The irreducible epistemic atoms underlying the curriculum. 4,828 atoms across 8 types and 2 molecules
Actively search for conditions under which your schema would fail and test those conditions specifically, rather than testing only where success is likely.
Define in advance what evidence would falsify your schema and commit to that standard before collecting data, preventing post-hoc rationalization of ambiguous results.
Structure your knowledge system to preserve and surface contradictions, counterarguments, and disconfirming evidence rather than curating only supporting material.
Attack your highest-confidence schemas most aggressively, because confidence indicates attachment that blinds you to flaws.
Shift validation questions from future-conditional ('what could go wrong?') to past-definite ('it failed—why?') to bypass motivated reasoning.
Allocate validation effort by multiplying probability of being wrong by impact of being wrong, not by treating all uncertainties equally.
Generate multiple independent observable consequences of a schema, then evaluate whether those consequences converge or diverge across different evidence types.
When direct testing is impossible, look for the schema that explains the most diverse types of observations under a single mechanism.
Validate important self-schemas using at least three independent evidence types: behavioral observation, external feedback, and outcome tracking.
When one piece of evidence contradicts four confirming pieces, treat the divergence as diagnostic of either measurement error or a boundary condition, not as noise to dismiss.
Choose schema reviewers who think differently from you, not those who share your framework, because divergent perspectives surface blind spots that similar perspectives cannot see.
Present schemas for review using three diagnostic questions: What assumption am I missing? What would disconfirm this? What alternative explains the same observations?
Receive schema feedback without defending for at least 24 hours, because immediate response activates argumentation mode rather than evaluation mode.
Document validation results immediately and contemporaneously, before hindsight bias rewrites your memory of what you predicted or expected.
Record five components for each validation: the schema before testing, the test performed, what you predicted, what actually happened, and what it means for the schema.
Write validation records in specific observational language ('she pushed back on three of five points') not interpretive language ('she mostly agreed'), to prevent retroactive interpretation drift.
Review validation logs periodically to identify meta-patterns—systematic tendencies in how your schemas fail—rather than treating each validation as an isolated event.
Document disconfirmations more carefully than confirmations, because confirmations are naturally retained in memory while disconfirming evidence is filtered out.
Document the specific conditions under which a schema was validated, as the boundary of tested conditions defines the boundary of warranted confidence.
Reformulate schemas with explicit boundary clauses that separate validated conditions from untested extrapolations.
When transferring a schema to a new domain, treat the transfer as a new hypothesis requiring independent validation rather than an extension of existing validation.
Treat scale transitions as requiring revalidation, as schemas validated at one scale often encounter emergent properties at different scales that invalidate their predictions.
Distinguish epistemic confidence (grounded in evidence) from psychological confidence (grounded in identity and familiarity) by requiring an external validation trail for the former.
Test schemas where you feel most certain first, as high confidence without testing history signals the highest risk of unwarranted certainty.