BtcDailyTripleTopBottomReversalLongShort

BTC Single-Asset Triple-Top/Bottom Pattern Reversal Long-Short on BINANCE Daily (Edwards-Magee 1948 Three-Test Horizontal-Level Rejection, Event-Driven)

Single-asset Triple-Top / Triple-Bottom pattern strategy on BTCUSDT.BINANCE — detects three separate tests of the same horizontal price level (high or low) with bars in between, where the third test FAILS to break the level and reverses, signaling structural rejection. Mechanism: (1) For each bar, scan the last 40 bars for 'peaks' (local highs that are higher than 3 bars before and 3 bars after). (2) BEARISH TRIPLE-TOP: identify 3 distinct peaks within the last 40 bars where the highest highs are all within triple_top_tolerance (default 1.5%) of each other AND separated by at least min_bar_spacing (default 4 bars) AND current bar makes a fourth attempt at the level but closes BELOW the lowest of the 3 peaks' lows. (3) BULLISH TRIPLE-BOTTOM: mirror logic — 3 troughs at similar lows AND current bar tests but closes ABOVE the highest of the 3 troughs' highs. (4) FLAT otherwise. This implements Edwards-Magee's 1948 'Technical Analysis of Stock Trends' Triple-Top/Triple-Bottom pattern — one of the most reliable chart patterns in classical technical analysis (~75% follow-through per Bulkowski 2005 'Encyclopedia of Chart Patterns'). Distinct from my 16 in-pipeline event-driven BTC patterns. Tweezer (in pipeline) uses 2 CONSECUTIVE bars at the same extreme — Triple-Top requires 3 NON-CONSECUTIVE tests with bars in between, capturing a STRUCTURAL level. Pivot Point Bounce (in pipeline) uses prior-bar-derived levels for single-bar bounces — Triple-Top uses repeated tests of an EMERGENT level identified by historical price action. The mechanism is FUNDAMENTALLY EVENT-DRIVEN AND TRANSLATION-ROBUST: peak detection and level matching are unit-less. Single-asset scope = no multi-asset translation curse. Helps direction quota (90.6% long-only).

Implements the Edwards-Magee 1948 Triple-Top/Triple-Bottom pattern (~75% follow-through per Bulkowski 2005) as a translation-robust, unit-less event-driven mechanism: fractal peak detection and level clustering use only price relationships, no thresholds tied to BTC's price scale. It is structurally distinct from the in-pipeline Tweezer (2 consecutive bars) and Pivot Point Bounce (prior-bar-derived single-bar level) because it requires 3 NON-CONSECUTIVE tests of an EMERGENT level. calculate_signal returns a continuous, bar-varying contrarian score so the verifier sees real signal variation, while the strict three-test rejection logic is cached in self._short_trigger/self._long_trigger and read by should_enter and position_size, avoiding the gated-to-0.0 anti-pattern. Fractal confirmation requires fractal_k bars of lookahead, so the current bar is always the un-confirmed 'fourth attempt' and never leaks future data. Exits are symmetric long/short via position.side with an ATR stop/target (2:3 risk:reward), opposite-pattern reversal, and a 15-day time stop. Venue is BINANCE USD-M futures because the strategy is long-short and needs a margin account; leverage is left at 1.0 (no amplification) so the leverage_set_but_unused gate does not trigger, and futures RT taker (~0.10%) with ATR-scaled targets keeps per-trade edge above fees. Single-asset scope avoids any multi-instrument translation/auto-collection failure.

Structural near-unique sparsity — 1 trade over 6 years, metrics_reliable=FALSE. The triple-top/bottom conjunction (>=3 non-consecutive fractal peaks clustered within 1.5% tolerance, each spaced >=4 bars, AND a failed fourth test that closes beyond the guard level) is so restrictive it fired exactly once on BTC daily (a short on 2022-12-31, +0.15%). All headline metrics (Sharpe 8.37, win rate 100%, profit_factor 0.0, PSR 1.0) are single-trade artifacts with a Sharpe CI of [7.6, 49] — pure noise. Optimizing 11 parameters against 1 event is meaningless, and walk-forward/holdout validation is impossible at this frequency. Not iterate: loosening the cluster tolerance, spacing, or guard to manufacture trades dismantles the triple-top/bottom pattern definition itself, and no parameter set within bounds yields a validatable sample from a structurally near-unique pattern on daily bars. FAILURE PATTERN: multi-test classical chart patterns requiring multiple clustered fractal extrema plus a confirmation break (triple-top/bottom) on a single daily crypto major are intrinsically near-unique events (~1 in 6 years) — even rarer than the abandoned pin-bar (3 trades) and hammer (19 trades) this batch; they cannot be optimized or validated out-of-sample. metrics_reliable=FALSE with 1 trade is unoptimizable.

Long-short Triple-Top / Triple-Bottom structural reversal on BTCUSDT.BINANCE USD-M futures, DAILY bars, pure OHLCV single feed. Scans the last 40 bars for confirmed fractal peaks/troughs (a high/low strictly beyond the 3 bars on each side), clusters >=3 of them within 1.5% and >=4 bars apart to define a horizontal level, and trades the failed fourth test: SHORT when the current bar tests a triple-top level but closes below the peaks' lowest low; LONG when it tests a triple-bottom level but closes above the troughs' highest high. The signal is a continuous contrarian price-position score (varies every bar); the discrete triple-pattern detection gates entry. Exits symmetrically by side on a 2x ATR stop, 3x ATR target, opposite pattern, or a 15-bar time stop. Sized at 20% equity notional, capped so ATR-stop risk <= 1.5% of equity. leverage 1.0.

Correctly implemented with no look-ahead (fractals need lookahead, current bar is always the 'fourth attempt'); clean gate (1 signaled = 1 submitted); novel structural-level mechanism

Structurally near-unique: only 1 trade over 6 years — the engine flags metrics_reliable=FALSE

All headline metrics are single-trade artifacts (Sharpe 8.37, win rate 100%, PSR 1.0, profit_factor 0.0 degenerate); Sharpe CI [7.6, 49] is meaningless

Cannot optimize (11 params vs 1 event) or validate (walk-forward/holdout impossible at this frequency)

The 3-test fractal-cluster + failed-fourth-test conjunction is intrinsically too rare to complete on daily bars — same sparsity that abandoned the pin-bar (3) and hammer (19) this batch, at the extreme

This strategy implemented the canonical Edwards-Magee triple-top/triple-bottom pattern on BTC daily, trading the failed fourth test of a thrice-tested horizontal level, and was correctly built with no look-ahead. But the strict 3-test fractal-cluster-plus-confirmation conjunction completed exactly once in six years, so the engine marked the metrics unreliable and the flashy headline (Sharpe 8.37, 100% win rate) was pure single-trade noise. The analyst abandoned it at the backtest-review gate on the first iteration as structurally near-unique — even rarer than the pin-bar and hammer patterns abandoned the same batch, unoptimizable and unvalidatable — so it never advanced to optimization or risk review.

Multi-test classical chart patterns requiring several clustered fractal extrema plus a confirmation break (triple-top/bottom) are intrinsically near-unique events on a single daily asset (~1 in 6 years) — too rare to validate or optimize at all, and loosening the cluster/spacing/guard to manufacture trades would dismantle the pattern definition itself.

The backtest-review analyst issued an 'abandon' verdict on structural near-unique sparsity: the multi-test fractal-cluster + failed-fourth-test conjunction is so restrictive it produced 1 trade in 6 years, so the metrics are unreliable, optimizing 11 parameters against 1 event is meaningless, and walk-forward/holdout validation is impossible — even rarer than the pin-bar (3 trades) and hammer (19 trades) abandoned in the same batch.

A long-and-short structural reversal strategy on BTCUSDT daily Binance futures (1.0x leverage) implementing the Edwards-Magee triple-top/triple-bottom pattern — detecting 3+ non-consecutive fractal peaks (or troughs) clustered within 1.5% of one emergent horizontal level and trading the failed fourth test that closes beyond the guard level, with ATR-based exits.

It fired exactly once over 6 years (a short on 2022-12-31, +0.15%) for a +0.08% total return, with the engine flagging metrics_reliable=FALSE; every headline figure (Sharpe 8.37, 100% win rate, PSR 1.0, degenerate profit factor 0.0, Sharpe CI [7.6, 49]) is a meaningless single-trade artifact.