The irreducible epistemic atoms underlying the curriculum. 4,828 atoms across 8 types and 2 molecules
When formal and intuitive schemas disagree on a decision, investigate the disagreement for thirty minutes rather than defaulting to either—write what your gut is reacting to and test whether it reveals a pattern your formal criteria missed or a bias you haven't examined.
When forced to make a binary decision after spectrum-based deliberation, document the richer multi-dimensional signal alongside the binary outcome so future analysis can recover what the compression discarded.
Before committing to a category assignment in high-stakes decisions, explicitly name what actions that category triggers and what you would do differently if the item belonged to an adjacent category.
For each major domain where you make decisions, explicitly write down the single deepest assumption everything else depends on, then list 5-10 decisions that would change if that root were different to verify you've found an actual root.
After updating a belief, identify one downstream decision where the revised model produces a different recommendation than the old one to ensure the update becomes operational rather than merely verbal.
When schema conflict persists after examining evidence, build a conditional routing rule specifying the exact conditions under which each schema applies rather than attempting to pick a universal winner.
When measurement data shows stable satisfactory performance with no identifiable bottleneck, redirect optimization effort to a different system rather than continuing to optimize the current one.
Set holding periods for contradictions based on cascade depth: one week for low cascade, two to four weeks for medium cascade, one month or longer for high cascade.
Before aggregating data across subgroups, check whether the relationship holds within each subgroup independently, as aggregate patterns can reverse at the disaggregated level (Simpson's paradox).
Before attempting to resolve any persistent organizational tension, apply the problem-vs-polarity test: can new information or analysis make one side permanently win? If no, design oscillation management rather than searching for resolution.
Address contradictions at the strategic principle level rather than re-adjudicating them at each tactical decision point to prevent decision multiplication.
Design agents only for decisions that score high on frequency (recurring often), stability (same answer each time), and low individual stakes, because these three properties determine whether automation saves resources without introducing unacceptable risk.
Do not automate decisions where the outcome is genuinely different each instance even if the category recurs (interpersonal conflicts, creative problems, novel diagnoses), because automating decisions with genuine novelty produces rigidity disguised as efficiency.
When a decision situation represents a once-in-career strategic pivot rather than a routine recurrence, suspend your decision agent and return to full deliberation even if the agent would produce an answer.
After each decision agent activation, update the agent's criteria based on whether they produced a good outcome, treating the agent as a living heuristic that improves with each use rather than a permanent law.
Build nutrition agents at the meal preparation layer (Sunday evening meal prep) rather than the consumption layer (what to eat when hungry), because deciding what to eat while hungry and facing an open refrigerator is too late to override depletion.
When discovering that your designed agents conflict with each other, resolve the conflict through documented priority hierarchies rather than case-by-case deliberation, making the resolution rule itself part of your agent system.
Match the number of qualifying conditions to the cost of false positives—use minimal guards for low-cost triggers and multiple defensive checks for high-consequence triggers.
Before beginning deliberation on any decision, classify it as one-way door (irreversible/high-stakes) or two-way door (reversible/low-stakes) and allocate analysis time proportionally—minutes to hours for two-way doors, days to weeks for one-way doors.
For decisions involving three or more options and four or more criteria, externalize the comparison into a weighted decision matrix rather than relying on intuitive averaging, because working memory cannot hold all dimensions simultaneously.
When building a decision matrix, assign criterion weights before scoring any options to prevent unconscious adjustment of weights toward your pre-existing preference.
Score one criterion at a time across all options rather than one option at a time across all criteria, to force apples-to-apples comparison and reduce halo effects.
For each decision framework you build, explicitly include five components: evaluation criteria, sequence of evaluation, time budget, kill conditions (automatic disqualifiers), and decision rights (who decides, who is consulted).
For reversible decisions, act when you have 50-60% of desired information because experiential learning from outcomes typically exceeds information gain from additional deliberation.