OiExpansionConfirmedBreakoutBasketLS

Open-Interest-Expansion-Confirmed Breakout, Long-Short (3-Asset USD-M Alt Basket SOL/AVAX/LINK, 4h Bars + Open-Interest Feed, Only Trade a Directional Range Break When OI Is SIMULTANEOUSLY Rising = Fresh Leveraged Capital Confirming the Move, Exit on OI Contraction or ATR Stop, 4-Parameter)

A LONG-SHORT, multi-instrument breakout on three liquid Binance USD-M alt perps (SOLUSDT, AVAXUSDT, LINKUSDT) on 4h bars, gated by an OPEN-INTEREST confirmation filter. Plain price-breakout strategies have failed repeatedly in this factory (failed-breakout XRP 1h and compression-breakout DOGE 15m both liquidated with negative expectancy) because they take EVERY range break — including the mechanical false breaks (stop-runs, short-covering pops) that have no fresh capital behind them and immediately revert. The structural fix here is to require that aggregate OPEN INTEREST is expanding at the moment of the break: a break accompanied by rising OI means NEW leveraged positions are being opened in the direction of the move (genuine trend initiation with fuel), whereas a break on flat/falling OI is positioning being unwound (squeeze / liquidation cascade) that mean-reverts. OI history on Binance USD-M is multi-year (Binance Vision metrics CSV, sumOpenInterest) — unlike the 90-day liquidations feed that killed the cascade strategies — so this backtests over the full window. A 3-name basket dilutes single-symbol overfitting. Deliberately 4-parameter to resist the overfit deaths (breakout_lookback, oi_lookback, oi_expansion_pct, atr_stop_mult).

Implements the hypothesis's structural fix — require OI expansion at the moment of the break so only breaks with fresh leveraged capital behind them are taken. Breakout is a validated (trend/continuation) edge family; the OI-expansion gate is the differentiator that kills the false-break deaths of plain price breakouts. Exit design deliberately avoids the inverted reward/risk trap that killed prior mean-reversion attempts: winners are held while OI keeps expanding, losers cut by the ATR stop. Venue: BINANCE USD-M futures (not spot) because the strategy is long+short and uses leverage>1, and OI is a futures data series; leverage=3.0 is consumed in the per-leg notional cap so it is not inert. Framework-constraint decision: the backtest engine loads open_interest/metrics supp for the PRIMARY instrument only (per-leg supp beyond funding is not provided), so per-name OI for AVAX/LINK is unavailable — those legs would otherwise never trade. Rather than silently degrade to a price-only proxy (forbidden) or leave two legs dead, the strategy reads the primary (SOL, most-liquid) OI series and applies it as an AGGREGATE ALT-MARKET OI-expansion regime filter to all three legs. This is the same edge mechanism expressed at the market level ('market-wide fresh leveraged capital confirming the move') and is arguably more robust than single-name OI; if no OI series is present at all, every leg stays flat. The 3-name basket dilutes single-symbol overfitting, and the 4 core parameters (breakout_lookback, oi_lookback_bars, oi_expansion_pct, atr_stop_mult) keep the design low-dimensional. Dry-run on synthetic bars confirmed the compound gate (breakout AND OI-expanding) fires entries and the OI-contraction exit triggers, guarding against the verification_loop zero-trade failure; the ~22-bar warmup is a small fraction of the 4h sandbox window.

No edge to optimize. Across a decisive 886-trade, ~5-year sample the OI-confirmed breakout basket nets profit_factor 0.976 (< 1.0, net loser), Sharpe -0.02 (CI [-0.66, 0.71] straddling zero), total_return -1.18%, and negative dollar expectancy (-$7.35/trade). Although avg_trade_return_pct 0.263% clears the fee floor (so this is not a fee-edge death), the signal has essentially zero directional edge — gross wins ~= gross losses — and the result is deteriorating (a -21.5% 2026 collapse, rolling Sharpe ending near -11). The OI-expansion confirmation filter was the whole structural thesis meant to fix the false-breakout problem that liquidated the XRP/DOGE breakout siblings; it did not create edge, leaving this in the OHLCV breakout class that has zero survivors (L9). Capacity is also a blocker: capacity_usd $2,071 with impact_cost_pct 694.8% means any marginal edge is real only at toy scale. There is no articulable path from PF 0.976 to a robust, deployable edge by tuning 4 parameters — it would be curve-fitting a near-breakeven signal. Abandon rather than spend 2 hours optimizing.

Long/short 4h Donchian breakout on a 3-asset Binance USD-M alt basket (SOLUSDT primary + AVAXUSDT, LINKUSDT extra legs), gated by an open-interest-expansion confirmation. Each leg computes a continuous channel-position signal (close minus prior-channel mid, in half-widths); |chan_pos|>=entry_thresh is a breakout. An entry (BUY on an up-break, SELL on a down-break) is only taken when aggregate open interest is expanding — new leveraged capital is confirming the move — filtering out the mechanical false breaks (stop-runs, short-covering) that revert. Positions exit on an ATR-from-entry stop or when OI contracts (fuel leaving). Sizing is risk-based (risk_frac of equity against an atr_stop_mult*ATR stop) and hard-capped per leg at equity*leverage*max_notional_pct/n_legs, with leverage (3.0) wired into the notional cap.

Abandon at backtest_review; Donchian breakout is a repeatedly-dead class and the OI differentiator did not clear fees.

STRONG ABANDON RECOMMENDATION (not a QA code fail). Decisively negative and well-sampled: 161 trades, total_return -27.18%, Sharpe -1.50 (CI [-3.09, 0.10]), PF 0.45, win_rate 0.35, avg_trade_return_pct -0.83%. The OI-expansion gate did not rescue the breakout; genuine edge failure, no L17 signature, balanced 87L/74S.

Design outcome, not tuning — abandon. A fix would require making the OI-contraction exit profit-conditional or a trailing stop (a redesign).

Claimed positive reward:risk is NOT realized: avg_win $398 < avg_loss $471 (0.84) at 35% win rate → doubly negative. The OI-contraction exit clips winners early (small wins) while the 2.5-ATR stop sizes losses full — the inverted RR the hypothesis said it avoids. Code implements the described exit (not a critical structural mismatch), but the payoff shape is opposite the intent.

Note two legs are proxy-confirmed; if per-leg OI supp can be loaded, read each leg's own series.

OI confirmation is MARKET-WIDE (SOL only), not per-name (disclosed: engine loads OI supp for the primary instrument only). AVAX/LINK breaks are gated by SOL's OI, so a LINK break with rising SOL OI but falling LINK OI is still taken — weakening the per-name 'fresh capital confirming the move' thesis. Disclosed and OI-based (flat if no OI series), so a warning not a critical.

No liquidation this time; max_drawdown 29.3% is contained (the leverage/notional cap worked)

Well-sampled (886 trades, ~5 years) so the flat result is decisive, not noise

avg_trade_return_pct 0.263% clears the futures fee floor — not a fee-edge death

The OI-confirmation filter is an honest, non-bug design (primary-leg OI used as a market-wide regime gate)

profit_factor 0.976 (< 1.0, net loser), Sharpe -0.02 (CI [-0.66, 0.71]), total_return -1.18%, expectancy -$7.35/trade — no directional edge

Deteriorating: a -21.5% collapse in 2026 offsets the 2023/2025 gains; rolling Sharpe ends near -11

capacity_usd only $2,071 with impact_cost_pct 694.8% — edge (if any) exists only at toy scale, a promotion blocker

The OI-expansion filter — the entire structural thesis — failed to convert the breakout into a positive edge; this is the OHLCV breakout class (zero survivors) with an OI gate bolted on

benchmark_meaningful=false (multi-instrument basket), so no favorable comparison to lean on

none

This strategy tried to fix the false-breakout problem that had blown up prior single-name breakout attempts by requiring open interest to be expanding at the moment of a range break — evidence of fresh leveraged capital fueling a genuine move — across a diversified SOL/AVAX/LINK alt basket. The risk controls worked this time (no liquidation, drawdown contained at 29.3%), but over a decisive 886-trade, five-year sample the OI filter failed to produce edge: profit factor 0.976, Sharpe -0.02, and a slightly negative total return, with results deteriorating into a -21.5% 2026 collapse. Capacity was also fatal — only $2,071 at 694% impact cost, meaning any marginal edge existed solely at toy scale. The analyst abandoned it at backtest review, judging it another member of the zero-survivor OHLCV breakout class where four-parameter tuning could only curve-fit a near-breakeven signal.

Bolting a plausible confirmation filter (rising open interest) onto a structurally edgeless breakout does not manufacture a directional edge — it merely reduces trade count while leaving gross wins roughly equal to gross losses; and a signal profitable only at ~$2,000 capacity with 694% impact cost is undeployable regardless of expectancy.

The analyst abandoned it at the pre-optimization backtest-review gate: the OI-expansion confirmation filter — the entire structural thesis meant to fix the false-breakout problem that had liquidated the XRP and DOGE breakout siblings — failed to convert the breakout into positive edge, leaving it in the OHLCV breakout class that has zero survivors, with near-breakeven expectancy and a toy-scale capacity blocker that four-parameter tuning could not rescue.

A long-short, 4h Donchian breakout on a three-name Binance USD-M alt basket (SOL/AVAX/LINK) that only took a directional range break when aggregate open interest was simultaneously expanding — treating rising OI as fresh leveraged capital confirming genuine trend initiation and filtering out fuel-less false breaks — with exits on OI contraction or an ATR stop.

Over a decisive ~5-year, 886-trade sample (444 long / 442 short) it netted essentially no edge: profit factor 0.976 (a net loser), Sharpe -0.02 (CI [-0.66, 0.71] straddling zero), total return -1.18%, expectancy -$7.35/trade, and max drawdown 29.3% (contained, no liquidation). avg_trade_return_pct of 0.263% cleared the fee floor, but capacity was only $2,071 with impact_cost_pct 694.8%, and the result was deteriorating (a -21.5% 2026 collapse, rolling Sharpe ending near -11).