Connecting to server...
Free-tier server may take up to 60s to wake up
Validated Backtest Results
See how each of the … backtested portfolio strategies performed during out-of-sample testing — returns, risk metrics, and consistency scores that weren't used to train the models.
Start with the Summary tab for the big picture, then drill into individual strategies. Green rows beat the S&P 500 benchmark. New here? Start with the Getting Started Guide or check the Glossary for unfamiliar terms.
How to read these results
- Live results and backtests will differ for the 9 quarterly variants, and the reason is known. Itop3_Q, Itop5_Q, Itop10_Q, Ntop3_Q, Ntop5_Q, Ntop10_Q, PI1, PI2, PN1 all rebalance every 63 sessions. At the 2026-04-08 and 2026-07-09 rebalances the live system selected from a truncated candidate book because of a bridge defect, fixed 2026-07-29. The backtests select from the full book. So a live-vs-backtest gap at those two dates is known-cause and expected, not evidence about the strategy.
- The 12 newly added variants carry one parameter set across all six walk-forward windows; the original 15 carry six per-window sets. A single set tested across six windows is a different and weaker claim than six sets each fitted to its own window — compare the two groups with that in mind.
- Six variants breach the −50% maximum-drawdown limit, which is why the deployment checklist reads all_max_dd_within_limits: false — that flag is honest, not broken. The six: Itop3_Q −60.3%, Ntop5_M −56.4%, Itop5_M −54.9%, Ntop3_M −54.2%, Itop3_M −52.7%, Itop5_Q −50.5%. Every other published variant is inside the limit.
- 30 variants implement 26 distinct strategies. Four pairs differ only by tax wrapper (IRA vs NET), not by holdings, and NET series are published pre-tax. Counting those pairs as independent strategies overstates the diversity of the fleet.
What's on this page
- Summary — Test period performance table (CAGR, Sharpe, Max DD, Calmar, Win Rate) and CAGR bar chart
- Annual Returns — Year-by-year test period returns (2020–2025) per variant with cross-year average
- Monthly Breakdown — Expected monthly return heatmap, monthly volatility heatmap, annual returns context table, cross-variant monthly bar chart, and good-vs-bad month expectation cards
- Overlay Impact — Crash overlay comparison: base vs overlay CAGR, CAGR cost, drawdown reduction, and risk-adjusted metrics (Calmar, Sharpe)
- Episodes — Roundtrip trade analysis: total episodes, win rate, avg win/loss, profit factor, best/worst episode, and max consecutive win/loss streaks
- Live Paper Trading — Real-time forward-test of every live variant since Jan 2026: portfolio value, return, position count, and status
What am I looking at?
Rolling test period results for … deployment variants tested on 2020–2025 data the model never saw during training. All returns are net of simulated taxes and transaction costs. Each variant was optimized on earlier data and tested forward. CAGR, Sharpe, Calmar, and Max Drawdown come from the test period equity curves.
| Variant | Name | Account | Test CAGR | vs SPY | Sharpe | Max DD | Calmar | Win Rate |
|---|
Test Period CAGR Comparison
The bottom line — how each variant's compound annual growth rate stacks up side by side. The dashed line is a static SPY allocation; bars above it are outperforming the benchmark.
What am I looking at?
Year-by-year test period returns for each variant (2020–2025). These are calendar-year returns from the rolling test period, net of taxes. 2022 was a bear market (most strategies had drawdowns). 2020 and 2023 were strong recovery/bull years. This shows how each variant performs across different market regimes.
| Variant | Name | Account | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | Avg |
|---|
Variants starting mid-year show partial-year returns which may appear unusually high or low. "--" indicates the variant was not active that year.
What am I looking at?
Monthly return patterns from the 6-year test period (2020–2025). Each cell shows the expected monthly return based on historical rebalance-period returns in that calendar month. Use this to set expectations: some months are historically strong (e.g., Nov, Dec) while others are weak (Feb, Sep). Compare against actual annual returns to understand how monthly patterns compound into yearly performance. Volatility shows the typical dispersion of returns within each month.
Expected Monthly Returns (Test Period 2020-2025)
Color intensity = magnitude. Derived from avg rebalance-period returns per calendar month.
| Variant | Acct | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec | Avg/Mo |
|---|
Monthly Volatility (Risk per Month)
Higher volatility = wider range of outcomes. Helps identify which months have the most uncertain returns.
| Variant | Acct | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec | Avg |
|---|
Annual Returns (for Context)
How monthly patterns compound into yearly results. Good months need to overcome bad months + taxes + slippage.
| Variant | Acct | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | CAGR | Best Yr | Worst Yr |
|---|
Cross-Variant Monthly Summary
Averaged across all … variants with a scorecard. Shows which months are systematically strong or weak for the strategy.
Setting Expectations: Good vs Bad Months
Why this matters — knowing which months are historically weak helps avoid panic-selling during normal seasonal dips. A −2% month in September is expected behavior, not a broken strategy.
What am I looking at?
Crash overlay impact on each variant. Overlays automatically reduce position size during detected crash regimes to protect capital. G2 Aggressive = reduces to 20% exposure in crashes, G1 Balanced = 30% min, G Light = 50% min. Table compares each variant's no-overlay baseline vs recommended overlay. Overlays typically sacrifice 5–10% CAGR but cut max drawdown by 50–75%, dramatically improving risk-adjusted returns (Calmar ratio).
| Variant | Overlay | Base CAGR | Overlay CAGR | CAGR Cost | Base MaxDD | Overlay MaxDD | DD Reduction | Overlay Calmar | Overlay Sharpe |
|---|
What am I looking at?
Roundtrip episode analysis for the full test period (2020–2025). Each "episode" is one complete hold period from buy to sell for a single stock position. Win Rate = % of episodes that were profitable. Profit Factor = gross wins / gross losses (above 1.0 means profitable system). Avg Win/Loss shows the typical magnitude of winning vs losing trades. Max consecutive wins/losses shows streak behavior.
| Variant | Total Episodes | Win Rate | Avg Win | Avg Loss | Profit Factor | Best Episode | Worst Episode | Max Consec W | Max Consec L |
|---|
What am I looking at?
Real-time paper trading of all … live variants since Jan 2, 2026. Each variant starts with positions carried from the 2025 backtest, normalized to $100,000. This is forward-testing to validate backtest results in live market conditions.
Page last updated: loading...