Decomposition asks you to split base rates from specifics. The deepest tension in that work is outside view versus inside view—terms from planning research (Kahneman and Lovallo) that superforecasting popularized for forecasters who must publish numbers, not adjectives.
Most retail blowups are inside-view stories told too loud: a clip, a leak, a thread. Outside view is the discipline of asking "how often does this happen in the class?" before the story seduces you. Neither view alone is enough; incoherent blending is worse than either alone.—language from planning research popularized in superforecasting as the fight between "what usually happens" and "what is special this time." Prediction-market losses often come from picking the wrong view for the question, or blending them without a paper trail.
Two ways of seeing the same contract
The outside view is statistical cousinhood: how often do incumbents win in this band, how often does the Court grant cert in this posture, how often does the Fed cut when inflation is above three percent? You are borrowing history from a reference class.
The inside view is causal narrative: this candidate's ground game, this trial's biomarker readout, this CEO's credibility after the earnings call. You are building a story about mechanisms.
Superforecasters start outside, then adjust inside with logged deltas. Weak traders start inside and never anchor; vivid anecdotes beat base rates even when the class is large and stable.
The planning fallacy, trading edition
Teams underestimate project completion time because they focus on inside plans, not outside distributions. Traders underestimate reversal risk because they focus on thesis slides, not class history. Politics traps sound like "but our polls are great" while open-seat base rates say otherwise. Biotech traps sound like "but the p-value looked amazing" while phase III success rates say otherwise. Macro traps sound like "but the chair sounded dovish" while cut frequency in this inflation regime says otherwise.
Time decay on a contract is outside-view discipline about when uncertainty collapses; price spikes are often inside-view traders reacting to one headline without checking whether the reference class moved.
Pre-mortem from the outside
Before sizing YES, finish the sentence: "If the outside view is right, I lose because…" If you cannot complete it, you have not anchored. Pre-mortems are outside-view empathy for your future self.
Worked example: certiorari
Contract: "Supreme Court grants cert in Case Z" YES.
Outside only: historical grant rate for similar posture, about twelve percent. Inside only: "circuit split plus SG brief" story pushing forty-five percent. A blended superforecaster might land near twenty-two percent: outside anchor plus roughly ten points of inside lift, not a 3.5× multiplication.
If market YES is twenty-eight cents, inside-only traders buy YES with enthusiasm; outside-first traders lean NO or pass unless tier-one evidence supports a large likelihood shift. If the Court grants cert, Brier punishes the forty-five percent forecast more than the twenty-two percent forecast—even if both lost money on a NO position sized too large.
When to weight outside versus inside
Favor the outside view when the reference class has dozens of clean cases, the narrative is not structurally novel, and long-run rates are stable. Favor the inside view when history offers few parallels, you have fresh verified micro-data, or the market is thin and emotional while your process reading of resolution text is unusually sharp.
Treat executable market price as one outside input—a crowd reference class—with bias from liquidity, jurisdiction, and who is allowed to trade. Mispricing is your justified blend differing from executable price, not your inside story differing from cable news.
If an inside adjustment would more than double the outside base without new verified data, pause and run the multi-perspective aggregation chapter before sizing.
Worked example: IPO timing
Contract: "Company A IPO before 2026-Q3" YES.
Outside: tech IPOs filed and profitable within eighteen months might be thirty-five percent. Inside: SEC comment-letter delay subtracts ten points; lead underwriter reputation adds five. Raw blend near thirty percent, shrunk toward market forty cents to thirty-three percent when factors are noisy. Inside-only "brand is hot" might say sixty percent; outside-first discipline lands near consensus with smaller, more Brier-durable bets.
Worked example: injury report on a thin book
Contract: "Star player plays in playoff game" YES. Market seventy-one cents on thin liquidity.
Outside: starters in this injury classification historically play about fifty-eight percent. Inside: morning skate rumor pushes eighty-two percent from one source. Blended with cap on inside bump might land sixty-four percent—lean NO versus ask if economics work, because vivid inside evidence without tier-one confirmation is a classic trap.
Foxes, hedgehogs, and narrative immunity
Hedgehogs run one big theory across every cycle. Foxes hold several partial models and revise. Narrative fallacy is the hedgehog tax: one story eats every lens. The chapter on why experts fail names the public version; the discipline here is private—argue the outside view alone for five minutes on an open position and see if the trade still makes sense.
Blend errors to avoid
Double inside counts the same driver three times under different names. Fake outside cherry-picks a reference class that flatters the trade. Market worship treats consensus as infallible. Frozen outside never updates when tier-one evidence arrives. "This time is different" without data is inside immunity, not insight.
Another subtle error is anchoring to the wrong outside. An incumbent senator is not the same class as an open-seat challenger; picking the flattering class is fake outside view dressed as statistics. Write the class you used and why.
Reconciliation when views fight
When outside and inside disagree by more than twenty points, treat that as a signal to slow down, not to average blindly. List what would have to be true for the inside view to dominate—verified data, not vibes. List what would have to be true for the outside view to dominate—large reference class, stable history. Only then pick a blend and log it.
Market price as outside view (careful)
Consensus mid is a crowd outside view with its own biases: who can trade, how deep the book is, whether resolution matches the story traders think they are pricing. Use it as one input in decomposition and dragonfly work, not as a substitute for your own outside class when the class is well defined.
Mispricing means your blend beats executable price for documented reasons—not that your inside narrative is louder than CNBC.
Inside view under time pressure
Before a major catalyst, write the outside anchor first. During the window, allow large inside moves only on verifiable tier-one facts. After the catalyst, compare the tape to the f you pre-registered; resist rewriting both to match price if your likelihood table does not justify it.
Event clocks and pre-announcement discipline
Before a debate or FOMC, lock an outside anchor in the journal. During the event, allow inside updates only from tier-one sources with URLs. Afterward, compare price change to pre-registered f rather than rewriting both to match the tape. That rhythm keeps inside view from becoming price-chasing with extra steps.
What comes next
Outside view is the anchor; inside view is the adjustment. The next chapter turns belief revision into a repeatable loop: priors, evidence tiers, likelihood ratios, and rules that stop you from rewriting forecasts to match price without news.
Key ideas to carry forward
Outside view is the anchor; inside view is the adjustment. Cap inside bumps without tier-one evidence. Wrong reference class is fake outside view. Log the blend.
When in doubt, write two forecasts: outside-only and inside-only, then blend with caps. The exercise feels slow; it is cheaper than a five-point calibration error taxed across dozens of trades.
Outside–inside discipline is the emotional core of superforecasting: humility about your story, respect for history, honesty about what is novel.
Next: Updating Probabilities: Bayesian Thinking in Practice