Core balanced
RELEASE CANDIDATE
NET RETURN
48.2%
NET SHARPE
1.298
SORTINO
2.678
MAX DRAWDOWN
-49.6%
Shrinkage Max Diversification · 25% signal blend · Legacy Risk Scaled
Research record
The release architecture is the result of controlled walk-forward experiments—not a collection of parameters chosen because they looked sophisticated. This page shows what improved, what failed, and what remains uncertain.
Release evidence
Aug 11, 2026, 11:14 PM UTC
Regenerated from committed reports
Ranking artifact
Jun 10, 2026
25 names · committed · not live
Candidate artifact
Jul 10, 2026
20 monitored names · not live
Execution status
Research only
No brokerage connection or live order routing
Three operating mandates
Core balanced
RELEASE CANDIDATE
NET RETURN
48.2%
NET SHARPE
1.298
SORTINO
2.678
MAX DRAWDOWN
-49.6%
Aggressive
STATIC 1.00X RESEARCH
NET RETURN
79.3%
NET SHARPE
1.222
SORTINO
2.819
MAX DRAWDOWN
-73.9%
Defensive
MINIMUM-VARIANCE ANCHOR
NET RETURN
45.5%
NET SHARPE
1.281
SORTINO
2.525
MAX DRAWDOWN
-44.0%
Core net return
48.2%
Simulated annualized net result
Core Sharpe
1.298
Risk-adjusted return across the full record
Core Sortino
2.678
Downside-risk-adjusted result
Core max drawdown
-49.6%
Historical simulated peak-to-trough decline
Annual out-of-sample record
Annual expanding-window fits produce a year-by-year view of the same locked architecture. This helps expose whether the result depends on one unusually favorable regime.
| Year | Core return | Core Sharpe | Aggressive return | Defensive return | Core drawdown |
|---|---|---|---|---|---|
| 2021 | -18.2% | -0.472 | -4.4% | -15.6% | -33.4% |
| 2022 | -19.6% | -0.614 | -44.7% | -14.3% | -26.9% |
| 2023 | 129.1% | 2.314 | 301.1% | 134.2% | -16.2% |
| 2024 | 77.1% | 1.670 | 97.7% | 71.2% | -11.9% |
| 2025 | 166.9% | 3.320 | 298.9% | 128.8% | -8.2% |
| 2026 | 64.7% | 1.620 | 164.4% | 52.3% | -12.6% |
Controlled decisions
Every decision below corresponds to a committed comparison. Failed hypotheses stay visible because a credible research platform records what did not work.
Did expanding the portfolio universe to roughly 2,000 names improve out-of-sample performance?
The broader point-in-time universe reduced drawdown in the risk-scaled policy, but materially weakened annualized return, Sharpe, and ranking IC. More names did not create a better portfolio model.
Decision
Retain Liquid-500 for Salarium 1.0 portfolio construction; keep broad coverage as a separate discovery and research funnel.
30.0%
Liquid-500 return
19.9%
Broad return
-0.167
Sharpe delta
-0.018
Broad IC
Was the original five-day target and five-day rebalance cadence too short and too active?
Separating prediction horizon from trading cadence showed that Salarium's signal is slower-moving. The 20D model traded every 10 days improved both return and Sharpe versus the original 5D/5D design.
Decision
Lock a 20-trading-day model horizon and 10-trading-day rebalance cadence.
74.5%
20D/10D return
40.5%
5D/5D return
+1.170
20D/10D Sharpe
+0.866
5D/5D Sharpe
Could a broader 20–75 name portfolio preserve alpha while reducing risk?
Additional breadth reduced volatility and turnover, but diluted return faster than it improved risk-adjusted performance. The model's useful alpha remained concentrated near the top of the ranking.
Decision
Keep Top-10 concentration with a rank-15 persistence buffer; manage joint risk through covariance rather than indiscriminate breadth.
74.5%
Top-10 return
35.6%
Top-75 return
+1.170
Top-10 Sharpe
+1.013
Top-75 Sharpe
Could covariance-aware construction preserve concentrated alpha while reducing redundant correlated risk?
A 60D Ledoit-Wolf maximum-diversification portfolio improved Sharpe and Sortino while modestly improving drawdown versus inverse-volatility weighting. The optimizer completed without fallback in the selected configuration.
Decision
Use 60D shrinkage maximum diversification as the primary risk anchor; retain minimum variance as the defensive comparator.
+1.216
Max-div Sharpe
+1.170
Inverse-vol Sharpe
75.4%
Max-div return
0.0%
Fallback rate
Should conviction influence weights after Top-10 selection and covariance optimization?
A 25% signal blend increased the balanced mandate's simulated return while leaving overall Sharpe nearly unchanged. Higher blends continued to raise return but progressively increased volatility and drawdown.
Decision
Blend 25% signal-aware weights with 75% covariance-risk weights under the 18% single-name cap.
48.2%
25% return
46.1%
0% return
+1.298
25% Sharpe
-49.6%
25% drawdown
Did the evidence justify using portfolio exposure above 1.00x?
The selected portfolio and exposure policies did not require leverage above 1.00x in the committed evaluation. The risk layer found more value in de-risking than in borrowing additional capital.
Decision
Retain a hard 1.25x governance ceiling as permission—not a target—and keep the selected mandate unlevered unless future risk evidence earns additional exposure.
+1.250
Hard ceiling
+1.000
Observed max
0.0%
Leveraged periods
+0.534
Average exposure
Covariance constructor
The 60-day shrinkage covariance tournament compared the original inverse-volatility baseline with minimum-variance and maximum-diversification portfolios while holding the upstream alpha signal fixed.
Portfolio constructor
Return
75.4%
Sharpe
1.216
Max drawdown
-72.9%
Volatility
61.2%
Portfolio constructor
Return
72.7%
Sharpe
1.215
Max drawdown
-64.7%
Volatility
58.7%
Portfolio constructor
Return
74.5%
Sharpe
1.170
Max drawdown
-75.3%
Volatility
65.2%
Portfolio constructor
Return
71.0%
Sharpe
1.156
Max drawdown
-73.7%
Volatility
63.0%
Signal-aware weighting
A 25% signal blend gives model conviction a meaningful vote while preserving the covariance engine as the primary portfolio-risk anchor.
Robustness
The selected blend is evaluated against the same-anchor 0% signal baseline across six annual test periods. The record is mixed rather than universal, which is why the signal share remains governed at 25%.
Each signal blend is compared with the same risk anchor at 0% signal influence.
| Blend | Sharpe wins | Return wins | Median Sharpe Δ | Median return Δ |
|---|---|---|---|---|
| 0% | 0/6 | 0/6 | 0.000 | 0.0% |
| 25% | 3/6 | 3/6 | -0.004 | 1.4% |
| 50% | 3/6 | 3/6 | -0.006 | 2.7% |
| 75% | 3/6 | 3/6 | -0.007 | 4.0% |
Each signal blend is compared with the same risk anchor at 0% signal influence.
| Blend | Sharpe wins | Return wins | Median Sharpe Δ | Median return Δ |
|---|---|---|---|---|
| 0% | 0/6 | 0/6 | 0.000 | 0.0% |
| 25% | 4/6 | 3/6 | +0.008 | 4.8% |
| 50% | 4/6 | 3/6 | +0.016 | 9.6% |
| 75% | 4/6 | 3/6 | +0.024 | 14.3% |
Research limitations
All displayed performance is historical research, not live account performance.
The current universe process retains a documented survivorship-bias limitation.
Multiple experiments increase the chance of choosing patterns that may not persist.
Borrow costs, market impact, taxes, and capacity can differ from research assumptions.
A 60-day risk estimate can fail when correlations change abruptly.
The system does not know any visitor's objectives, constraints, or risk tolerance.