Platform Insights

Venue-wide Polymarket activity, trader outcomes, bot participation, and resolution analytics as indexed by Polyanna.

Overview#

The Platform Insights page aggregates the same indexed fill history used across leaderboards and trader profiles. It is designed to answer platform-level questions: how much activity has happened, how traders are distributed by PnL, how much activity appears automated, and how resolved markets behaved.

Volume, Fees & Activity#

Total Volume sums indexed USD trading volume across fills. Orders Filled counts the number of indexed fill events. Total Fees sums estimated protocol fees paid across those fills after converting them to USD. These are activity metrics, not profitability metrics, so a month can have high volume and high fees even if traders lose money overall.

PnL & Profitability#

Platform Insights reuses the same PnL and win-rate definitions used across trader profiles and leaderboards, but it aggregates them across the trader population:

• Median PnL is the middle trader PnL in the all-time distribution.
• Top Trader PnL and Worst Trader PnL are the highest and lowest total trader PnL values in the current analytics snapshot.
• Avg Win Rate is the arithmetic mean of each trader's win rate. It is not weighted by volume, trade count, or account size.

Winners & Losers#

Winners are traders with positive total PnL. Losers are traders with negative total PnL. Break-even wallets are excluded from both groups. Total Won sums positive PnL across profitable traders, while Total Lost sums the absolute value of negative PnL across losing traders. This section is about trader outcomes, not the count of individually winning trades.

PnL Distribution#

The distribution chart bins every trader by total all-time PnL into seventeen decade-scaled buckets, symmetric around a ±$1 break-even band. Buckets range from < -$10M through -$10 to -$1, ±$1, $1 to $10, up to > $10M. Bucket edges are left-inclusive; a trader on an edge lands in the bucket starting at that edge.

Each row shows how many traders fell into that bucket and the sum of their PnL. The All traders toggle shows the full population; the Bots only toggle restricts the view to wallets with bot score at or above Polyanna's display threshold. Because the population is power-law distributed, tail buckets with only a handful of wallets often account for the bulk of the money in or out of the platform.

Market Coverage#

Markets Tracked counts distinct markets with at least one indexed trade. Avg Markets / Trader is the arithmetic mean of distinct market counts across traders in the all-time base. It gives a quick sense of whether the population is more concentrated in a few markets or spread broadly across the venue.

Monthly Activity#

Monthly Trades counts indexed fill events per calendar month. Monthly Events counts distinct events with at least one indexed trade in that month, grouped by event identifier so multi-market events count once rather than once per market. Monthly Markets uses the same “at least one indexed trade this month” rule but counts each underlying market separately, so a three-way election contributes once to Monthly Events and three times to Monthly Markets. The ratio tells you how much of each month's activity comes from multi-market events. The Monthly Events tooltip's Top event label is the busiest event in that month by indexed fill count.

Bot Participation#

Bot Volume Share measures what fraction of total platform USD volume came from wallets that meet Polyanna's bot filter threshold. Bot Trade Share measures the same concept but counts filled orders instead of USD volume, which is useful when bots execute many small trades that contribute little volume but dominate the fill count. Both are shares of activity, not shares of wallets. The underlying bot score methodology is documented in Bot Detection.

Resolution Analytics#

Resolution Analytics looks at every binary market that settled on Polymarket and asks whether the crowd got it right. A market qualifies if it has a final YES or NO outcome, is not part of a multi-outcome event, and recorded at least 20 indexed fills. Voided markets are reported separately and excluded from the YES/NO split. Multi-outcome markets are excluded because their routing means the on-tape midpoint is not directly comparable to a clean two-outcome binary market.

This section is auth-gated. Sign in with a free account to unlock the calibration curve, biggest upsets, and crypto up/down market cohorts.

Favorite Win Rate#

For every qualifying market we take the time midpoint between its first and last observed fill and evaluate two independent crowd-belief signals:

• Price favorite — the side priced strictly above 50% at the midpoint. This is the classic midpoint-based signal.
• Midpoint volume — the side with more cumulative USD volume before the midpoint. Measures whether money-weighted conviction predicts outcomes better than price alone.

The Favorite Win Rate for each signal is the fraction of markets where that signal's favorite matched the eventual winner. The toggle in the resolution analytics section lets you switch between both signals in the monthly trend chart, while the summary card shows both side by side.

We deliberately avoid the last-trade price at market close: by then the outcome is often visible off-chain and prices collapse toward 0 or 1 in an end-of-market information cascade. Using that contaminated signal would overstate how well the market "predicted" the outcome. The midpoint snapshot is earlier, less biased, and closer to what a trader entering in the middle of the market would have seen.

Calibration Curve#

We bucket every resolved market by its midpoint YES price into 10 bins of 10% width each and compute the fraction of markets in that bin that actually resolved YES. Plotting those points gives the calibration curve, with the dashed diagonal (y = x) representing a perfectly calibrated market.

• If the curve sits above y = x in a bin, YES prices in that range systematically under-estimated the true YES probability. The crowd was too bearish on YES.
• If the curve sits below y = x, the crowd was too bullish on YES in that range. YES prices over-estimated the true YES probability.
• A curve that hugs the diagonal says the market is well-calibrated on average, even if individual markets were surprising.

Biggest Surprises#

Resolved markets where the underdog won, ranked by surprise cost = (1 − winner midpoint price) × total volume. Both factors matter: a high-volume near-coin-flip carries more aggregate loss than a tiny market priced at a fraction of a percent, so ranking by lowest price alone surfaced noise. Each row links through to the underlying event on Polymarket when a slug is available.

Crypto Up/Down Cohorts#

Polymarket hosts a large family of short-horizon crypto markets that resolve based on whether an asset is up or down across a fixed window. We group these by asset and interval.

Each row in the cohort table shows Up %, Down %, Void %, the same midpoint-based favorite-win rate, and the average midpoint price of the eventual winner, giving a compact readout of how predictable each cohort was. Small cohorts are hidden to avoid surfacing noisy samples.

YES / NO / Void Share (Monthly)#

Stacked area chart of YES / NO / Void share across every resolved binary market, bucketed by the month of its last fill timestamp as a close-time proxy. The mix is not stable. Months with heavy election activity can skew toward NO because many binary political questions end up unresolved as-written.

Crowd Accuracy Trend#

Monthly favorite-win rate shows how often the side priced above 50% at midpoint actually won. Anything above 50% means the crowd carried real information that month. Months with too few eligible markets are suppressed so early-dataset noise does not swing the line.

Crypto 24h Distribution#

For short-horizon crypto up/down cohorts, we bucket each resolved market by the UTC hour of its last fill and compute the share that resolved UP at that hour. The asset selector lets you switch between assets or an aggregate view. A flat distribution means timing-of-day carries no predictive edge over the cohort's baseline; a tilted one suggests systematic structure, such as the impact of US market open on BTC or session-based mean reversion.

Data Quality#

Platform Insights uses the same indexed fill base and validation flow as the broader analytics system. Polyanna cross-checks trader activity against Polymarket's public data and verifies that refreshed platform metrics are derived from the same validated activity.