Concepts
The irreducible epistemic atoms underlying the curriculum. 4,828 atoms across 8 types
Ask AI for multiple sequences of your notes, not one best…
Present AI systems with your atomic notes and ask for multiple possible sequences (chronological, causal, problem-solution) rather than asking for a single best structure, using AI to discover sequences rather than impose them.
Split notes at 800 words or 3 topics — decomposition…
When a note exceeds 800 words or covers three distinct topics, decompose it into 2-4 separate atomic notes and rewrite the connections between them to reveal causal chains invisible in the original structure.
Before splitting a note, decide
When splitting a compound note during refactoring, make explicit decisions about which idea is the core claim, what was supporting evidence versus separate argument, and how the pieces causally relate before completing the split.
Let AI detect structural debt — make refactoring decisions…
Use AI to audit your knowledge base for structural debt (compound notes, duplicates, orphans, broken connections) but perform the actual refactoring decisions yourself to gain the cognitive benefit.
Sequence gaps during refactoring are specs for new notes…
When refactoring reveals that notes in a sequence jump or break, treat those gaps as specifications for new atoms to write rather than as sequence failures.
Uncertain about atomicity during capture? Write it now…
When unable to determine if a note contains one idea or two, write it as-is during capture, then return during a dedicated review session to attempt decomposition without the pressure of real-time capture.
Improve one thing every time you touch a note — kaizen…
Each time you review or link a note, make one small improvement (sharpen title, add missing context, split tangled claim) rather than scheduling separate cleanup sessions.
Keep raw inbox captures out of AI retrieval scope
Index only processed permanent notes in AI-searchable systems while keeping unprocessed inbox captures outside retrieval scope, because AI systems cannot distinguish epistemic status and will retrieve raw captures with equal confidence to verified knowledge.
Ask four diagnostic questions during weekly review…
During weekly reviews, ask four metacognitive questions—what did I capture well, what did I almost lose, where did I over-capture noise, and what am I avoiding—to monitor system health rather than just processing lists.
Digitize analog notes weekly — handwriting for capture…
For analog captures intended for long-term use, implement a pipeline that photographs or transcribes key entries into digital storage during weekly review—preserving handwriting's cognitive benefits during capture while enabling digital searchability and AI-readability for retrieval.
Keep separate Candidate and Confirmed pattern lists…
Maintain two separate lists—'Pattern Candidates' and 'Confirmed Patterns'—promoting candidates only after they survive three independent observations, one alternative-explanation check, and one successful prediction.
Quarterly
Schedule quarterly depreciation reviews where you scan captured notes and bookmarks for information that has exceeded its useful life, then either update with current data, archive with context, or delete entirely to prevent outdated information from corrupting current decisions.
Write 'This connects to [X] because [Y]' after consuming…
After consuming any piece of information, write one connecting sentence that relates it to existing knowledge using the structure 'This connects to [X] because [Y]'; if you cannot write this sentence within two minutes, classify the content as non-compounding noise regardless of its intrinsic quality.
Domain facts need aggressive updating; structural…
Distinguish domain-specific facts (treatment protocols, software frameworks, market conditions) requiring aggressive temporal updating from structural principles (logic, mathematics, core psychological mechanisms) where age indicates Lindy-tested robustness, applying opposite update strategies to each type.
Record decisions with five fields
Document decisions using five fields: what you decided, alternatives considered, information available and missing, optimization criteria, and conditions for revisiting—rather than recording only conclusions.
Structure learning notes as: claim, evidence, connection…
Write learning in the structure: claim (one sentence, your words), evidence (why believe it), connection (how it relates), question (what's unresolved) to force generation rather than transcription.
Document your knowledge system in 5 parts
Document system operations in five components—capture rules, processing workflow, retrieval method, review protocol, and evolution history—because each component addresses a distinct failure mode in knowledge system sustainability.
Give AI your full knowledge system, not isolated questions
Feed complete externalized system context to AI assistants rather than isolated queries, because AI reasoning quality scales with the completeness and structure of the personal knowledge base it can traverse.
Attach scope metadata to every schema — where it was built…
Store each schema with explicit scope documentation specifying the domain where it was built and the structural conditions it assumes, treating scope as mandatory metadata rather than optional annotation.
For each category, complete
Document the purpose each category serves by completing the sentence 'this category exists to [do what] for [whom]' to distinguish functional infrastructure from inherited furniture.
Match classification architecture to domain
Design multi-class classification systems with mutually exclusive categories when items can only be one type, and multi-label systems when items can legitimately belong to multiple categories simultaneously.
Audit categories against your values — missing values…
For each top-level category in your knowledge system, write one sentence explaining what value that category protects or promotes, then identify missing categories that would operationalize values you hold but aren't currently encoding.
A growing 'Miscellaneous' category signals missing…
When a 'Miscellaneous' or 'Other' category grows faster than named categories, it signals that your classification dimensions are missing a meaningful distinction that reality contains.
Ground every abstraction with 3+ examples from different…
Connect each abstract concept in your knowledge system to at least three concrete examples from different domains, because single examples invite surface-feature overgeneralization while multiple examples force attention to shared structural patterns.