Dad's Channel Trade EXPLORATORY — uncertified

Faithful port of the handwritten note: touch-based 21/10 Donchian on highs/lows, both sides, gated by a trend-direction filter. Transcription in written-notes/transcription.md.
The note (MetaStock): HL:=HHV(Ref(H,-1),21); LL:=LLV(Ref(L,-1),10); — enter long on touch of HL, exit long on touch of LL. Short is the mirror: HS:=HHV(Ref(H,-1),10); LS:=LLV(Ref(L,-1),21); — enter short on touch of LS, exit short on touch of HS. "Trend Direction must be identified before" trading either side — the note doesn't say how, so the filter below is selectable (see the recommendation box).
Read before believing a result: entries/exits fill on an intra-bar TOUCH (resting stop order) — with daily bars the order of touches inside a bar is unknowable, so same-bar entry+exit assumes the WORST ordering (whipsaw loss taken). Gaps fill at the open. Shorts imply perps: funding/borrow are NOT modeled (short results flattered). The trend gate applies to entries only; open positions run to their channel exit. Nothing here inherits the certification.
Each trade posts its margin at the chosen leverage (notional = margin × leverage). Fixed $: trade size never changes — profits pool but don't compound. % of capital: each trade posts that percent of the CURRENT balance — wins compound the size up, losses shrink it (100% at 1× = all-in compounding). The account is compared against buying & holding the WHOLE starting capital in the asset. Isolated: a liquidation costs that trade's margin + entry fee, the rest survives. Cross: the liquidation price is much farther away (whole balance absorbs the loss) but ONE liquidation wipes the account. Date window: channel levels and the trend filter warm up on the FULL history before the window (no cold start); only trades ENTERED inside the window count — a position that would already be open at the start is skipped.
Filter recommendation (tested BTC/ETH/SOL/SUI, daily, full history): the 200-day SMA gate is the most robust choice — positive on every coin tested (worst coin: SUI 1.9× vs 0.75× B&H), ~40% fewer trades than ungated, similar or smaller drawdowns. Running with no filter is boom-or-bust: it looked great on BTC/ETH history but returned ~0% over three years on SUI and takes −60…−78% drawdowns — the same inconsistency that failed stop-and-reverse in the gauntlet project. The short side adds trades but little robustness: long-only + 200d SMA had the best worst-case MAR. The 50/200 cross reacts too slowly; the D55/20-state gate over-shorts sideways markets.