Principles
2,201 principles · page 34 of 46
Track classification decisions as hypotheses with version…
Track classification decisions as hypotheses with version history rather than as immutable facts, enabling systematic improvement of category definitions over time.
Layer facets, tags, or cross-references on top…
Layer facets, tags, or cross-references on top of hierarchical taxonomies to handle items that legitimately span multiple branches, using hierarchy for primary navigation and facets for multi-dimensional access.
Assign types to fields and objects to restrict valid inputs…
Assign types to fields and objects to restrict valid inputs and operations before execution, making entire categories of error structurally impossible rather than detectable after the fact.
Track status as a state within a defined lifecycle rather…
Track status as a state within a defined lifecycle rather than as a categorical property, defining valid state transitions and triggers to convert status tracking into executable workflow.
Type items by priority during batch classification sessions…
Type items by priority during batch classification sessions rather than at moment of action, converting continuous judgment into categorical lookup that preserves cognitive resources for execution.
Define priority types with one-sentence decision rules…
Define priority types with one-sentence decision rules that specify membership criteria, not descriptions, ensuring that classification can occur without deliberation about tier assignment.
Work through priority types sequentially rather…
Work through priority types sequentially rather than scanning the full list, consulting only the highest-priority tier that contains items and never reviewing lower tiers until higher tiers are empty.
Type along both urgency and importance dimensions…
Type along both urgency and importance dimensions simultaneously to counteract the mere urgency effect, creating explicit categories for important-but-not-urgent items that would otherwise be chronically deprioritized.
Balance role type distribution across teams rather…
Balance role type distribution across teams rather than optimizing for individual capability, because team performance depends on covering necessary functions not maximizing any single dimension.
Audit classification systems for duplicates, dead…
Audit classification systems for duplicates, dead categories, overstuffed catch-alls, and ambiguous labels on a regular schedule to prevent compound interest on classification debt.
Audit your classification system by identifying what's…
Audit your classification system by identifying what's prominent (operational values), what's absent (neglected values), and what's miscellaneous (blind spots), because category structure reveals value structure.
For high-stakes classifications, explicitly name adjacent…
For high-stakes classifications, explicitly name adjacent categories and assess asymmetric costs of each error direction before committing to action protocols.
Set review triggers for important…
Set review triggers for important classifications—time-based, evidence-based, or outcome-based—because categories that persist without verification inherit compounding error from initial misclassification.
Calculate your actual prediction accuracy across documented…
Calculate your actual prediction accuracy across documented tests rather than relying on felt sense of how often you're right, because subjective confidence systematically exceeds objective accuracy.
Reclassify when boundary cases become the majority, when two
Reclassify when boundary cases become the majority, when two categories collapse into functional synonyms, or when a single category becomes an undifferentiated catchall, because these signal that current categories no longer carve reality at its joints.
Add categories for values you hold but haven't…
Add categories for values you hold but haven't operationalized, and remove categories that exist by inertia, because classification systems calcify around past values and make un-encoded values invisible.
When using AI tools that classify on your behalf, explicitly
When using AI tools that classify on your behalf, explicitly examine whose values are embedded in the AI's classification categories, because automated classification at scale makes embedded values operationally invisible.
Design categories around prototypical central examples…
Design categories around prototypical central examples rather than exhaustive definitional rules, as prototype-based classification matches cognitive processing speed and flexibility better than rule-checking.
Test category systems with boundary cases that resist clean…
Test category systems with boundary cases that resist clean classification, as items at category edges reveal missing dimensions, vague boundaries, and incorrect framing more reliably than central members.
Log items that resist classification rather than forcing…
Log items that resist classification rather than forcing them into nearest categories, as accumulation of misfits reveals systematic problems in category structure that individual cases obscure.
Decompose classification into independent facets when items…
Decompose classification into independent facets when items genuinely belong to multiple categories simultaneously, rather than forcing single-parent assignment.
Test each candidate classification dimension by asking…
Test each candidate classification dimension by asking whether it enables a question you actually ask and cannot currently answer - dimensions that don't resolve real queries add maintenance cost without retrieval benefit.
Audit what your classification system discards by explicitly
Audit what your classification system discards by explicitly listing the information lost in each category compression, then check whether any discarded dimensions matter for decisions the system supports.
Adjust classification granularity when categories either…
Adjust classification granularity when categories either collapse distinctions needed for decisions (over-compression) or preserve distinctions never used for decisions (under-compression).
Evolve classification systems through specific operations…
Evolve classification systems through specific operations (splitting, merging, renaming, promoting, retiring) triggered by measurable signals (category size imbalance, unused categories, hesitation in classification).
Treat classification systems as living infrastructure…
Treat classification systems as living infrastructure that evolves rather than finished products, recognizing that domains change, understanding deepens, and purposes shift over time.
When classification purpose shifts (from organizing by topic
When classification purpose shifts (from organizing by topic to organizing by project, or from product area to customer segment), redesign the primary organizational dimension rather than forcing the old structure to answer new questions.
Map relationships between concepts as explicitly as you map…
Map relationships between concepts as explicitly as you map the concepts themselves, because insight emerges from the connections rather than from individual nodes.
Externalize assumed relationships by writing them down…
Externalize assumed relationships by writing them down with explicit entities, relationship type, and evidence basis, because implicit connections cannot be examined, tested, or corrected.
State the relationship type (causal, correlational…
State the relationship type (causal, correlational, temporal, enabling, inhibiting) separately from the entities being related, because different relationship types license different inference patterns and conflating them produces systematic errors.
Classify your evidence for each relationship as verified…
Classify your evidence for each relationship as verified, plausible, or assumed, and prioritize testing the assumed relationships that drive consequential decisions.
Use the swap test for every relationship
if reversing the direction changes the meaning, the relationship is directed and must preserve the arrow in your representation.
When a relationship appears bidirectional, map it as two…
When a relationship appears bidirectional, map it as two separate directed arrows with potentially different strengths and mechanisms rather than collapsing it into a single undirected connection.
For causal relationships, establish direction through…
For causal relationships, establish direction through intervention or counterfactual reasoning rather than correlation alone, because observational data cannot distinguish causation from association.
Invest maintenance effort disproportionately in weak ties…
Invest maintenance effort disproportionately in weak ties that bridge clusters, because they provide structural value but are most vulnerable to decay through neglect.
For knowledge relationships, separate epistemic confidence…
For knowledge relationships, separate epistemic confidence (evidence quality) from identity centrality (emotional importance) when scoring strength, because these dimensions often diverge and conflating them corrupts decision-making.
Break learning goals into explicit prerequisite chains…
Break learning goals into explicit prerequisite chains before attempting them, working bottom-up from verified fundamentals rather than top-down from aspirations.
When learning fails repeatedly despite effort, trace…
When learning fails repeatedly despite effort, trace backward through prerequisite chains to find the missing foundation rather than pushing harder on the target skill.
Identify and invest disproportionately in enabling…
Identify and invest disproportionately in enabling conditions that create cascading improvements across multiple downstream outcomes rather than targeting each outcome independently.
Validate enabling relationships by articulating the specific
Validate enabling relationships by articulating the specific mechanism through which one condition creates another, not just their co-occurrence.
Surface and explicitly label contradictions between beliefs…
Surface and explicitly label contradictions between beliefs rather than resolving them prematurely, treating the tension as data about incomplete frameworks.
Map the boundary conditions under which each side…
Map the boundary conditions under which each side of a contradiction holds true rather than attempting to determine which side is universally correct.
Before attempting to resolve any contradiction, construct…
Before attempting to resolve any contradiction, construct the strongest possible argument for each side in a form that advocates of each position would recognize as their own.
Verify that supporting evidence comes from genuinely…
Verify that supporting evidence comes from genuinely independent sources by tracing each back to its origin and checking for shared roots.
Ground every abstract concept in at least three concrete…
Ground every abstract concept in at least three concrete examples from different domains to enable pattern recognition and transfer.
Sequence instruction from concrete enactive experience…
Sequence instruction from concrete enactive experience through iconic representation to abstract symbolic notation rather than starting with abstractions.
Test understanding of abstractions by generating…
Test understanding of abstractions by generating explanations using only concrete examples without definitions or jargon.
Re-ground abstractions periodically with fresh concrete…
Re-ground abstractions periodically with fresh concrete examples from recent experience to prevent drift between symbols and referents.