Rebuilt from first principles out of homily's FMZ digest 'Quantitative Speed-Rate Trading Strategy' (https://www.fmz.com/digest-topic/4987, sections 'Indicator design' and 'Trading strategy design'); this is not a faithful port. The source takes the midpoint of the 35-bar highest high and lowest low (Donchian midline), smooths it with a 5-bar SMA, and computes the 35-bar OLS regression slope 'ss' of that series (its 'speed'). It then flips long/short EVERY time ss rises or falls versus the previous bar. That version is always-in and flips on every tick of slope-of-slope: 1,261 trades a year on 1H, a flip on nearly every other bar. The per-flip move is far below our ~0.20% double-notional flip cost, and lessons L150, L167 and L166 all say that design dies. Its claimed Sharpe was 0.117 even with 2019 fees. What we keep is the economic idea: the SLOPE of the smoothed channel midline is a low-noise measure of trend velocity. Velocity accelerating PAST a meaningful threshold marks trend onset, and velocity DECELERATING marks exhaustion, where the source exits. Changes to the source: (1) Use 4H bars on BTCUSDT.OKX, the OKX USDT perpetual swap and the modern USDT-margined twin of the source's OKX BTC quarterly. This keeps the source's venue and underlying, with catalog history from 2019-12 (~6.7 years, ~14,600 bars). (2) Normalise the slope: S = ss * 35 / ATR(35), i.e. how many ATRs the fitted midline travels across the regression window, so one threshold applies across 2020-2026 price levels. (3) Replace the slope-tick flip with a threshold ENTRY (|S| crossing 1.0) and a deceleration EXIT (S moving toward zero for 3 consecutive bars, the source's 'momentum weakening' rule made persistent). This adds an explicit FLAT state, expected ~50-60% of the time. (4) Add a 3-ATR chandelier stop. Why BTC, given its over-concentration in the corpus: OKX is 0.3% of experiments; the source mechanism was specified and published on OKX BTC; OKX alt swaps in our catalog only start 2026-09-24 (SOL/XRP/DOGE/LTC 4H), so only BTC/ETH/BCH have multi-year OKX history; ETH OKX already has a trend-flip strategy in the pipeline (EthOkxSuperTrendFlipChandelier); and BCH is thinner, which L136 warns against. No regime gate and no supplementary feed (L130, L160). Implementation note, given the sandbox timeouts in recent results: every indicator must be O(1) per bar. Use monotonic deques for HHV/LLV, a running sum for the SMA and ATR, and the closed-form rolling OLS slope with fixed x = 0..34: slope = (sum(x*y) - 17*sum(y)) / 3570. Recomputing it over a 35-element deque each bar is O(35) and fine. Never rescan the full history.