After you line up prices for one event across venues, the next question is blunt: what single probability belongs in your model? Consensus is not the mean of mids. It is a meta-forecast—a weighted blend you document before trading, then update with evidence using Bayesian habits from earlier modules.
Simple averages fail because the same traders appear in multiple apps, thin venues distort tails, and rule mismatches poison legs. Medians ignore liquidity. Raw volume weighting invites wash prints. Your goal is a reproducible recipe logged in the journal so future you knows how today’s π was built.
Weighting dimensions that matter
Rule match should dominate: a fatally different contract gets zero weight. Depth at your size matters next—executable price near mid for the clip you actually trade. Recent dollar volume signals informed flow versus one print. All-in fees change net edge. Historical calibration on similar contract classes is useful but easy to overfit. Recency matters more inside thirty days to resolution than a year out.
Illustrative recession blend: a regulated venue at 34¢ might earn large weight with tight rules and depth; a global venue at 38¢ slightly more if volume and recency are stronger; a capped retail 41¢ leg down-weighted; play money excluded entirely. Arithmetic might yield consensus near 36¢ in a toy example—your trade compares your updated view to that blend, not to any single app icon.
Write weights in the journal as plain language: “regulated 40%, global 45%, capped 15%, play 0%.” Future you should reconstruct π without guessing.
When to drop a leg
Exclude venues with fatal rule diff, books stale for a week without volume, open disputes freezing settlement, play-money markets, and temporary manipulation suspects after spike audits. Down-weight but do not delete legs with minor basis differences—note the bias direction in the dossier.
Bayesian workflow
Treat consensus as prior π before new information. List what your inside view adds—polls, models, primary reporting. Estimate how likely the evidence is if YES is true versus false, even qualitatively. Posterior p′ is what you trade. If jobs data pushes your unemployment contract posterior to 18¢ while the best ask is 21¢, fading can still be correct after a fast market move—the crowd updated π but you believe it lagged your posterior.
Poll-versus-market foundations remind you that markets aggregate paid beliefs; your update may include unpaid data—justify why it is not already priced.
Dispersion is its own signal
Low dispersion (a few cents) suggests arbitrage pressure worked—edge may live elsewhere. Medium dispersion may reflect segmented pools—let the leader venue anchor π until proven otherwise. High dispersion demands dossier repair: rules, caps, or stale books. Widening dispersion after quiet periods often precedes headline risk; narrowing after news may mean incorporation, not momentum license.
Track cross-venue standard deviation in the journal. Rising dispersion without thesis is a size reducer.
Multi-outcome slates
Nomination or policy slates need consensus per outcome, then check sums near 100% after fees. A single blended mid on “the nominee market” is meaningless—you need a vector of πᵢ for each candidate, then compare each to its ask. Normalize each venue before blending if one book shows 103% and another 99%—otherwise consensus inherits juice. Tree coherence material applies when sibling contracts must add up logically; rich and cheap legs may be the real trade, not the headline outcome.
Time decay and weights
Far from expiry, weights are stable. Inside thirty days, recency and depth should dominate. Inside seventy-two hours, a single leader venue may effectively be π for execution planning. Recompute weights weekly; election weights do not belong on a CPI week contract.
Building π in twenty minutes (conceptual)
Open the dossier. Drop failed legs. Score survivors on rule, depth, volume, fees, history, recency. Compute weighted π. Log dispersion and leader. Write inside p in one sentence. Bayes-update if fresh evidence today. Compare p′ to executable asks per venue. Route orders to cheapest expression of edge. After resolution, score calibration on π versus outcome.
Sensitivity and ethics
Stress-test π: drop the lowest-weight venue—if belief jumps several cents, your blend was fragile. Double a capped retail weight—if π lurches, caps dominated. Keep an independent p′; consensus informs, it does not replace accountability. Wash volume and whale ticks distort blends—cross-check open interest and use executable prices, not one print.
False consensus happens when every leg shares the same narrative Twitter feed but different caps—diversity of rules matters more than diversity of logos.
Simple blends and why they fail
A simple mean of mids treats a play-money quote equal to a regulated book. A median ignores that one outlier might be the only venue with depth. Volume-only weighting rewards wash and headline churn. Expert-only blends discard market information you paid nothing to ignore—unless you have verifiable edge in the expert layer.
The weighted recipe is more work; it is also what you can defend in a journal when a trade goes wrong.
Worked disagreement: jobs report beats whisper
Suppose weighted π on “unemployment above 4.5% by December” sits at 28¢ pre-release. A strong jobs print pushes risk-off; Kalshi moves 28¢ → 22¢ in three minutes. Your posterior is 18¢ because the print changes the path, not because you enjoy being contrarian. If the best ask is 21¢, you still should not buy YES—the market moved toward you but not far enough. Consensus updated; your p′ updated faster. The trade is in the gap between p′ and executable price, not in the gap between your ego and π.
Core concepts to remember
π is a tool, not truth. Weights must be written before clicks. Dispersion is signal. Exclusions are as important as inclusions. Bayes links outside data to p′. Calibration after resolution teaches whether your weights were wise.
Common misconceptions
“Averaging apps is democratic.” It is statistical noise. “π moved so I must trade.” Movement without edge is entertainment. “One venue is always right.” Leaders rotate by event class. “Consensus replaced my judgment.” It narrowed your prior—you still own p′.
After resolution: calibrate the blend
When the event settles, score whether π one day before was closer to truth than any single leg. Over years you learn which weights work for macro versus politics—data beats intuition. Adjust weights in writing; do not silently drift.
What comes next in this module
Consensus closes the multi-venue thread. Economic-indicator chapters place π beside traditional macro tools; sentiment chapters explain text-driven moves; the capstone mispricing chapter forces p′ to beat executable prices net of costs. Together they turn parallel quotes into one decision.
Document your weight formula once per event class—politics, macro, sports—and reuse it. Ad-hoc weights each trade are indistinguishable from storytelling.
When two venues disagree, ask which pool has the capital you respect for this topic before splitting the difference. If you cannot defend the weight split in one sentence, you are not ready to trade the blend.
Record π, p′, and dispersion in the journal on one line: π=36¢ | p′=42¢ | σ=4¢. Months of those rows teach more than any single chapter. Election night U.S. flow may deserve more weight on regulated books; token-policy contracts may deserve more on global venues. The answer is empirical—log outcomes, adjust weights.
What comes next
A disciplined π can sit beside traditional macro tools as a timely indicator—when rules and liquidity allow.
Next: Using Prediction Markets as Economic Indicators