How we compute our metrics

Data & snapshots

Every figure on this site comes from Polydata’s own indexed copy of Polymarket’s public on-chain activity — trades, positions, resolutions — updated continuously since the start of the platform. We do not use third-party aggregators.

Each article states its data snapshot time (UTC) in the source note at the bottom. Markets move after we publish; the numbers in an article are correct as of its snapshot, not as of your reading time.

Whale trades & flow

A whale move is a single trade of at least $10,000 within the last 24 hours. A matched trade has a buyer and a seller; we collapse the two sides so one fill is never counted twice.

A single trade becomes a story only from $250,000 — and only when the data offers context beyond size: the market-implied probability at the time, the wallet’s other trades in the window, or the trade being the largest of the day.

Whale flow for a market is the sum of whale buys versus whale sells over 24 hours. When we say a trade “joins” or “runs against” the broader whale flow, that is the comparison being made — direction of the trade versus the sign of the market’s net flow.

Wallet profiles quoted in whale stories (accuracy, leaderboard rank, lifetime volume, days of activity) are computed from that wallet’s full public history using the same definitions described on this page. Addresses and usernames are public on-chain data.

Smart-money divergence

We flag a divergence when net whale flow into one side of a market reaches $150,000 over 24 hours while the market prices that side at least 10 percentage points away from even. Big money accumulating an outcome the order book calls unlikely is the signal — not a prediction that the whales are right.

Labels like “against the consensus” or “contrarian” are used only when this computed condition holds, never as editorial colour.

Realized PnL

Realized PnL is computed cash-basis over resolved markets only. Open positions are excluded entirely — unrealized gains do not count. This makes our figures deliberately conservative: a wallet holding a large winning position will show a smaller PnL here than in the Polymarket UI until that market resolves.

Leaderboard ranks quoted in articles (“#10 on the 30-day leaderboard”) are by realized PnL over the stated window, among all wallets active in it.

Accuracy & Brier score

Accuracy is the share of a wallet’s resolved positions where the backed outcome actually happened. Brier score measures forecast quality: the average squared gap between stated confidence and the real outcome — 0 is perfect, lower is better.

Both are computed on resolved markets only, with a minimum of 20 resolved positions per wallet. The raw top of any accuracy board is dominated by micro-stake accounts with trivially perfect records, so before we feature a “top forecaster” we additionally require real volume (at least $10,000) and at least 50 resolved positions.

Calibration, where shown, compares stated probability with realized outcome rate: a perfectly calibrated forecaster sits on the diagonal.

The PMX index

PMX is Polydata’s proprietary sentiment composite for prediction markets on a scale where 50 = neutral; higher means more active or risk-on. It aggregates five sub-indices:

• PMX-Volume — trading volume versus its 30-day norm;
• PMX-Whale — the balance of whale buying versus selling;
• PMX-Politics — activity in political markets;
• PMX-Crypto — activity in crypto markets;
• PMX-Sentiment — the average market-implied probability across active markets.

Readings at or below 35 we describe as risk-off territory; at or above 65 as running hot. A PMX story is published only when a sub-index reaches an N-day extreme against its own daily history — a drifting composite with nothing extreme is not news.

Weather markets

For every city with active weather markets we track the money (total volume across its brackets) and the forecast: today’s high as predicted by multiple independent sources. The spread is the gap between the highest and lowest source — wide disagreement between models is exactly what makes a weather market tradable.

The daily roundup covers all cities with non-zero volume; its table lists the largest by volume plus the city with the widest forecast split. Where available, the previous day’s actual reading is shown to close the loop between forecast and reality.

What becomes a story

Coverage is event-driven. Beyond whales and divergence, the thresholds are:

• Volume spike — at least $5M of new volume in a market since our previous pass;
• Price break — a market on Polymarket’s Breaking feed with at least 15 percentage points of YES-price move in 24 hours and $5K+ in 24-hour volume, enriched with Polydata whale flow. Up to two movers per pipeline cycle (~every 10 minutes); the same market is republished only when YES price, the 24h delta, or whale net flow shifts materially since the last story;
• New market — a newly listed market that has already drawn $100K+;
• Category spotlight — the largest market of a category, minimum $5M volume;
• Resolution — a settled market with at least $1M traded;
• Daily recap — the last complete day versus the prior multi-day average, published once per day.

Markets past their end date are never presented as active. Each story is keyed to its underlying event, so the same fact is not re-published day after day.

Automation & editorial standards

Polydata News is a data-first newsroom: stories are drafted by our publishing pipeline directly from the metrics above, and every draft passes automated verification before it can appear on the site. Every figure in the prose must match the source data exactly — a draft containing any number that cannot be traced to the snapshot is rejected and rewritten. Evaluative labels (“longshot”, “against the consensus”) are permitted only when the corresponding computed condition holds. Near-duplicate drafts are blocked before publication.

Charts are rendered programmatically from the same data snapshot as the text — they are never illustrations, always evidence.

On-chain addresses are public information. We report positioning; we do not give trading advice, and nothing on this site is a recommendation to buy or sell anything.

Found a figure that looks wrong? Write to us — data errors are corrected and noted in the article.