims, to enable proportionality assessment and multi-argument reuse.
Before forcing resolution of contradictory observations or beliefs, accumulate multiple instances in a contradiction log to enable pattern detection impossible from individual contradictions.
When AI retrieval quality degrades despite good source material, diagnose whether notes are self-contained units or fragments requiring external context, because fragmentation produces context confusion that corrupts AI reasoning.
Match note granularity to retrieval frequency and question complexity: create fine-grained atomic notes (single claims) for domains where you need precise retrieval, and coarser aggregated notes for domains where you need high-level orientation.
When using AI systems with your knowledge base, maintain multiple granularity levels of the same material (fine-grained for precise retrieval, coarse-grained for contextual reasoning) rather than forcing a single chunk size, because different query types require different resolutions.
Store well-formed questions as first-class atoms in your knowledge system with the same structural treatment (unique identifiers, bidirectional links, metadata) as claims and answers, because questions organize attention and generate persistent search filters.
When a question receives a partial answer, preserve the original question as a persistent atom and link the answer to it rather than replacing the question, creating a visible record of how understanding evolves from open inquiry to accumulated evidence.
When using AI to analyze accumulated evidence around an open question, provide the constellation of linked notes (question + partial answers + contradictions + gaps) as context rather than asking the AI to answer from scratch, because the accumulated context enables pattern recognition your cold query cannot access.
For every high-stakes term in your reasoning (quality, success, productive, fair), write an operational definition specifying observable conditions that must be true for the term to apply, then store that definition as a canonical reference atom in your knowledge system.
When two people or two parts of your own thinking use the same term with persistent conflict, pause the debate and conduct a definition audit: have each party write their operational definition independently, then compare—if definitions diverge, the conflict is definitional not factual and should be resolved at the definition level.
Feed your operational definitions to AI systems as explicit context before generating analysis or recommendations, treating your personal glossary as the translation layer between the model's probability-weighted semantics and your specific conceptual framework.
When encountering the same insight expressed in three or more separate notes across different contexts, extract the shared structural pattern into a single canonical note with a precise name, then replace the duplicate instances with links to the canonical abstraction.
When considering whether to merge two similar notes, test whether the underlying structure is identical (same entities, same relationships, same claims) rather than whether the vocabulary overlaps, because structural identity warrants abstraction while surface similarity does not.