Secret Arena — descriptive crypto market screen
A plain description of the liquid crypto market: regime, trend, RSI, volatility and liquidity for each asset, computed from public Binance candles. It describes the market; it does not predict it, and it does not tell anyone what to buy.
Measured verdict
The score shown in this screen is not a proven trading signal. Over 162 days (30 April to 8 October 2026), its Top 5 did what a random draw from the same universe does.
None of these differences is significant. Fees alone would cost more than the point estimate. The only factor that kept the right sign is the volatility penalty (the most agitated assets did worse afterwards), and it was found after the fact: it remains a hypothesis to pre-register and test, not a rule.
The screen: liquid assets only
Universe: Crypto.com USDT spot pairs. Main screen: assets with at least $10M of Binance volume over 24 h. Currencies, stablecoins and tokenized gold (EUR, PAXG, XAUT…) are kept out of the screen, as are illiquid assets.
| Asset | Price | Volume 24 h | Regime (MA20/50, 1 h) | Trend 5 h | RSI 14 (1 h) | Volatility 20 h | 7 d | 30 d | Score* |
|---|---|---|---|---|---|---|---|---|---|
| … | |||||||||
Click a column to sort. * Descriptive score: a hand-weighted sum of the columns (regime, trend, RSI, 7 d and 30 d momentum, liquidity, volatility). It was never calibrated and showed no predictive power (see the verdict above). Volatility = standard deviation of hourly returns over 20 h.
Assets kept out of the screen
Research: multi-timeframe lead-lag
Lagged correlations between 18 assets on 1 h, 4 h and 1 d candles (“does A move before B?”). This is a research layer: no bot, score or signal reads it. Re-estimated every week and tested on the following period (walk-forward), the relations did not hold: 47.3% (1 h), 48.7% (4 h) and 50.2% (1 d) of signs right, and no relation was still there in the next period. A round trip costs 10 to 40 times the measured edge.
Archives: shadow bot V10.1 and social confluence V7.2
Archived, measured result: −67.5%, mostly from feesBoth paper bots are archived as signals and paused since 8 October 2026. Their history is kept and stays readable.
The shadow bot did not follow the Secret Arena score: it followed the lead-lag meta-signal V6. The social confluence (X Alpha and News Pulse) only filtered that bot's positions: at equal exposure its lanes lost as much, and the “consensus” lanes looked better mainly because they hardly traded (average exposure 0.02 to 0.05).
Kept history: /api/secret-arena/shadow-bot · /api/secret-arena/social-confluence/history
Method
- Sample: one stored snapshot per day (the first after 00:00 UTC with a full Top 20), 162 days from 30 April to 8 October 2026.
- No look-ahead: entry at the open of the next 1 h candle, exit at the open 24 h, 168 h or 720 h later. All prices from Binance.
- Random reference: an equal-weight draw from the same universe, so the result is the excess over the universe average (about 106 assets listed at each date).
- Intervals: Newey-West 95% CIs with a lag of the horizon minus one day for overlapping returns. Independent periods: 162 at 24 h, 22 to 23 at 7 d, only 4 to 5 at 30 d, so no 30-day verdict is possible yet.
- Factor by factor: the score was rebuilt on the whole universe at each date from 1 h candles, then a daily Spearman IC was computed between each factor and the future return (EUR, PAXG and XAUT excluded).
- Reservations: 11 factors × 3 horizons = 33 tests on a single 5-month period. The negative momentum sign may belong to that period; it is not turned into a contrarian rule without pre-registration and an out-of-sample test.
- Shadow bot: replayed on clean Binance prices with and without fees, and compared with 300 random portfolios using the same schedule and weights.
How the screen is refreshed: every 15 minutes, outside the web server. A refresh whose universe is empty or abnormal never overwrites the previous screen: it is logged and the screen keeps its date, shown at the top of this page. General method of the site: /en/methodology
Research, not advice. This page describes public market data. It places no order, connects to no account, and nothing in it is an invitation to buy or sell.