The irreducible epistemic atoms underlying the curriculum. 4,828 atoms across 8 types and 2 molecules
Accumulate post-action reviews in a single searchable location to enable pattern recognition of systemic errors that recur across multiple tasks.
Convert every identified gap between expectation and outcome into a concrete process change, not a resolution to try harder or be more careful.
Use structured questioning protocols to counteract memory reconstruction biases that distort unstructured reflection into self-serving narratives.
When designing checklists, include items that trigger observation of actual state rather than retrieval of intention, converting 'Did I do X?' into 'Is X currently true?'
Implement each identified root cause as an encoded structural mechanism—a process step, default setting, checklist item, or automated check—that prevents the error class from recurring.
Break long dependency chains into shorter parallel paths to prevent error propagation.
Insert independent verification checkpoints at every point where one process automatically consumes the output of another without review.
Define kill conditions before beginning any multi-step process to enable early detection of upstream errors.
Design explicit operating modes (full, reduced, minimal) for every critical system before constraint forces improvisation.
Distribute critical functions across multiple redundant mechanisms at different scales rather than relying on single points of function.
Practice running systems in degraded mode during normal conditions to rehearse partial failure before real constraints force it.
Design recovery procedures to function under the degraded conditions that typically accompany failures, not ideal conditions.
Implement cognitive checkpointing by capturing the complete state of ongoing thinking at regular intervals to enable rollback without full reconstruction.
When investigating errors, ask 'what conditions made this failure likely?' rather than 'who is responsible?' to shift analysis from person to system.
Commit to structural changes (process, checklist, information flow, defaults) rather than personal resolutions as the test of learning from error.
Group errors into clusters and treat three or more instances of the same pattern as a signal of structural weakness rather than coincidence.
For skill-based errors, fix the environment and triggers; for rule-based errors, improve situation discrimination; for knowledge-based errors, update mental models.
Deploy automated detection for high-frequency, pattern-based errors and reserve human judgment for contextual, judgment-dependent errors.
Externalize vigilance requirements to tools or protocols rather than relying on sustained human attention.
Catch errors at the earliest point in a process where detection is possible, as correction cost increases exponentially with propagation distance.
Track the full cost of error correction including direct cost, opportunity cost, context-switching cost, and propagation cost, not just the visible effort of fixing.
Invest in preventing error-producing conditions rather than optimizing correction speed when total correction cost is high.
Extract structural information from errors by separating mechanism from emotion and identifying the incorrect assumption that produced the error.
Treat errors as high-information signals that reveal system boundaries and assumptions, allocating attention disproportionately to failures rather than successes.