Superforecasting for one person already demands decomposition, outside-first anchoring, and disciplined Bayes. Markets already aggregate many participants; dragonfly is how you approximate that aggregation before you add your capital—so your single bet is closer to a small committee than a monologue.
The failure mode is false diversity: five lenses that all hear the same podcast. Independence is a modeling assumption you must enforce, not hope for. Real markets aggregate many eyes—crowd wisdom, cross-venue consensus, dossiers that compare rules. Dragonfly eye is Tetlock's metaphor for seeing one target through many lenses before you lock a single probability. One lens is a trap; several weakly correlated lenses, aggregated with rules, are a forecast.
What the metaphor demands
Dragonflies combine partial images into sharp vision. Good Judgment teams used dissenting views, alternative reference classes, red-team narratives, and parallel model tracks. Solo traders can replicate the spirit: deliberately produce three to five independent estimates, then aggregate—mean, median, or weighted—before you trade.
Your favorite pollster alone is one lens. Poll plus fundamentals plus executable consensus plus outside base plus a steel-man NO case is dragonfly. Treating venue mid as the final word is one lens wearing a crown.
A five-lens template for binary contracts
Outside lens: reference class rate. Structural lens: polls, regressions, simulations. Market lens: executable consensus across venues. Red-team lens: best story for the other side. Skeptic lens: what if you are overconfident—shrink toward base.
Default to equal weights unless you have evidence one lens is systematically better for this domain—deeper liquidity might justify a higher market weight; a backtested model might earn extra weight on elections you have scored for a year.
Suppose lenses read forty, fifty-five, fifty-two, thirty-five, forty-eight percent with modest extra weight on the model. A weighted aggregate near forty-eight percent contrasts with a monolithic gut of sixty-two percent. Dragonfly often lowers overconfidence and saves calibration tails.
When lenses disagree wildly
If spread between highest and lowest exceeds twenty-five points, pause. Polls may be stale, manipulation may be in play, or resolution text may differ across venues you compared. Dragonfly discipline says do not trade until the gap has a named cause—not "I trust my guy."
Ten to twenty-five points suggests a missing factor: which lens lacks data? Under ten points, proceed to economics with normal caution.
Red team before size
Assign yourself—or a buddy—the job of arguing the opposite side with steel-man evidence. What resolution wording kills YES? What path wins for NO? What would consensus be if your thesis is wrong? Which lens is cheerleading your tribe?
Good Judgment aggregation beat individuals because errors were partially independent. Two traders who submit f before discussion, then median, often beat either alone—if they are not the same ideology in two accounts.
Independence is a choice
You enforce independence by sourcing lenses from different methods: history, model, market, adversarial story, personal shrink. Two poll-based lenses are one lens. A model and a poll are closer to two. Document correlation decisions in the journal so future you knows why weights were halved.
Worked example: climate bill
Event: landmark climate bill passes the House by mid-2026.
Outside historical base for similar bills might be eighteen percent. Whip-count structure thirty-two percent. Polymarket mid twenty-seven percent. Red team Senate block story twelve percent. Sentiment index twenty-two percent—likely correlated with whip count, so down-weight sentiment half versus structure. Recomputed aggregate near twenty-four percent versus thirty-cent YES suggests rich YES to fade if net EV clears fees and NO liquidity exists.
Correlated lenses dressed as five independent voices are one lens with makeup. Sentiment plus Twitter plus "vibes" counts once.
Fox habits versus hedgehog traps
Hedgehogs run one driver; foxes update several and synthesize. Narrative fallacy is building one story that eats every lens. Dragonfly is the fox habit operationalized.
Aggregation methods
Simple mean works when lenses are independent and you have no track record. Median helps when one zealot lens exists. Trimmed mean helps with five or more lenses. Weight market lens higher when liquidity is deep and rules match your dossier. No trade when spread is unexplained—capital preservation is a lens too.
Crowds work when errors are partially independent. Duplicate lenses wearing different hats are not a crowd.
Scoring multi-lens discipline
Primary score is Brier on the aggregate f you locked. Secondary review: which lens would have been best ex post? Adjust weights empirically after twenty or thirty resolved events—not because one loud lens felt smart.
Handoff from signals to trustworthy belief
You arrived in this module from mispricing hunts with candidates. You should leave with defensible probabilities: decomposed, anchored, updated, multi-lens, ready for calibration drills and daily habits. Calibration turns "I said sixty-two percent" into "sixty-two percent means what it claims."
Solo trader dragonfly session (thirty minutes)
Minutes zero to five: write resolution and pick the contract. Five to ten: fill outside, structural, and market lenses. Ten to fifteen: red-team NO with steel-man bullets. Fifteen to twenty: skeptic shrink and pre-mortem. Twenty to twenty-five: aggregate, note lens spread. Twenty-five to thirty: economics and lock f if you trade.
If you skip red team, cap size at probe levels. Full size requires full dragonfly.
When aggregation beats you alone
Track ex post which lens had best Brier over twenty events. Increase that lens weight slightly next quarter. Decrease weight on lenses that looked smart narratively but scored poorly. Empirical weights beat ego weights.
Common mistakes
Skipping red team because you are bullish. Letting market lens dominate because it is easy to read. Averaging lenses that share one data source. Trading wide lens spread because "someone must be wrong."
What comes next
Aggregation produces a number; calibration training asks whether your numbers mean what they say across many events.
Key ideas to carry forward
Five lenses, red team, down-weight correlation, aggregate, note spread, economics gate, lock f. Wide unexplained spread means pause.
Dragonfly is how you operationalize humility without collapsing to fifty-fifty everywhere. You still take stands—after synthesis, not after the first compelling lens.
Dragonfly is optional only in the sense that skipping it is optional if you accept worse Brier tails.
Treat lens spread as a risk gauge the way you treat bid-ask spread: wide unexplained spread means smaller size or no trade.
Next: Calibration Training: How Accurate Are Your Probabilities?