The affect heuristic paints probability before you open a spreadsheet. Confirmation bias is what happens next: you hunt evidence that preserves the feeling and quietly discard the rest. In prediction markets this bias is especially expensive because the book already priced much of what you are about to read on social feeds.
This chapter deepens the confirmation row from trader psychology catalogs and connects to coherence fallacies, multi-lens review, and journaling that forces disconfirming sources.
Definition and market symptom
Confirmation bias means preferentially gathering, interpreting, and remembering information that supports a pre-existing belief while discounting disconfirming data.
At the search stage you follow only bullish analysts and miss NO edge. At interpretation you call an unfavorable poll “noise” while your probability stays high. At memory you remember hits and forget misses, inflating a private Brier score. At sharing you post wins and hide losses, feeding social proof for others.
You are not dumb—you are efficient at defending a thesis. Superforecasters institutionalize active open-mindedness; casual traders institutionalize likes. The book is a compressed argument that already swallowed much public information; confirmation makes you re-litigate the case with a one-sided brief.
Why “more research” can make it worse
Another hour on Twitter rarely fixes confirmation—it deepens it. Research quality beats research quantity: one primary document that could falsify your view outweighs twenty threads that cheer your side. Schedule disconfirm time the way you schedule earnings releases.
Confirmation versus affect and overconfidence
Affect starts with feeling; confirmation starts with the belief you want to keep. Overconfidence makes the defended belief too narrow and too large. Anchoring sticks to the first number that felt fair.
Confirmation often follows affect: you feel YES hope, then you only read YES paths. Fixing confirmation without naming affect is patching a leak upstream.
Why liquidity does not save you
Deep books tempt “the market is wrong because my thread is right.” Thin books tempt “nobody knows my secret.” Post-spike moves get read as validation instead of noise. Cross-venue disagreements get ignored when the venue you like confirms you.
Structured audit habits fail if “read news” means refresh only confirmatory feeds.
Election polling cherry-pick (worked example)
Contract: “Candidate J wins general” YES. You entered YES at 54¢ with thesis probability 62%.
You noticed a subsample poll plus three, a viral yard-sign video, and a follower who bought YES. You skipped the aggregate tracker sitting flat, mixed early-vote math, and Kalshi trading near 56¢ that you dismissed as dumb money.
Discipline: steel-man NO with turnout and incumbency base rates; blended probability falls near 57% against a 56¢ market—about one cent of edge, not worth size. Action: trim or exit; journal tag confirmation. Base-rate neglect often rides confirmation: one vivid poll beats a reference class.
Crypto ETF “approval inevitable” (worked example)
Contract: “ETF approved by date D” YES. Narrative community says 90%; market near 72¢.
Confirmatory stories stack: prior applications filed, politician tweet, issuer marketing. Disconfirmatory stories avoided: SEC delay history, court losses, rule ambiguity.
A dragonfly pass—legal, political, technical, market lenses—might place outside approval rates near 38% with capped inside bumps toward 58% against 72¢—sell YES edge. Confirmation cost: you held YES from 45¢ mentally at ninety percent and never re-read the NO resolution path.
Steel-man drill (operational, not inspirational)
Before material size, write “NO wins because…” in three bullets with specific mechanisms, not strawmen. Find the best counter source—a URL in the journal, not a dismissive meme. If NO wins, which likelihood moves, numerically? If you cannot write the bullets, no trade today.
Pair every thesis field with a counter-URL. Empty counter fields in weekly review are a smoking gun.
Disconfirm hunt before you trade
What would make you exit—write invalidation before entry. What is the strongest fact against you—if none found after ten more minutes of search, bias is likely. Does another venue disagree—read the comparison matrix before hero trades. Did probability ever move down on bad news—if never, confirmation is suspect. Are you explaining away misses—run a pre-mortem.
Feeds, AI, and portals
Algorithmic feeds optimize engagement, not truth. AI summaries sound confident without citations—demand sources. News portals that pair headlines with lessons still need both sides in your notes. Discord whale alerts are social proof until you verify size.
Product builders can suggest counter-lesson links in editorial workflow—a nudge, not censorship. Traders still own the click.
Bayesian update you refused (worked example)
Prior: 40% YES on a rate cut. Tier-one CPI prints hot.
Correct Bayes cuts odds toward 28% as the market drops from 35¢ to 30¢. Confirmation path says “hot print priced in” and adds YES because it feels cheap. Log the likelihood you used. If you cannot write probability of data given YES versus NO, you are narrating, not updating.
Tie-ins worth keeping
Team review that argues the opposite before averaging helps. Structured forecast routines treat review as disconfirm pass. Conjunction and narrative fallacies glue confirmatory stories. Weekly bias audits and journals with counter-links catch confirmation before overconfidence sizes it up.
Red flags
You muted accounts that disagree—feed as bunker. You say “data is wrong” without a source—protecting thesis. Edge exists only on an illiquid venue—picking a confirming price. Journal thesis lengthens—moving goalposts. You avoid Brier review because scores already know the disconfirming facts.
Weekly confirmation audit (twenty-five minutes)
Export recent journal rows. Highlight trades with empty counter-source fields. Spend eight minutes steel-manning each; note if edge flips. Count how often probability moved only on favorable news; if more than sixty percent, shave a few points off future calls until the next audit. Re-read the fallacy section that appeared most in your notes. If the same narrative appeared in three losing trades, ban that phrase from journal titles for a week—language shapes what you notice next.
Passes are evidence too
Confirmation also hides in passes you never logged. You skipped the NO trade because one bullish thread felt sufficient; later the contract resolved NO and you claim you “almost” traded it. Shadow logs and pass notes keep the sample honest for calibration. Survivorship without passes trains you to believe you are always on the right side of the forks you remember.
Contrarian confirmation (the mirror trap)
Fading the crowd can be confirmation inverted: you only read bear cases because you want to be the genius against π. The fix is the same—steel-man the side you dislike. Contrarians without counters are just hipster confirmation.
What comes next
Confirmation defends belief; anchoring defends the first number. The next chapter explains why opening prints, entry prices, and poll headlines stick—and how to re-estimate probability fresh before every decision.
Next: Anchoring: First Prices Stick in Your Mind