PSMI Methodology

The PSMI is a daily 0–100 reading of how active, how broadly profitable, and how capital-committed a verified cohort of top Polymarket traders is, measured against that same cohort’s own recent history. A reading of 50 means the cohort is behaving in line with its trailing 90-day baseline.

What it is not. It is not a forecast, not a recommendation, and not a directional view on any market. Prediction markets are thousands of separate binary questions, not one asset with a price that goes up or down, so “bullish” and “bearish” framing does not apply and is not used. The index describes cohort behaviour. Any relationship between that behaviour and future prices is an empirical question we have not claimed to have settled.

The cohort is the product. A wallet is included when all of the following hold:

1. Parity-verified. The wallet is published with a passing parity receipt — our computed profit and loss reconciles to its on-chain totals. Wallets whose numbers we cannot verify are excluded regardless of how good they look.
2. Not a market maker. Wallets classified as market-making, or scoring at or above 70 on our market-making score, are excluded. Market makers are the largest single source of mechanical volume; leaving them in would let inventory management masquerade as conviction.
3. Not an arbitrage bot. Wallets scoring at or above 70 on our arbitrage score are excluded.
4. Recently ranked. The wallet carries a current 30-day profit-and-loss rank, which is also the activity gate.

From the wallets that qualify, the index takes the top 1,000 by 30-day rank. Membership is recomputed monthly and every change is logged. Cohort size, exclusion counts, and active-wallet counts are stored with each daily value.

The size is fixed while the pool it is drawn from grows, so the cohort becomes more selective over time without the index definition changing — which keeps history comparable. When the published pool is smaller than the target, the real size is what gets stated on the page; we do not describe the cohort as larger than it is.

Each component is normalised so that 50 equals the cohort’s own trailing 90-day baseline. Signed quantities use z-scores against trailing distributions, never ratios of signed values.

Weighting rationale. Momentum leads because it is the headline behaviour and the most interpretable signal, but at 30% it can no longer carry the index by itself. Breadth and participation are weighted equally as independent confirmations. Capital conviction is capped at 20% while the underlying cash-flow metrics remain the newest part of our warehouse.

When no cohort activity exists in a window, every component returns 50 rather than an extreme. A data gap is an absence of information, and the index says so instead of reporting it as a signal.

The composite is the weighted average of the four components, rounded and clamped to 0–100. Zone labels describe activity levels and are anchored to the historical distribution rather than to round numbers, so that each label stays meaningfully rare:

• QUIET — Smart money is sitting out. Activity and conviction are well below this cohort’s own baseline.
• WATCHING — Smart money is engaged but cautious. Activity is running below its usual pace.
• ACTIVE — Smart money is trading at a healthy clip, at or above its own baseline.
• SURGING — Smart money is unusually active. This is one of the busiest, most broadly profitable stretches on record.

The target shares are the point of the design: if SURGING were reached every other week the label would be meaningless, and if it were never reached it would be dead weight. Cut points are derived from the observed distribution and frozen at each methodology version.

One canonical value per day, computed after the daily warehouse run, timestamped as a UTC calendar day. The live gauge refreshes every 30 minutes so the page is current, but intraday values are never stored. Only daily closes are archived, charted, downloadable, and citable.

Published values are never restated. If the methodology changes, the version number changes with it, the change is recorded in the changelog, and the new definition applies going forward — creating a new series segment rather than rewriting days that may already have been quoted. The write path enforces this: a value stored under one methodology version cannot be silently overwritten by another.

The index launches with reconstructed history so it has a chart and a track record from day one. Backfilled values are computed by the same code as live values, from the same warehouse tables, and every reconstructed observation is flagged is_backfill = true in the downloadable data.

Cohort look-ahead in reconstructed history. Historical values are computed using cohort membership selected from current rankings, because point-in-time membership is not available in our serving layer yet. This biases reconstructed history upward: it measures the past behaviour of wallets we already know went on to rank highly. Backfilled readings should therefore be read as “how this cohort behaved then”, not as “what the index would have printed at the time”. Live readings from launch onward carry no such bias.

• Wash trading. A meaningful share of Polymarket volume is wash or self-matched. Excluding market-making and arbitrage wallets removes the largest source, but the participation component may still include some residual inflation.
• Share volume, not notional. Some warehouse volume measures are share-based rather than dollar-based. The participation component uses fill counts specifically to avoid this ambiguity.
• Capital conviction is a deployment proxy. The component currently measures net capital deployed into positions (buys minus sells) rather than net USDC deposited to the exchange. Deposit-based conviction requires a daily cash-flow table that is not yet in our serving layer; when it ships, that change will be a versioned methodology update, not a silent swap.
• Cohort concentration. With 1,000 wallets, a small number of very large traders can move the momentum and conviction components more than the breadth component. That is why breadth exists and why it is weighted independently.
• Polymarket only. The index measures one venue. It is not a cross-platform prediction-market indicator.

Every part of the published reading is available without a signup:

• Daily history (CSV) — score, zone, all four components, methodology version, backfill flag.
• Free API endpoint — current score, zone, and date. No key required.

Cite as: Polycopy's Polymarket Smart Money Index (PSMI), with the reading date and methodology version — for example, “Polycopy's Polymarket Smart Money Index (PSMI) stood at 62 (ACTIVE) on 12 March 2026, methodology 2.0.0”.