On Nimb, the map reliably finds active Polymarket markets, but some analytics are still limited: holder lists are partial, themes are keyword groups, correlations rest on 30 days of data, and there is no history, API or alerts. Each limit is listed below with its number and status.
Every row is dated. When a fix ships, its row changes here and in the changelog.
Why does Nimb publish its limitations?
Because a number you cannot judge is worse than no number. Before launch, Nimb tested its own model on live Polymarket data (snapshot of 2026-11-07, 228 events, 770 markets, 2,982 holder wallets). The tests found a strong discovery interface with analytics that were often noise. Several fixes shipped the same day; the figures below say what is fixed and what is not.
Market reading is working out what a prediction market's price, flow and holders say about an event, and how much to trust it. This page is the "how much to trust" part for Nimb itself.
What can't Nimb tell you?
| Area | Limitation | Number | Status |
|---|---|---|---|
| Holders | Only the largest holders of each outcome are fetched; the API returns 500 or more. Concentration and "who is in" are partial. | 12 per side before launch; 25 per side now; net positions under $500 hidden | Partly fixed · 2026-11-07 |
| Market makers | Wallets present in 12 or more events are treated as market makers and kept off the map (still listed in holder tables). This is a rule of thumb, not a test of maker activity. | An earlier filter hid 46% of top-holder value | Open · 2026-11-07 |
| Hedged positions | Wallets holding Yes and No in the same market (hedged positions) used to be drawn as two separate positions. | 58.3M gross shares carried 0.55M net | Fixed: netted out · 2026-11-07 |
| Wallet "skill" | No tested measure of whether a wallet is good at forecasting. The earlier "profit wallets lean" signal was circular. | Matched the 30-day price direction in 63% of events | Signal removed; score not built · 2026-11-07 |
| Moves with / against | Correlations between markets are noisy on short histories. | 81–87% of flagged pairs rested on one observation; 43–71% expected false positives | Partly fixed: 30D only, ≥ 30 joint observations, |ρ| ≥ 0.5, leave-one-out · 2026-11-07 |
| Themes | Themes are topic groups from reviewed keyword rules, not statistical factors. Titles with unusual wording can be misgrouped. | 1 of 20 themes passed a statistical test after cleaning; about 13 of 228 events were mislabelled | Partly fixed: "explains X%" claim removed · 2026-11-07 |
| Price history | Hourly prices are forward-filled when nothing traded, so charts show flat steps. | 69% of hourly and 43% of 12-hour returns were exactly 0 | Partly fixed: correlations skip filled hours · 2026-11-07 |
| Short-window volume | Polymarket publishes 24-hour, 7-day and 30-day volume but not hourly. Shorter views estimate volume from taker trades scaled to Polymarket's total. | Taker trades are about 0.40× total volume; scale factor 1–5, default 2.4 | Open · 2026-11-07 |
| Money flow | Flow counts traders who crossed the spread (taker flow). It shows direction, not all trading. | Ranks match Polymarket's volume (Spearman 0.97) | By design, labelled · 2026-11-07 |
| Selection stability | No hysteresis: markets near the cut-off can swap between 5-minute refreshes. | Replay: 26.4 changes per hour unsmoothed vs 0.7 smoothed (1-hour window) | Open · 2026-11-07 |
| Selection tests | Tests were run at 80 markets and a 30% cap; the live rule uses 110 and 40%. | Near-certain leads 12.5–24% → 0–1.2% | Re-test pending · 2026-11-07 |
| Outcomes shown | Up to 5 outcomes per event are drawn (priced 1–99%). Size counts every outcome. | The old top-5 cap dropped 49.6% of event volume from sizes | Sizes fixed; display cap by design · 2026-11-07 |
| Markets entering between builds | Loaded in your browser with the most recent 1,000 trades per outcome, so their flow may be cut short until the next 6-hour build. | Build fetches up to 8,000 trades per outcome | Open · 2026-11-07 |
| History, API, alerts | No stored history beyond 30 days, no export or API, no alerts for new large positions. | — | Not available · 2026-11-07 |
| Other venues | Polymarket only. Kalshi data is not included. | — | Not available · 2026-11-07 |
Details on each area: holders and wallet lines, moves with and against, themes and factors, market selection.
How reliable are themes and correlations?
Treat themes as a reading aid: they group markets about the same topic so you can filter the map. They do not claim that those markets move together. For links between markets, Nimb's tests measured expected false positives with a shuffled-time null: 61% of flagged pairs at 1H–3D, 71% at 7D and 43% at 30D under the old rule. Nimb now shows links only on 30D and only when at least 30 joint 12-hour steps agree, |ρ| is 0.5 or more, and the link still holds (same sign, |ρ| ≥ 0.5) after dropping the single most influential step. At |ρ| ≥ 0.5 the tests estimate a false-discovery rate of about 20%, so roughly one link in five may still be chance. A rank-based (Spearman) cross-check and a multiple-testing correction are not applied.
How many holders does Nimb fetch?
25 per side (Yes and No) for each outcome on the map, with entry price and profit and loss. After netting wallets that hold both sides, net positions worth less than $500 are dropped. So a market with 2,000 holders shows at most 50, and the "largest holders" are the largest among those fetched. Holder P&L is recomputed on the net position.
What was fixed before launch?
| Date | Change |
|---|---|
| 2026-11-07 | Selection moved to the score and gates; recurring markets kept and collapsed; end date no longer drops live markets; finance titles reclassified. |
| 2026-11-07 | Circle size follows Polymarket's own volume across all outcomes, not holdings. |
| 2026-11-07 | Hedged Yes+No positions netted; circular wallet signal removed. |
| 2026-11-07 | Moves with / against limited to 30D with stricter tests; theme "explains X%" claim removed. |
| 2026-11-07 | Live prices over WebSocket, volumes every minute, selection re-run every 5 minutes in the browser. |
| 2026-11-07 | Holders fetched raised from 12 to 25 per side; category room follows volume (floor 4, cap 40%). |
How should you use Nimb's analytics, then?
As a starting point. Check the order book, the resolution rules and the market itself on Polymarket before you act. Any apparent edge, or alpha, you find in these signals may be noise or eaten by fees.
Last reviewed 2026-11-07 · Data from Polymarket · Not financial advice · Nimb is not affiliated with Polymarket.