Derived by the Research Lead from Tushar Chande's Trend Score as published on ProRealCode (Source URL: https://www.prorealcode.com/prorealtime-indicators/tuschar-chandes-trend-score/). The source defines the score: compare today's close with each close from k=11 to k=20 bars ago, score +1 if today is higher and -1 if lower, and sum the ten votes (range -10..+10). The source then smooths the score (SmoothingPeriod=5). The source is an indicator only; the trading rules below are the Research Lead's operationalization.
Mechanism: the score is a sign-vote on the medium-horizon trend that deliberately skips the most recent 10 bars. It measures how persistent the trend is and ignores the short-term reversal noise that kills raw-momentum crossovers on alts.
Market: five Binance COIN-M inverse alt perpetuals that the corpus has never tested (LINK, DOT, BCH, ETC and FIL _PERP.BINANCE_CM, 0 prior experiments each). BINANCE_CM is 2.6% of the corpus against a quota of at least 5%. Bars are 12-HOUR, and each perp is traded independently as its own netted position.
Rules: go long when the smoothed score is >= +entry_threshold and short when it is <= -entry_threshold. Stay flat in between; the strategy is not forced to always hold a position. Exit when the score crosses back through the exit band, or on a platform-matched intrabar ATR trailing stop (a reduce-only _submit_stop_market; no exchange-resting order is required). Size each position with contracts_for_notional() on the inverse USD-face contracts at 18% of equity notional per perp, unlevered (max gross 90%).
Fee arithmetic: on 12H alts, a trend that persists across the 11-20 bar (5.5-10 day) vote window usually spans a multi-day move of 6-15%. Capturing a third of that gives about 2.0-3.0% gross per trade. Costs are about 0.06% COIN-M round-trip taker plus about 0.05% impact and funding drag, so the expected net is >= 1.0% per trade, more than 10x cost.
Expected frequency: about 10-15 trades a year per perp, 55-70 a year across the five, and 300+ over the 2020-09..2026-10 history.
This is distinct from the many 4H/1D dual-timeframe EMA-confluence variants already in the corpus. It has no higher-timeframe filter, uses a lagged sign-vote instead of price versus a moving average, and trades on a coin-margined venue.