Research note
Score Weight Sweep: optimization is not validation
Lens searches and threshold sweeps rank candidates. No frozen winning IC, uplift, returns or passing verdict is inventoried.
Backtest verdict: Unknown — no frozen result recorded in the inventory. Lens-weight optimization, threshold sweeps and nightly lens records are distinct research products; a winner is not a validated edge.
Open this research tool in Backtesting
The rule
The weight search explores nonnegative lens weights summing to one against forward SPY returns. A sign-aware variant also explores lens flips; constrained, baseline, top-candidate and cross-validation/OOS outputs remain distinct. The on-demand request samples 2,000 candidates and returns a top-ten view; those settings are not independent observations.
The opening range is 10 years at a 14-day cadence, with a 20-day IC horizon; 5/20/60-day horizons are selectable. The separate threshold sweep defaults to the directional family and also exposes contrarian; API plumbing/buy-and-hold support remains distinct. This is a parameterized search, not a single trade rule. Entry/exit and cost behavior belong to the selected score strategy; no common optimizer cost or invented stop/sizing model is imposed.
Methodology
Evidence as of: the INVENTORY review dated November 5, 2026; this is the consolidation planning baseline, not a new observation window. Publication does not extend any study's sample.
The current search uses cached historical inputs and ranks lens configurations by its native IC criteria. Threshold results use their own strategy objective. Review the baseline, candidate constraints, every fold, CV/OOS fields and full sweep rather than only the displayed optimum. Selection and multiple testing can inflate the chosen score; overlapping forward horizons reduce effective independence.
H19 nightly lens runs are saved separately from on-demand output. No dated candidate/fold/sweep payload, seed, winning IC, uplift or historical outcome is inventoried. Exact sample dates, benchmark alignment and price basis must come from that original payload, not today's settings.
Results
| Study / variant | Inventory evidence | Window / basis | Strategy return / CAGR | Benchmark | Max drawdown | Closed trades / observation N | Significance / IC | Native verdict |
|---|---|---|---|---|---|---|---|---|
| Constrained / sign-aware weights, CV/OOS | D08 / F16 / H18 | not recorded | not recorded | not recorded | not recorded | not recorded | not recorded | Unknown — no frozen result |
| Threshold sweep | D08 / H17 | not recorded | not recorded | not recorded | not recorded | not recorded | not recorded | Unknown — no frozen result |
| Nightly saved lens records | D08 / H19 | not recorded | not recorded | not recorded | not recorded | not recorded | not recorded | Unknown — no frozen result |
The linked workspace retains the complete native output and notes when available. Missing evidence is not a zero return, a successful empty result or a failed strategy.
Verdict and limitations
Unknown — no result/verdict recorded in this inventory. Positive IC, higher Sharpe in a selected sample or an optimizer rank is not PASS. Current on-demand results and nightly stored proposals must not be combined into a fictional run archive. No new weights are applied by reading this article.
Sources and live research
INVENTORY D08, F16/H17–H19. Sources to check: frontend/components/optimize/ScoreOptimizerBoards.tsx, frontend/lib/score-optimizer-page.ts, backend/app/scoring/optimization.py and strategy_backtest.py. The missing evidence is a dated candidate/fold/sweep payload with its precise selection basis. Read existing nightly lens records without rerunning them.