# Direction and portfolio goal v8 Exploratory retrospective tests, not untouched validation. Histories have already been inspected repeatedly. Data through November16,2026. User's 40% gain request is interpreted as cumulative account return within each annual test period (2026 is partial), not40% per trade or annualized. 80 combinations per market: four entry families and five filters from v6 × fixed3/5/10/20 daily bars. Signals computed at completed closes; next-open entries with the original gap restrictions. No stop or profit target in this distinct fixed-horizon direction test. Prediction succeeds only when sign × (horizon close / next open −1)>0; ties fail. Direction accuracy and after-fee profitable-trade rate are separate. No future-price target is claimed. One position per symbol; no overlapping same-symbol entries. Only horizons completed within the period are eligible, defined before performance is measured. Crypto requires consecutive UTC days. Entry ranking on ties is ticker order, not subsequent outcome. Portfolio starts with capital1. Each entry allocates10% of then-current opening equity, capped by available cash collateral and ten positions. Both longs and hypothetical shorts reserve full notional; no borrowed capital or cash reuse before an exit's daily close. Positions marked to daily closing prices, including unrealized losses. Previous closes carry only when that asset has no bar. Opening equity uses current opens for positions with a bar, never the current close. Fees:half of assumed round-trip cost at entry and half at exit, each based on initial trade notional. Stockcost20bp/high50bp; cryptocost60bp/high100bp. Separate base/high-cost portfolios may accept different positions because fees consume cash. Idle cash earns0. No tax, dividends, interest, historical borrow or crypto funding is modeled. Shorts therefore remain hypothetical. No stops means adverse moves can exceed losses in earlier target studies. Selection:2024 only; >=50 eligible predictions AND >=50 portfolio trades; >=80% correct direction on both eligible predictions and accepted portfolio positions; portfolio gain>40%; positive higher-cost portfolio return. Freeze selection before checking2025/2026 using identical criteria. Long,short,combined scored independently. No development candidate passed. 720 additional stop/target configurations (v7) remain separately archived in expanded-setups-v7.json; they do not represent fixed-horizon direction predictions. Earlier v6's1200 configurations remain unchanged. All search results are disclosed; later-year best rows are descriptive hindsight and cannot select a promoted live model. ## Evaluation extension v9 The same80 configurations now include benchmark and descriptive uncertainty fields in direction-evaluation-v9.json; v8 results remain preserved. There are no new entry filters or selections in this extension. `node scripts/evaluate-direction-models.mjs` recomputes this extension. The baseline invests100% of starting capital at the first available open and exits the last available close for that year. It charges the same initial-notional fees as the active strategy and holds SPY for stocks / BTCUSD for crypto. This is a price-only reference; stock dividends are omitted and different exposure is not adjusted. 95% Wilson intervals describe each observed binomial count. Cross-asset signal dependence and repeated model selection are not corrected; these ranges are not established future win-probability intervals. Both all-prediction and accepted-trade denominators remain available. Time invested is the share of market dates when a position exists after opening entries, before closing exits. Average gross exposure is the sum of reserved initial notionals divided by opening equity, averaged across all market dates (zero on idle dates). It is not a volatility or beta-adjusted exposure estimate.