think-first

The goal: Never implement a significant decision without structured analysis first. Never lose specific constraints by abstracting them into vague qualifiers.

<essential_principles>

1. Detect Decision Moments

Decision moments include:

• Architectural choices ("should we use X or Y?")
• Business model decisions ("pricing, positioning, target market")
• Feature prioritization ("what to build first?")
• Trade-off analysis ("speed vs quality, scope vs time")
• Strategy questions ("how should we approach this?")
• Pivots or direction changes ("should we change course?")

1b. Randomness Audit (Taleb) -- BEFORE trusting ANY outcome evidence

Before evaluating options or trusting evidence, ask whether the evidence itself is contaminated by randomness. This is the skeptic's corrective to both Klein's intuition-trusting (RPD) and Kahneman's structured judgment -- even structured judgment fails if the input data is noise masquerading as signal.

Check	Question	If YES
Survivorship bias	Am I only seeing the winners? What do the failures I cannot see look like?	Discount the evidence. Seek disconfirming cases. Ask "what happened to the ones who tried this and failed?"
Narrative fallacy	Am I constructing a tidy story to explain an outcome that randomness explains equally well?	Strip the narrative. Look at base rates. Would this outcome be surprising if I had no story?
Attribution bias	Am I attributing this outcome to skill when it could be luck?	Apply the "alternative histories" test: in how many of the possible paths would this outcome NOT have occurred despite the same skill?
Sample path overfitting	Is this strategy successful only because the specific sequence of events happened to be benign? How long since a rare event tested it?	The longer the calm, the more suspicious. Strategies untested by adversity are fragile, not proven.
Observation frequency trap	Am I checking data too often? (The more frequently you check, the more noise vs signal you see.)	Reduce observation frequency. Taleb: over 1 year, ~0.7 noise per 1 signal. Over 1 hour, almost entirely noise.
Domain type	Is this a "dentist" domain (steady, low-variance, skill dominates) or a "trader" domain (high-variance, luck dominates)?	In trader domains, be deeply skeptical of all skill claims. In dentist domains, trust track records.

The key discipline: "It is more random than we think" -- not "it is all random." The question is always whether you have enough evidence to distinguish skill from luck in THIS specific case.

Ergodicity check (for strategy decisions): Does this strategy that looks good across many people at one point in time also survive for one person across time? Ensemble averages (many traders, one moment) differ from time averages (one trader, many moments). A strategy with positive expected value can still ruin the individual if it has ruin risk on any single path.

Reference: 05 Library/Reference/Fooled by Randomness - Application Guide.md

1c. Noise Check (Kahneman/Sibony/Sunstein) -- BEFORE routing to RPD

Before trusting intuition, assess whether this domain is HIGH-NOISE. Klein shows where intuition works; Kahneman shows where it systematically fails; Taleb (1b above) shows where even the data itself is unreliable. High-noise domains require structured judgment, not recognition-primed decisions.

Signal	Noise Level	Implication
Multiple qualified people would give substantially different answers	HIGH NOISE	Do NOT use RPD. Use decision hygiene: decompose, get independent assessments, aggregate.
Judgment depends on mood, time of day, recent cases, or order of information	OCCASION NOISE	Add structure to counter. Sequence information carefully. Delay holistic judgment.
The same person would give a different answer on a different day	OCCASION NOISE	Seek a second opinion (even from yourself at a different time -- "the crowd within").
Domain has clear feedback loops, repeated experience, regular patterns	LOW NOISE	RPD applies. Proceed to 1c below.
Domain has long feedback delays, rare events, complex causation	HIGH NOISE	Use algorithms or structured protocols instead of intuition. Simple rules beat experts in predictive tasks.