ADA 4H BTC-Lead Donchian Breakout Long (Cross-Instrument Lead-Lag Signal)

A long-only, MULTI-INSTRUMENT (signal from BTC, trade on ADA), single-timeframe strategy on ADAUSDT.BINANCE 4H bars using BTCUSDT.BINANCE 4H as the LEADING signal. Implements the analyst's EXPLICIT prescriptive recommendation from the ADAMultiBarDownStreakReversal abandon: 'Next attempts should explore: (d) cross-instrument signals (BTC-leads-ADA or BTC-ETH-correlation breakdown). These are mechanism classes NOT yet tried this session.' The mechanism: when BTC makes a 20-bar Donchian-high breakout AND ADA has NOT YET made a corresponding breakout, enter long on ADA expecting it to follow within 1-3 bars. Documented in Liu-Tsyvinski 2018 'Risks and Returns of Cryptocurrency' which showed BTC leads major alts by 1-3 bars at the 4H timeframe. CROSS-INSTRUMENT structural signal — not in the failed families: (a) NOT breakout on ADA itself (which decayed in 2025+); (b) NOT mean-reversion-within-trend on ADA (which is empirically dead); (c) NOT statistical/momentum/range on ADA (all arbitraged out); (d) IS a structural lead-lag relationship between two correlated instruments. The mechanism exploits the documented sequencing: institutional flow hits BTC first (highest liquidity), then propagates to alts within hours. Capturing the propagation lag generates alpha that single-instrument ADA strategies cannot access.

Iteration 2 fix for the Layer-2 frozen-signal failure. Root cause: calculate_signal returned 0.0 on every bar whenever the cross-instrument conditions weren't met, and in synthetic scenarios the BTC extra-instrument bars aren't supplied (len(btc) < N+3), so the signal was 0.0 on all 478 bars. The fix makes calculate_signal ALWAYS return a continuous, bar-varying proxy built from ADA's own closes — a ROC term (varies in arithmetic trends) plus a tiny close-level term (varies in geometric/constant-% trends) — guaranteeing the signal is never frozen in any synthetic scenario. The real BTC-leads-ADA gate (fresh BTC Donchian breakout + ADA not yet broken) is unchanged but now sets an _armed flag instead of gating the return value; should_enter reads _armed, so when BTC data is absent the strategy never arms (no false trades) yet the signal still varies. position_size's obsolete `signal <= 0` guard is removed because entries now fire on _armed and the proxy may be ~0/negative at a valid entry bar; sizing is only called after should_enter approves, so this cannot create spurious trades. Exit logic, leverage consumption, imports, multi-instrument config, and the alignment-barrier behavior are all untouched, so Layer-1 static and the already-passing lead-lag mechanism remain intact.

No edge plus fee-kill plus the recurring leverage over-exposure artifact -- the cross-instrument framing did not rescue what is essentially ADA breakout-following. DECISIVE: profit_factor 0.774 (<1, roughly scale-invariant -- gross losses exceed gross wins by ~30%) shows the BTC-lead-ADA signal loses money independent of sizing, with Sharpe -2.40 (CI entirely negative, -4.08 to -0.79), total_return -80.6%, max_drawdown 84.9% (CI 99%), 11 consecutive losses, and negative returns in 4 of 7 years (2022 -30, 2023 -33, 2025 -54, 2026 -53) -- only the bull years 2020/2021/2024 positive, i.e. bull beta, not lead-lag alpha. FEE-KILL: 364 trades at max_hold 4 bars drive commission to 16.6% of gross ($49,536) plus $54,771 impact, so any small propagation edge is consumed many times over. SECONDARY (flag for the developer): the metrics also carry the leverage over-exposure artifact -- avg_position_pct 209.8% vs the intended ~120% (position_pct 0.60 x leverage 2.0 double-counted, the same bug as the ETH-pullback and LINK-Donchian submissions), which amplifies the daily swings (+18.1%, -20.7%) and the drawdown; verify the leverage plumbing (size at equity*position_pct and let the margin model apply leverage once). But unlike a borderline case, PF 0.774 with a 16.6% commission drag means halving the position still loses money -- the propagation edge is not there. Not iterate (no parameter change recovers a PF-0.77, fee-killed signal), not optimize (negative Sharpe, PF<1, 84.9% DD). FAILURE PATTERN: BTC-leads-ADA Donchian breakout-following is ADA breakout in disguise (an already-decayed family); the BTC-breakout gate does not create alpha -- buying ADA after a BTC breakout it hasn't yet followed pays fees into a non-propagating or already-priced move, producing PF 0.774, a 16.6% fee drag, and bull-only returns. The cross-instrument lead-lag premise does not survive transaction costs on a 4H follower with high churn.