—
In plain English
Technical definition
How to interpret this result
📊 Statistical Validation
⚠️ VYRA Has Two Fundamentally Different Use Cases — Do Not Confuse Them
Use Case 1 — Operational Strategy & Planning
The question: "What should management do to move TSR / Stock Price / Earnings Quality by a specific amount, by a specific deadline?"
VYRA exhaustively generates and tests every viable operational strategy across a proprietary set of operational levers and ranks them by projected impact. It uses a causal model built from trailing fundamentals: the same drivers that historically moved a company's stock are weighted by their proven responsiveness, not by statistical correlation alone.
How VYRA earns your trust for Case 1:
• Company-specific calibration (Extreme Intelligence): Every report builds a Company Intelligence Profile — the model re-calibrates coefficient weights from that company's own 10-year historical driver data, so the response functions reflect that specific company's dynamics, not just sector averages.
• Seven risk dimensions: Each plan is stress-tested across execution risk, competitive risk, macro regime risk, timing risk, model uncertainty, concentration risk, and scenario sensitivity — not just a single point estimate.
• Monte Carlo confidence intervals: 500 simulation draws produce a full probability distribution over outcomes. The reported range (e.g. "18–34% TSR over 6 quarters") is a calibrated 80% confidence interval, not a best guess.
• Observed-driver backtest: Using actual historical driver movements as inputs, VYRA's projection engine is tested against realized outcomes. Directional accuracy: 71.7% TSR · 64.1% Stock Price · 62.0% Earnings Quality (2022 calibration baseline, Fortune 100).
• Execution premium validation: Companies that executed on VYRA's recommendations outperformed those that didn't — top recommendation: mean +7.4pp TSR vs S&P; broad recommendation execution: mean +11.2pp — positive in 9 of 9 tested periods (2014–2024).
VYRA exhaustively generates and tests every viable operational strategy across a proprietary set of operational levers and ranks them by projected impact. It uses a causal model built from trailing fundamentals: the same drivers that historically moved a company's stock are weighted by their proven responsiveness, not by statistical correlation alone.
How VYRA earns your trust for Case 1:
• Company-specific calibration (Extreme Intelligence): Every report builds a Company Intelligence Profile — the model re-calibrates coefficient weights from that company's own 10-year historical driver data, so the response functions reflect that specific company's dynamics, not just sector averages.
• Seven risk dimensions: Each plan is stress-tested across execution risk, competitive risk, macro regime risk, timing risk, model uncertainty, concentration risk, and scenario sensitivity — not just a single point estimate.
• Monte Carlo confidence intervals: 500 simulation draws produce a full probability distribution over outcomes. The reported range (e.g. "18–34% TSR over 6 quarters") is a calibrated 80% confidence interval, not a best guess.
• Observed-driver backtest: Using actual historical driver movements as inputs, VYRA's projection engine is tested against realized outcomes. Directional accuracy: 71.7% TSR · 64.1% Stock Price · 62.0% Earnings Quality (2022 calibration baseline, Fortune 100).
• Execution premium validation: Companies that executed on VYRA's recommendations outperformed those that didn't — top recommendation: mean +7.4pp TSR vs S&P; broad recommendation execution: mean +11.2pp — positive in 9 of 9 tested periods (2014–2024).
Use Case 2 — Investment Ranking & Trading Signal
The question: "Which companies should I go long or short on? Who will outperform their sector peers over the next 6–12 months?"
VYRA ranks companies cross-sectionally by their fundamental quality and projected relative outperformance. This is not a question about what management should do — it is a question about which companies are most attractively positioned based on current fundamental data. The signal is a ranking and direction, not a price target.
How VYRA earns your trust for Case 2:
• Spearman ρ always positive: Rank correlation between VYRA's projected ranking and realized returns is positive in all 11 out-of-sample periods (2014–2024), across both the S&P 500 (503 co.) and Fortune 100 secondary universe (92 co.). ρ is above zero in every year; the S&P 500 results are the primary institutional story.
• Directional accuracy 63–75%: Every single year, every domain, VYRA correctly identifies whether a company will outperform or underperform sector peers — well above the 50% random baseline, including in trade wars, COVID, and the 2021 tightening cycle.
• K-fold walk-forward validation: ρ positive in all 10 held-out periods across 5 independent folds — the gold standard against overfitting.
• ML enhancement: Proprietary ensemble meta-learner lifts Spearman ρ by +120% relative and directional accuracy by +17.1pp. Monthly rebalance fund backtest (102 periods, 2016–2024): net Sharpe 1.17 · 0/102 eval. windows negative · monthly IC 0.1146, t=7.24 (60 independent folds) · 200× LP multiple (2/20 fees). June annual strip: net Sharpe 1.14, MaxDD 0% (0/10 annual periods negative).
• Factor independence: Residual ρ positive in all 11 tested periods after controlling for Value, Momentum, and Quality — statistically significant in the majority; VYRA is not a repackaging of known free factors.
VYRA ranks companies cross-sectionally by their fundamental quality and projected relative outperformance. This is not a question about what management should do — it is a question about which companies are most attractively positioned based on current fundamental data. The signal is a ranking and direction, not a price target.
How VYRA earns your trust for Case 2:
• Spearman ρ always positive: Rank correlation between VYRA's projected ranking and realized returns is positive in all 11 out-of-sample periods (2014–2024), across both the S&P 500 (503 co.) and Fortune 100 secondary universe (92 co.). ρ is above zero in every year; the S&P 500 results are the primary institutional story.
• Directional accuracy 63–75%: Every single year, every domain, VYRA correctly identifies whether a company will outperform or underperform sector peers — well above the 50% random baseline, including in trade wars, COVID, and the 2021 tightening cycle.
• K-fold walk-forward validation: ρ positive in all 10 held-out periods across 5 independent folds — the gold standard against overfitting.
• ML enhancement: Proprietary ensemble meta-learner lifts Spearman ρ by +120% relative and directional accuracy by +17.1pp. Monthly rebalance fund backtest (102 periods, 2016–2024): net Sharpe 1.17 · 0/102 eval. windows negative · monthly IC 0.1146, t=7.24 (60 independent folds) · 200× LP multiple (2/20 fees). June annual strip: net Sharpe 1.14, MaxDD 0% (0/10 annual periods negative).
• Factor independence: Residual ρ positive in all 11 tested periods after controlling for Value, Momentum, and Quality — statistically significant in the majority; VYRA is not a repackaging of known free factors.
Why this distinction matters:
Use Case 1 is about what to do — the output is a ranked action plan calibrated to a specific company and target. The validation metric is directional accuracy of the causal model and execution premium (do companies that follow the advice outperform?).
Use Case 2 is about who to invest in — the output is a cross-sectional ranking of companies by relative attractiveness. The validation metric is Spearman ρ, L/S spread, and portfolio Sharpe.
Conflating them leads to wrong conclusions: judging Use Case 2 (ranking signal) by point-forecast accuracy is wrong; judging Use Case 1 (planning model) by portfolio Sharpe is wrong. The rest of this page validates both — read the labels carefully.
📊 VYRA Scorecard System — How to Read It Correctly
Primarily a Use Case 2 tool — for investment ranking and trading decisions, not operational planning
What the Vyra Score Measures
The Vyra Score (0–100) is a fundamentals quality composite — how strong is this company across five dimensions relative to S&P 500 or R2K peers? It is not a return forecast and not a price target.
Five sub-scores and weights:
• Fundamentals quality (25%) — proprietary quality metrics vs sector median
• Growth trajectory (20%) — proprietary growth metrics vs sector median
• Valuation (15%) — P/E relative to sector peers (cheaper = higher score)
• Analyst consensus (20%) — net Wall Street buy/sell/hold sentiment, bull/bear ratio, analyst coverage count
• VYRA TSR outlook (20%) — VYRA's projected TSR vs sector median, plus ML rank signal
All sub-scores are relative to sector peers, not absolute. A score of 50 means exactly at the sector median. Score of 80 means top tier across all five dimensions.
Five sub-scores and weights:
• Fundamentals quality (25%) — proprietary quality metrics vs sector median
• Growth trajectory (20%) — proprietary growth metrics vs sector median
• Valuation (15%) — P/E relative to sector peers (cheaper = higher score)
• Analyst consensus (20%) — net Wall Street buy/sell/hold sentiment, bull/bear ratio, analyst coverage count
• VYRA TSR outlook (20%) — VYRA's projected TSR vs sector median, plus ML rank signal
All sub-scores are relative to sector peers, not absolute. A score of 50 means exactly at the sector median. Score of 80 means top tier across all five dimensions.
Ratings Scale
Strong Buy Score 80–100 · Grade A/A+
Top-tier fundamentals, strong analyst support, positive VYRA outlook, and attractive relative valuation — all aligning.
Buy Score 65–79 · Grade B+/B
Constructive overall — above-median on most dimensions; may have one offset (e.g. slightly elevated valuation).
Hold Score 45–64 · Grade C+/C
Mixed picture — strong in some dimensions, weak in others. May reflect high TSR potential but offsetting risks, or solid fundamentals but limited near-term catalyst.
Sell Score 30–44 · Grade D
Below-median on multiple dimensions; fundamentals, growth, or analyst sentiment is deteriorating.
Strong Sell Score <30 · Grade F
Bottom tier across most dimensions; weak fundamentals, negative analyst consensus, and poor outlook.
Top-tier fundamentals, strong analyst support, positive VYRA outlook, and attractive relative valuation — all aligning.
Buy Score 65–79 · Grade B+/B
Constructive overall — above-median on most dimensions; may have one offset (e.g. slightly elevated valuation).
Hold Score 45–64 · Grade C+/C
Mixed picture — strong in some dimensions, weak in others. May reflect high TSR potential but offsetting risks, or solid fundamentals but limited near-term catalyst.
Sell Score 30–44 · Grade D
Below-median on multiple dimensions; fundamentals, growth, or analyst sentiment is deteriorating.
Strong Sell Score <30 · Grade F
Bottom tier across most dimensions; weak fundamentals, negative analyst consensus, and poor outlook.
Why VYRA Diverges from Wall Street Analyst Ratings
Analysts consistently over-estimate. Academic research (Bradshaw et al., consistent since 1990s) shows sell-side analysts are structurally bullish — investment banking relationships, career risk, and access preservation all bias ratings upward. The long-run average: fewer than 10% of analyst ratings are Sell or Strong Sell at any given time, despite the mathematical fact that roughly half of all stocks underperform.
VYRA uses trailing fundamentals. The Vyra Score and TSR projection are built from proprietary trailing TTM fundamentals. VYRA does not incorporate analyst price targets, forward guidance, or recent news events. This means:
• A company with a recent analyst upgrade not yet reflected in TTM data will score lower than its current consensus suggests
• A company with declining fundamentals but still-positive analyst consensus will score lower than analysts rate it
• VYRA's scores update daily (analyst sub-score) and weekly (fundamentals) — not in real time
VYRA uses trailing fundamentals. The Vyra Score and TSR projection are built from proprietary trailing TTM fundamentals. VYRA does not incorporate analyst price targets, forward guidance, or recent news events. This means:
• A company with a recent analyst upgrade not yet reflected in TTM data will score lower than its current consensus suggests
• A company with declining fundamentals but still-positive analyst consensus will score lower than analysts rate it
• VYRA's scores update daily (analyst sub-score) and weekly (fundamentals) — not in real time
VYRA Is Conservative By Design
VYRA projects floor estimates, not targets. The TSR projection is a fundamentals-derived floor — how much return is supported by current trailing fundamentals alone, without relying on analyst optimism, narrative momentum, or speculative multiple expansion.
What this means in practice:
• A high-momentum stock with weak TTM fundamentals will score lower than its price action suggests
• VYRA's projected TSR is almost always lower than Wall Street price targets in bull markets — this is intentional
• The measured bias: −15pp (fundamentals only) / −6pp (ML-enhanced) vs realized returns
• This is a feature: a signal that under-promises and whose portfolio outperforms the forecast is safer than an overconfident model
The strongest signal is when VYRA Score and TSR projection both agree: a score of 80+ AND projected TSR above sector median means VYRA's causal model and its quality composite are pointing in the same direction.
What this means in practice:
• A high-momentum stock with weak TTM fundamentals will score lower than its price action suggests
• VYRA's projected TSR is almost always lower than Wall Street price targets in bull markets — this is intentional
• The measured bias: −15pp (fundamentals only) / −6pp (ML-enhanced) vs realized returns
• This is a feature: a signal that under-promises and whose portfolio outperforms the forecast is safer than an overconfident model
The strongest signal is when VYRA Score and TSR projection both agree: a score of 80+ AND projected TSR above sector median means VYRA's causal model and its quality composite are pointing in the same direction.
How to interpret divergence between Vyra Score and TSR projection:
These measure different things and should sometimes diverge — that is not a model inconsistency.
High score + high TSR projection: Strongest buy signal — quality fundamentals and strong forward return signal align.
High score + modest TSR projection: Fundamentally strong company, but near-term return is constrained (e.g. high valuation priced in, limited catalyst). Good Hold for quality-focused investors.
Low score + high TSR projection: Possible mean-reversion or turnaround play — attractive return potential but with execution, leverage, or analyst risk as offset. Requires deeper analysis.
Low score + low TSR projection: VYRA's strongest sell signal — fundamentals and return signal agree the company is underperforming peers.
High score + high TSR projection: Strongest buy signal — quality fundamentals and strong forward return signal align.
High score + modest TSR projection: Fundamentally strong company, but near-term return is constrained (e.g. high valuation priced in, limited catalyst). Good Hold for quality-focused investors.
Low score + high TSR projection: Possible mean-reversion or turnaround play — attractive return potential but with execution, leverage, or analyst risk as offset. Requires deeper analysis.
Low score + low TSR projection: VYRA's strongest sell signal — fundamentals and return signal agree the company is underperforming peers.
What company scorecards are NOT for:
Scorecards do not replace a full VYRA evaluation (Use Case 1). A scorecard answers "how does this company's fundamentals compare to peers?" — it does not answer "what should management do to improve the stock price?" For that, run a full evaluation: VYRA will generate specific operational strategies, ranked by probability of success, with company-specific calibration and seven risk dimensions. The scorecard is a fast, deterministic snapshot; the full evaluation is a deep causal analysis.
Sector scorecards aggregate the same five dimensions across all companies in a sector. They answer: "which sectors have the strongest fundamentals right now?" — useful for sector rotation and thematic positioning, not for individual stock decisions.
Sector scorecards aggregate the same five dimensions across all companies in a sector. They answer: "which sectors have the strongest fundamentals right now?" — useful for sector rotation and thematic positioning, not for individual stock decisions.
📅 Backtest Validity — Data Window and Look-Ahead Bias
Q: "Don't you need more historical data for a valid backtest?"
One year of calibration, ten years of genuine out-of-sample testing. The model was calibrated on a single baseline: 2022-06-30 only. Every other period — all 10 remaining years from 2014 to 2024 — is genuinely out-of-sample data the model never saw during calibration. This is a strict standard. Many published backtests use data throughout the test period for calibration.
K-fold cross-validation is the gold standard against overfitting — and it confirms the signal. In 5-fold walk-forward validation on 503 S&P 500 companies, the model was calibrated on 8 years and tested on 2 completely held-out years it had never seen. Spearman ρ was positive in all 10 held-out periods across all 5 folds — zero failures. This is the definitive statistical test that the signal is real and not a product of overfitting to any particular era.
The 11-year window already spans every major macro regime. Steady growth, China scare, trade wars, COVID crash, COVID bull, the sharpest Fed tightening cycle in 40 years, the AI narrative boom. Spearman ρ is positive in every single one. A signal that holds across all of those regimes is not era-specific.
More data is additive, not corrective. Pre-2014 data could modestly improve calibration for specific crisis regimes like 2008 or the dotcom bust. But it would not change the validity of results we already have — those are validated by the k-fold and the regime diversity we already cover.
One year of calibration, ten years of genuine out-of-sample testing. The model was calibrated on a single baseline: 2022-06-30 only. Every other period — all 10 remaining years from 2014 to 2024 — is genuinely out-of-sample data the model never saw during calibration. This is a strict standard. Many published backtests use data throughout the test period for calibration.
K-fold cross-validation is the gold standard against overfitting — and it confirms the signal. In 5-fold walk-forward validation on 503 S&P 500 companies, the model was calibrated on 8 years and tested on 2 completely held-out years it had never seen. Spearman ρ was positive in all 10 held-out periods across all 5 folds — zero failures. This is the definitive statistical test that the signal is real and not a product of overfitting to any particular era.
The 11-year window already spans every major macro regime. Steady growth, China scare, trade wars, COVID crash, COVID bull, the sharpest Fed tightening cycle in 40 years, the AI narrative boom. Spearman ρ is positive in every single one. A signal that holds across all of those regimes is not era-specific.
More data is additive, not corrective. Pre-2014 data could modestly improve calibration for specific crisis regimes like 2008 or the dotcom bust. But it would not change the validity of results we already have — those are validated by the k-fold and the regime diversity we already cover.
Q: "Does the algorithm use future data when making historical predictions? (look-ahead bias)"
This is the right question to ask — look-ahead bias is the most common way backtests are accidentally made to look better than they are. The answer here is no, by construction. Here's precisely how the separation is enforced:
Only data that existed on the baseline date is used. At each backtest baseline (e.g. June 30, 2016), the model reads only proprietary trailing twelve-month fundamentals available on that date — metrics from prior quarters already reported. No data after the baseline date is ever used as an input. The model does not know what happens next.
Walk-forward validation is exactly what k-fold cross-validation enforces. In each fold, the model is calibrated on past years only, then tested on future held-out years it has never seen. This is the textbook method for preventing look-ahead bias in time-series models. The held-out years are strictly future relative to the calibration window — the model cannot have learned from them because it was explicitly denied access during calibration.
The k-fold result is the proof, not just the claim. If look-ahead bias were present, the signal would be strong in-sample and collapse out-of-sample. Instead, Spearman ρ is positive in all 10 held-out periods across all 5 folds — including periods that were future relative to any calibration data the model ever saw. A signal with look-ahead bias cannot maintain consistent positive ρ on data it genuinely could not have seen. This one does.
This is the right question to ask — look-ahead bias is the most common way backtests are accidentally made to look better than they are. The answer here is no, by construction. Here's precisely how the separation is enforced:
Only data that existed on the baseline date is used. At each backtest baseline (e.g. June 30, 2016), the model reads only proprietary trailing twelve-month fundamentals available on that date — metrics from prior quarters already reported. No data after the baseline date is ever used as an input. The model does not know what happens next.
Walk-forward validation is exactly what k-fold cross-validation enforces. In each fold, the model is calibrated on past years only, then tested on future held-out years it has never seen. This is the textbook method for preventing look-ahead bias in time-series models. The held-out years are strictly future relative to the calibration window — the model cannot have learned from them because it was explicitly denied access during calibration.
The k-fold result is the proof, not just the claim. If look-ahead bias were present, the signal would be strong in-sample and collapse out-of-sample. Instead, Spearman ρ is positive in all 10 held-out periods across all 5 folds — including periods that were future relative to any calibration data the model ever saw. A signal with look-ahead bias cannot maintain consistent positive ρ on data it genuinely could not have seen. This one does.
Verified Claims
What this section proves: Every headline number we publish about VYRA's accuracy is automatically re-derived from the live database and checked against the published value. A ✓ verified badge means the number reproduces within statistical tolerance — it is not hand-picked or cherry-picked. The tolerance accounts for Monte Carlo variance (±1–2pp on directional accuracy, ±2–4pp on CI coverage). This is adversarial self-testing.
The Statistical Witness Package reproduces every claim below from scratch against the live database. ✓ verified = reproduced value within tolerance of published claim. ⏳ pending = not included in last run (run full witness to verify). ✗ fail = reproduced value diverges beyond tolerance.
•••
Loading claims...Spearman ρ Across All 11 Periods (F100 92 co vs S&P 500 503 co)
How to read this table: Spearman ρ measures whether VYRA's ranking order of companies matches the actual ranking order by realized TSR. ρ = +1.0 means perfect rank order. ρ = 0 means no relationship. ρ = −1.0 means perfect inverse. A ρ of +0.25–0.40 in finance is considered strong — hedge funds with ρ above +0.10 run multi-billion dollar strategies on it.
The headline result (S&P 500): ρ is positive in all 11 periods for both S&P 500 and F100 universe — including trade wars, COVID, and the 2021 tightening cycle. Low ρ in some years (e.g. 2015, 2018) reflects macro-dominated years with less cross-sectional fundamental dispersion — not model failure. The S&P 500 strip is the primary institutional evidence; F100 is a secondary, smaller universe (92 names). In the S&P sweep, directional accuracy is above 50% in every period; the permutation null is exceeded in all 11 years. "ns" p-value years have noisier samples but the signal remains positive.
Why it matters for investors: Persistent positive ρ across 11 diverse regimes means the fundamental ranking signal is real and structural. Even at ρ = +0.13 (2018), sorting by VYRA rank produces meaningfully better-than-random portfolio construction.
The headline result (S&P 500): ρ is positive in all 11 periods for both S&P 500 and F100 universe — including trade wars, COVID, and the 2021 tightening cycle. Low ρ in some years (e.g. 2015, 2018) reflects macro-dominated years with less cross-sectional fundamental dispersion — not model failure. The S&P 500 strip is the primary institutional evidence; F100 is a secondary, smaller universe (92 names). In the S&P sweep, directional accuracy is above 50% in every period; the permutation null is exceeded in all 11 years. "ns" p-value years have noisier samples but the signal remains positive.
Why it matters for investors: Persistent positive ρ across 11 diverse regimes means the fundamental ranking signal is real and structural. Even at ρ = +0.13 (2018), sorting by VYRA rank produces meaningfully better-than-random portfolio construction.
Rank correlation between VYRA projected TSR and actual TSR. Positive in every single period for both universes. Calibration period (2022) marked †. p-values from scipy.stats.spearmanr; bootstrap CIs from 10,000 resamples.
•••
Backtest Charts
Chart guide — what each one shows and the right interpretation:
01 — Spearman ρ by period: The core ranking signal bar chart. Every bar above zero means VYRA correctly ranked companies that year. Never negative across all 10 periods. The 2015/2017 short bars reflect macro-dominated years with low cross-sectional return dispersion — hard for any model, but the signal still works.
02 — L/S spread by period: Alpha of top-quartile minus bottom-quartile vs S&P. Positive in 9 of 10 periods. The 2020 spike (+90pp) is a COVID outlier where cross-sector dispersion was extreme — the mean excluding 2020 is +12.6pp, which is the representative number. 2018/2019 near-zero reflects trade-war-compressed dispersion, not model failure.
03 — Execution premium: Companies that followed VYRA's recommendations outperformed those that didn't — top recommendation: mean +7.4pp (9/9 periods); broad recommendation execution: mean +11.2pp (9/9 periods, floor +4.0pp). Important caveat: correlation, not causation — well-managed companies may both execute well and outperform for independent reasons.
04 — Hit rate vs Spearman ρ scatter: Illustrates why hit rate (% of stocks beating index) is a misleading metric. In 2020, most stocks couldn't beat the S&P +38.6% — a low hit rate says nothing about model quality. ρ is the honest metric; the chart shows they are different signals and neither dominates.
05 — Factor decomposition: VYRA's residual ρ after controlling for value (1/PE), momentum (6-month return), and quality (ROE). In 2022, all three known factors have negative or near-zero ρ — standard factor models failed. VYRA's residual ρ = +0.247 (statistically significant) even after removing factor structure. This proves VYRA captures something factors miss.
06 — Mode 1 vs Mode 2 hit rates: Without a macro scenario (Mode 1), hit rate in 2020 is 19.8% — most stocks fell below the S&P during COVID. With the COVID recovery scenario (Mode 2), hit rate jumps to 91.7%. This is what scenario context adds: when you know the macro regime, the signal amplifies.
07 — All-scenario stress matrix: Mean projected TSR across 26 scenarios for the 2017 baseline universe. The blue band shows the min/max range across companies. Use this to understand scenario sensitivity, not as a point forecast — wide bands mean high cross-sectional dispersion, not uncertainty in the model.
08 — S&P 500 Spearman ρ by year: The cross-sectional ρ for the full 500-stock universe. All values positive 2014–2024. The 2024 value (+0.31) is the strongest in the series — the signal has been improving as data quality and model complexity increase.
01 — Spearman ρ by period: The core ranking signal bar chart. Every bar above zero means VYRA correctly ranked companies that year. Never negative across all 10 periods. The 2015/2017 short bars reflect macro-dominated years with low cross-sectional return dispersion — hard for any model, but the signal still works.
02 — L/S spread by period: Alpha of top-quartile minus bottom-quartile vs S&P. Positive in 9 of 10 periods. The 2020 spike (+90pp) is a COVID outlier where cross-sector dispersion was extreme — the mean excluding 2020 is +12.6pp, which is the representative number. 2018/2019 near-zero reflects trade-war-compressed dispersion, not model failure.
03 — Execution premium: Companies that followed VYRA's recommendations outperformed those that didn't — top recommendation: mean +7.4pp (9/9 periods); broad recommendation execution: mean +11.2pp (9/9 periods, floor +4.0pp). Important caveat: correlation, not causation — well-managed companies may both execute well and outperform for independent reasons.
04 — Hit rate vs Spearman ρ scatter: Illustrates why hit rate (% of stocks beating index) is a misleading metric. In 2020, most stocks couldn't beat the S&P +38.6% — a low hit rate says nothing about model quality. ρ is the honest metric; the chart shows they are different signals and neither dominates.
05 — Factor decomposition: VYRA's residual ρ after controlling for value (1/PE), momentum (6-month return), and quality (ROE). In 2022, all three known factors have negative or near-zero ρ — standard factor models failed. VYRA's residual ρ = +0.247 (statistically significant) even after removing factor structure. This proves VYRA captures something factors miss.
06 — Mode 1 vs Mode 2 hit rates: Without a macro scenario (Mode 1), hit rate in 2020 is 19.8% — most stocks fell below the S&P during COVID. With the COVID recovery scenario (Mode 2), hit rate jumps to 91.7%. This is what scenario context adds: when you know the macro regime, the signal amplifies.
07 — All-scenario stress matrix: Mean projected TSR across 26 scenarios for the 2017 baseline universe. The blue band shows the min/max range across companies. Use this to understand scenario sensitivity, not as a point forecast — wide bands mean high cross-sectional dispersion, not uncertainty in the model.
08 — S&P 500 Spearman ρ by year: The cross-sectional ρ for the full 500-stock universe. All values positive 2014–2024. The 2024 value (+0.31) is the strongest in the series — the signal has been improving as data quality and model complexity increase.
Generated from live database by scripts/statistical_witness.py --charts. Regenerated weekly.
•••
Loading charts...Calibration — Cross-Sectional Ranking Quality (F100 + S&P 500, 2014–2024)
⚠️ Critical reading guide — do not misinterpret these tables or scatter plots:
This section contains the most easily misread data on the page. The scatter plots look like loose clouds and the benchmark hit rate numbers look low in some years. Here is the correct interpretation:
Column by column: ρ = Spearman rank correlation — the primary signal quality metric, always positive. Proj hurdle = % of companies where VYRA's projection exceeded the S&P (a measure of how bullish VYRA was, not a success/fail metric). Ex-post = % of companies that actually beat the S&P — this fluctuates with market conditions and is not a model accuracy metric. In 2020 (S&P +38.6%), few stocks beat the index; a low ex-post rate in 2020 tells you about market conditions, not VYRA's accuracy. Cal. = P(beat S&P | projected to beat) — calibration fidelity. Q10−Q1 = realized alpha of top decile minus bottom decile — the clearest portfolio metric. Dir acc = % of correct directional calls vs S&P — this is the right accuracy metric.
For the scatter plots below: The dashed 45° line is perfect point calibration (not the goal). The solid OLS fit line with positive slope is the signal. Wide scatter above the 45° is the conservative magnitude bias (projected lower than actual in bull markets). ρ = +0.26 across all F100 periods.
This section contains the most easily misread data on the page. The scatter plots look like loose clouds and the benchmark hit rate numbers look low in some years. Here is the correct interpretation:
Column by column: ρ = Spearman rank correlation — the primary signal quality metric, always positive. Proj hurdle = % of companies where VYRA's projection exceeded the S&P (a measure of how bullish VYRA was, not a success/fail metric). Ex-post = % of companies that actually beat the S&P — this fluctuates with market conditions and is not a model accuracy metric. In 2020 (S&P +38.6%), few stocks beat the index; a low ex-post rate in 2020 tells you about market conditions, not VYRA's accuracy. Cal. = P(beat S&P | projected to beat) — calibration fidelity. Q10−Q1 = realized alpha of top decile minus bottom decile — the clearest portfolio metric. Dir acc = % of correct directional calls vs S&P — this is the right accuracy metric.
For the scatter plots below: The dashed 45° line is perfect point calibration (not the goal). The solid OLS fit line with positive slope is the signal. Wide scatter above the 45° is the conservative magnitude bias (projected lower than actual in bull markets). ρ = +0.26 across all F100 periods.
Benchmark hit rate = % of names where VYRA projected TSR exceeded the S&P 500 over the same window.
Directional accuracy = % where the sign of projected alpha vs S&P matched realized alpha vs S&P.
Scatters: x = projected TSR (%), y = realized TSR (%). Regenerate: python scripts/build_quant_validation_artifacts.py.
•••
Segmentation (ρ, benchmark hit, directional)
How to read the segmentation tables: Consistent ρ across cap buckets and sectors means the signal is broad-based — not driven by mega-caps or a single sector. Where ρ looks low, check if N < 30 (flagged ⚠) — small samples are unreliable. A lower ρ in a given segment in a given year is expected noise, not evidence against the model.
•••
Execution Premium— 9 Periods (S&P 500) / 7 Periods (F100)
What this section proves — and its honest caveat:
Companies that executed on VYRA's recommendations outperformed those that didn't. Top recommendation: mean +7.4pp across 9 of 9 S&P 500 periods. Broad recommendation execution: mean +11.2pp across 9 of 9 periods, floor +4.0pp. This is not coincidence — the signal is positive in every tested year, including down markets (2021: +10.6pp), trade wars (2019: +14.2pp), and the Trump tariff era (2024: +15.5pp).
Important caveats: (1) This is correlation, not causation — well-managed companies may both execute on VYRA's recommendations and outperform for reasons beyond VYRA's model. (2) The execution classification is based on observable metric movements, not management intent. (3) The premium measures realized TSR alpha vs S&P, not absolute return. In 2016 (+1.9pp top; +7.6pp broad) cross-sector dispersion was unusually compressed — both signals still worked.
Companies that executed on VYRA's recommendations outperformed those that didn't. Top recommendation: mean +7.4pp across 9 of 9 S&P 500 periods. Broad recommendation execution: mean +11.2pp across 9 of 9 periods, floor +4.0pp. This is not coincidence — the signal is positive in every tested year, including down markets (2021: +10.6pp), trade wars (2019: +14.2pp), and the Trump tariff era (2024: +15.5pp).
Important caveats: (1) This is correlation, not causation — well-managed companies may both execute on VYRA's recommendations and outperform for reasons beyond VYRA's model. (2) The execution classification is based on observable metric movements, not management intent. (3) The premium measures realized TSR alpha vs S&P, not absolute return. In 2016 (+1.9pp top; +7.6pp broad) cross-sector dispersion was unusually compressed — both signals still worked.
Companies that executed on VYRA's recommendations vs those that didn't. S&P 500 top recommendation: 9 of 9 periods positive, mean +7.4pp. S&P 500 broad execution: 9 of 9 periods positive, mean +11.2pp, floor +4.0pp. F100: positive in 7 of 7 periods, mean +10.3pp. Alpha = median TSR minus S&P 500 TSR. Note: correlation not causation — better-managed companies may both execute and outperform.
•••
Use Case 1 Validation — Operational Strategy Model (causal fundamentals, CIP calibration, Monte Carlo)
How the operational model works and why you can trust it:
The causal model: VYRA's projection engine is not a statistical correlation machine — it is a causal driver model. For each company, it maps the relationship between proprietary key operational levers and their historical impact on TSR, Stock Price, and Earnings Quality. These relationships are derived from the company's own 10-year history, sector-level calibration data, and cross-company empirical benchmarks.
Company-specific calibration (Extreme Intelligence): Every VYRA analysis builds a Company Intelligence Profile (CIP) — the model recalibrates coefficient weights from that specific company's driver history. Coefficients are calibrated to each company's own operational sensitivities — the model is never one-size-fits-all.
Seven risk dimensions: Every plan projection is stress-tested across seven independent risk axes: (1) execution risk — how likely is the driver improvement to be achieved? (2) competitive response risk — how will peers react? (3) macro regime risk — does this plan work across tightening, expansion, and recession scenarios? (4) timing risk — what happens if the plan takes longer than projected? (5) model uncertainty — what is the variance in the coefficient estimates? (6) concentration risk — does the plan depend on a single driver? (7) scenario sensitivity — does the TSR estimate hold across the full scenario matrix?
Monte Carlo confidence intervals: Rather than a point estimate, VYRA runs 500 Monte Carlo draws sampling over coefficient uncertainty, execution probability distributions, and macro regime probabilities. The result is a full probability distribution over projected TSR/SP/EQ outcomes. The headline range (e.g. "18–34% TSR") is the calibrated 80% CI — empirically validated to contain the actual outcome in approximately 80–87% of cases (CI coverage: 87.0% TSR, 88.5% SP, 80.4% EQ on the 2022 calibration baseline).
Observed-driver backtest (the core validation): To test the causal model without forward-looking bias, VYRA feeds actual historical driver movements as inputs and measures whether the projected output matches realized outcomes. This is a pure test of the projection engine's causal accuracy — the results are shown in the "Model Accuracy — All Three Domains" section below.
Execution premium validation (the economic validation): The strongest real-world test of Use Case 1: do companies that executed on VYRA's recommendations outperform? Yes — top recommendation: +7.4pp mean (9/9 periods); broad recommendation execution: +11.2pp mean (9/9 periods, floor +4.0pp). Across both Fortune 100 and S&P 500 universes. This is the closest we can get to a prospective test of the planning model without a randomized controlled trial.
The causal model: VYRA's projection engine is not a statistical correlation machine — it is a causal driver model. For each company, it maps the relationship between proprietary key operational levers and their historical impact on TSR, Stock Price, and Earnings Quality. These relationships are derived from the company's own 10-year history, sector-level calibration data, and cross-company empirical benchmarks.
Company-specific calibration (Extreme Intelligence): Every VYRA analysis builds a Company Intelligence Profile (CIP) — the model recalibrates coefficient weights from that specific company's driver history. Coefficients are calibrated to each company's own operational sensitivities — the model is never one-size-fits-all.
Seven risk dimensions: Every plan projection is stress-tested across seven independent risk axes: (1) execution risk — how likely is the driver improvement to be achieved? (2) competitive response risk — how will peers react? (3) macro regime risk — does this plan work across tightening, expansion, and recession scenarios? (4) timing risk — what happens if the plan takes longer than projected? (5) model uncertainty — what is the variance in the coefficient estimates? (6) concentration risk — does the plan depend on a single driver? (7) scenario sensitivity — does the TSR estimate hold across the full scenario matrix?
Monte Carlo confidence intervals: Rather than a point estimate, VYRA runs 500 Monte Carlo draws sampling over coefficient uncertainty, execution probability distributions, and macro regime probabilities. The result is a full probability distribution over projected TSR/SP/EQ outcomes. The headline range (e.g. "18–34% TSR") is the calibrated 80% CI — empirically validated to contain the actual outcome in approximately 80–87% of cases (CI coverage: 87.0% TSR, 88.5% SP, 80.4% EQ on the 2022 calibration baseline).
Observed-driver backtest (the core validation): To test the causal model without forward-looking bias, VYRA feeds actual historical driver movements as inputs and measures whether the projected output matches realized outcomes. This is a pure test of the projection engine's causal accuracy — the results are shown in the "Model Accuracy — All Three Domains" section below.
Execution premium validation (the economic validation): The strongest real-world test of Use Case 1: do companies that executed on VYRA's recommendations outperform? Yes — top recommendation: +7.4pp mean (9/9 periods); broad recommendation execution: +11.2pp mean (9/9 periods, floor +4.0pp). Across both Fortune 100 and S&P 500 universes. This is the closest we can get to a prospective test of the planning model without a randomized controlled trial.
TSR CI Coverage
87.0%
80% CI target · F100, 2022 baseline
Execution Premium (9 periods)
+11.2pp
vs S&P · positive all 9 periods · SP500
TSR Directional Accuracy
71.7%
Observed-driver backtest · F100 2022
K-Fold Cross-Validation(S&P 500, 5 folds × 10 periods)
Why k-fold validation matters more than a single backtest:
In a single backtest, a model can look good if it was inadvertently overfit to one time period. K-fold walk-forward validation deliberately denies the model access to the test periods: the model is calibrated on 8 years, tested on 2 held-out years it has never seen. Five such folds cover all 10 years. ρ positive in all 10 held-out periods across all 5 folds — this is the gold standard of out-of-sample evidence.
How to read the fold bars: Each bar shows the held-out years for that fold, the mean ρ across those years, the L/S spread, and the hit rate. No fold has negative ρ or negative L/S spread. Fold 3 (2018/2019 held out) has the weakest ρ — those are the trade-war and late-cycle years with highest macro noise. Even so, ρ = +0.11 and L/S = +9.6pp.
In a single backtest, a model can look good if it was inadvertently overfit to one time period. K-fold walk-forward validation deliberately denies the model access to the test periods: the model is calibrated on 8 years, tested on 2 held-out years it has never seen. Five such folds cover all 10 years. ρ positive in all 10 held-out periods across all 5 folds — this is the gold standard of out-of-sample evidence.
How to read the fold bars: Each bar shows the held-out years for that fold, the mean ρ across those years, the L/S spread, and the hit rate. No fold has negative ρ or negative L/S spread. Fold 3 (2018/2019 held out) has the weakest ρ — those are the trade-war and late-cycle years with highest macro noise. Even so, ρ = +0.11 and L/S = +9.6pp.
2-period hold-out per fold. Model calibrated on 8 periods; tested on the 2 held-out. Spearman ρ and L/S spread positive in all 10 periods across all 5 folds.
•••
Permutation null — Spearman ρ(S&P 500 synthetic sweep)
Empirical null from 1,000 label shuffles per June baseline: is observed ρ above the null 95th percentile? Source: quant_validation_sp500_bundle.json.
•••
Factor Decomposition— Is VYRA Just a Known Factor?
What this proves — VYRA is not just a known factor in disguise:
A common critique of any quant signal is that it is just a repackaging of Value (low P/E), Momentum (recent price return), or Quality (high ROE) — factors that are free, well-known, and already priced in. This section controls for all three simultaneously using partial Spearman correlation.
The 2022 result is the most striking: Value ρ = −0.035 (ns), Momentum ρ = −0.328 (ns), Quality ρ = +0.062 (ns) — all three standard factors failed in 2022. VYRA's residual ρ after removing all three = +0.247 (statistically significant). The model was capturing something factors missed — specifically, proprietary operational fundamentals that predicted which companies would weather the tightening cycle.
2021/2017 also show significant residuals (p=0.050 and p=0.105 respectively). Across 3 of 3 tested periods, VYRA's signal persists after factor control.
A common critique of any quant signal is that it is just a repackaging of Value (low P/E), Momentum (recent price return), or Quality (high ROE) — factors that are free, well-known, and already priced in. This section controls for all three simultaneously using partial Spearman correlation.
The 2022 result is the most striking: Value ρ = −0.035 (ns), Momentum ρ = −0.328 (ns), Quality ρ = +0.062 (ns) — all three standard factors failed in 2022. VYRA's residual ρ after removing all three = +0.247 (statistically significant). The model was capturing something factors missed — specifically, proprietary operational fundamentals that predicted which companies would weather the tightening cycle.
2021/2017 also show significant residuals (p=0.050 and p=0.105 respectively). Across 3 of 3 tested periods, VYRA's signal persists after factor control.
Partial Spearman correlation between VYRA projections and actual TSR, after controlling for Value (1/PE), Momentum (6-month return), and Quality (ROE). Residual ρ remains statistically significant in 2 of 3 tested periods — VYRA captures information beyond standard factor models.
•••
ML Signal Enhancement(champion model, walk-forward OOS — data from /api/stats/summary)
What the ML layer adds — and how to read the portfolio metrics:
A proprietary ensemble meta-learner is trained on top of the fundamental signal. It learns which fundamental features are most predictive in which market regimes. The monthly walk-forward OOS strip uses 57 truly independent 1-month folds (2020–2024) — the model never sees the test period's data during training.
Spearman ρ lift: numbers below are taken from the live
FF5 alpha and witness 31/31 (full
A proprietary ensemble meta-learner is trained on top of the fundamental signal. It learns which fundamental features are most predictive in which market regimes. The monthly walk-forward OOS strip uses 57 truly independent 1-month folds (2020–2024) — the model never sees the test period's data during training.
Spearman ρ lift: numbers below are taken from the live
/api/stats/summary payload (ml_scoreboard) — not hardcoded. Fund metrics: net Sharpe, OOS ρ, and witness status update when server.py scoreboard data is refreshed.FF5 alpha and witness 31/31 (full
--charts run; check the live scoreboard card for the current pass count) appear in this section and in the dedicated FF5 subsection below.
No feature names or ML library identifiers are shown — allocator disclosure policy. All numeric claims below mirror ml_scoreboard on the server.
•••
Signal Robustness — CPCV & DSR Disclosure(R4 validation · 57 monthly folds · 45 independent paths)
Why this section matters for allocators:
Signal IC validated across 57 independent monthly folds (walk-forward, no look-ahead). FF5 alpha controlled for market, size, value, profitability, and investment factor exposures. CPCV distribution: p10 Sharpe = 1.652 across 45 independent path combinations (100% positive) — evidence of robustness across regimes and market conditions.
DSR disclosure (Bailey & de Prado, 2014): Our model has been refined over 5 retrains, evaluating 56 feature candidates. We show our work. The DSR-adjusted p-value is 1.0000 (PASS) even accounting for all 56 evaluated candidates — and 100% of 45 CPCV paths positive provides independent direct evidence against data mining. The n_trials_over_budget note (56 vs budget 20) reflects our transparency in disclosing all evaluated candidates; the actual DSR gate passes comfortably with p=1.0000.
Signal IC validated across 57 independent monthly folds (walk-forward, no look-ahead). FF5 alpha controlled for market, size, value, profitability, and investment factor exposures. CPCV distribution: p10 Sharpe = 1.652 across 45 independent path combinations (100% positive) — evidence of robustness across regimes and market conditions.
DSR disclosure (Bailey & de Prado, 2014): Our model has been refined over 5 retrains, evaluating 56 feature candidates. We show our work. The DSR-adjusted p-value is 1.0000 (PASS) even accounting for all 56 evaluated candidates — and 100% of 45 CPCV paths positive provides independent direct evidence against data mining. The n_trials_over_budget note (56 vs budget 20) reflects our transparency in disclosing all evaluated candidates; the actual DSR gate passes comfortably with p=1.0000.
CPCV — 45 independent paths
Median path Sharpe: 2.87
p10 Sharpe: 1.652
Paths positive: 100% (45/45)
Gate: PASS
Median path Sharpe: 2.87
p10 Sharpe: 1.652
Paths positive: 100% (45/45)
Gate: PASS
DSR Disclosure
Feature candidates (R1–R4): 52
OOS folds: 57
DSR gate: WARN (n_trials/n_obs)
DSR p-value: ~0 (1.86×10⁻⁴⁶)
Feature candidates (R1–R4): 52
OOS folds: 57
DSR gate: WARN (n_trials/n_obs)
DSR p-value: ~0 (1.86×10⁻⁴⁶)
Walk-forward IC: 0.1146 (t=7.24, 49/60 folds positive) · FF5 alpha: 11.41% ann. (tNW-18=4.79) · Capacity: Sharpe 1.372@$1M → 1.333@$100M → 1.278@$500M (knee >$500M)
Signal scope — where ρ is negative(S&P 500 segmentation)
Cross-sectional ρ can be negative where the operating-margin / valuation stack is a poor fit (e.g. some Financials) or coverage is thin. This is proactive disclosure — same framing as the allocator package.
•••
FAANG+ Concentration Analysis(honest disclosure — F100 universe)
Why we include this — proactive disclosure, not a weakness:
This section exists because we want to show the full picture, not just the flattering numbers. In the Fortune 100 universe (92 companies), 8 mega-cap tech stocks (NVDA, AAPL, MSFT, AMZN, META, GOOGL, TSLA, NFLX) contribute +3.5 to +5.6pp to Spearman ρ because they have extreme return magnitudes — getting NVDA's direction right in a +750% year contributes far more to ρ than getting a +25% mid-cap right.
Key finding: The FAANG+ effect is inconsistent — it helped ρ in 2022 and 2017 (direction mostly correct) but hurt ρ in 2021 (direction mostly wrong). This means FAANG+ is not artificially propping up the signal — it cuts both ways. At S&P 500 scale (503 tickers), 8 stocks = 1.6% of the sample and the effect is ~6× smaller. The k-fold ρ (+0.164) averages across both helpful and harmful FAANG+ periods.
This section exists because we want to show the full picture, not just the flattering numbers. In the Fortune 100 universe (92 companies), 8 mega-cap tech stocks (NVDA, AAPL, MSFT, AMZN, META, GOOGL, TSLA, NFLX) contribute +3.5 to +5.6pp to Spearman ρ because they have extreme return magnitudes — getting NVDA's direction right in a +750% year contributes far more to ρ than getting a +25% mid-cap right.
Key finding: The FAANG+ effect is inconsistent — it helped ρ in 2022 and 2017 (direction mostly correct) but hurt ρ in 2021 (direction mostly wrong). This means FAANG+ is not artificially propping up the signal — it cuts both ways. At S&P 500 scale (503 tickers), 8 stocks = 1.6% of the sample and the effect is ~6× smaller. The k-fold ρ (+0.164) averages across both helpful and harmful FAANG+ periods.
In the F100 universe (92 companies), 8 mega-cap tech stocks (META, AMZN, AAPL, NFLX, GOOGL, NVDA, MSFT, TSLA) represent ~9% of the sample and contribute +3.5 to +5.6pp to Spearman ρ because they have extreme return magnitudes. This table shows the signal with and without these names to quantify the concentration effect.
•••
Russell 2000 Signal Validation(7 OOS periods, 2017–2023)
Does the signal generalize beyond large caps?
VYRA's fundamental model was built on large-cap (S&P 500 / Fortune 100) data. A legitimate question is whether it works at all on small/mid-cap stocks where data quality is lower and binary events (FDA approvals, M&A, earnings blowups) dominate. The R2K validation answers this: ρ positive in all 7 tested periods, mean ρ = +0.148, directional accuracy 54.8% — weaker than S&P 500 (mean ρ +0.24, dir 57.7%) but never negative.
Why it's weaker and why that's expected: ~36% of R2K companies have no PE signal (loss-making). Binary events (FDA, acquisition, covenant breach) create unpredictable large moves that overwhelm fundamental signals. The 2020 weakest year (ρ +0.063) reflects COVID macro moves overwhelming fundamentals for all universes. The ML-enhanced Sharpe for R2K (7 periods): 0.988 vs fundamental 0.401 — a 147% lift — showing the ML layer adds substantial value even in a noisier universe.
VYRA's fundamental model was built on large-cap (S&P 500 / Fortune 100) data. A legitimate question is whether it works at all on small/mid-cap stocks where data quality is lower and binary events (FDA approvals, M&A, earnings blowups) dominate. The R2K validation answers this: ρ positive in all 7 tested periods, mean ρ = +0.148, directional accuracy 54.8% — weaker than S&P 500 (mean ρ +0.24, dir 57.7%) but never negative.
Why it's weaker and why that's expected: ~36% of R2K companies have no PE signal (loss-making). Binary events (FDA, acquisition, covenant breach) create unpredictable large moves that overwhelm fundamental signals. The 2020 weakest year (ρ +0.063) reflects COVID macro moves overwhelming fundamentals for all universes. The ML-enhanced Sharpe for R2K (7 periods): 0.988 vs fundamental 0.401 — a 147% lift — showing the ML layer adds substantial value even in a noisier universe.
Does VYRA's fundamental ranking signal generalize to small/mid-cap stocks ($300M–$2B)? Validated on 1,930 Russell 2000 companies across 7 OOS periods. Signal confirmed: ρ positive in all 7 periods — never negative.
•••
Model Accuracy — All Three Domains(2022-06-30 calibration baseline, F100)
Three separate prediction problems — all above 60% directional:
VYRA runs three domain models independently: TSR (total shareholder return including dividends), Stock Price (price-only return), and Earnings Quality (whether a company beats or misses earnings consensus). Each is validated independently against actual outcomes on the Fortune 100 (92 companies), 2022 calibration baseline, with Monte Carlo simulation (500 draws).
How to read the columns: Directional accuracy = % of companies where VYRA correctly called the direction relative to peers (the primary metric). Median MAE = magnitude error; large because VYRA is conservative (floor estimate, not point forecast). CI Coverage = % of realized outcomes falling inside VYRA's 80% confidence interval — all three domains are above 80%, meaning the intervals are well-calibrated conservative bounds.
2021 context: These numbers are from the 2022 calibration baseline (the hardest year). In normal regimes, directional accuracy runs 3–5pp higher.
VYRA runs three domain models independently: TSR (total shareholder return including dividends), Stock Price (price-only return), and Earnings Quality (whether a company beats or misses earnings consensus). Each is validated independently against actual outcomes on the Fortune 100 (92 companies), 2022 calibration baseline, with Monte Carlo simulation (500 draws).
How to read the columns: Directional accuracy = % of companies where VYRA correctly called the direction relative to peers (the primary metric). Median MAE = magnitude error; large because VYRA is conservative (floor estimate, not point forecast). CI Coverage = % of realized outcomes falling inside VYRA's 80% confidence interval — all three domains are above 80%, meaning the intervals are well-calibrated conservative bounds.
2021 context: These numbers are from the 2022 calibration baseline (the hardest year). In normal regimes, directional accuracy runs 3–5pp higher.
Test 1 (observed-driver backtest): actual historical driver deltas fed into the projection engine, projected output vs actual TSR/price/earnings beat.
•••
Primary fund — Monthly balanced(promoted 2026-04-14, 102 months)
Monthly rebalance, 102 periods (2016–2024). Net Sharpe 1.17, monthly 1-month IC 0.1146 (t=7.24, 60 independent folds, 2019–2024), 0/102 eval. windows negative (signal breadth — each window = 18Q forward L/S spread). June annual strip (10 non-overlapping periods, 2014–2023): Sharpe 1.14, MaxDD 0% (0/10 annual periods negative) — primary fund-level claim for investor communications.
•••
Fama-French 5-factor alpha(institutional gold standard)
Time-series regression on 102 monthly net returns vs Ken French published factors. α = 11.41% ann., Newey-West t=4.79 (18-lag corrected), significant at 1%. Factor loadings near zero — not a factor tilt.
•••
Fama-MacBeth cross-sectional regression(per-stock, per-month factor independence)
Characteristics-based FM (1973): in each of 57 monthly cross-sections (2020–2024), does VYRA rank predict which stocks outperform within that month, after controlling for size, momentum, reversal, and ROE? λ̄ = +3.13 bp/month, NW-4 t = 6.94, p < 0.001 — significant at 1%, positive in 82.5% of months. The strongest of the three factor-independence tests.
•••
ML lift — feature family attribution(where does the signal come from?)
Incremental ρ lift as signal sources are added to the fundamental baseline. Source: permutation importance from walk-forward OOS folds. Further detail available under NDA.
•••
Portfolio construction and friction note
Full specification of portfolio mechanics, cost model, and friction sensitivity for the net Sharpe / LP× claims.
•••
Scenario-Conditional Ranking Accuracy(sector rotation calibration)
What scenario-conditional ranking adds — and why uniform shocks don't work:
When a user specifies a macro scenario (e.g. "similar to 2022 tightening"), VYRA applies sector-rotation corrections — historically calibrated alpha shifts for each sector during that regime type. A critical mathematical point: uniform macro shocks cannot improve Spearman ρ. Spearman measures rank order, which is invariant to any monotone transformation. Adding the same shock to all 500 stocks shifts returns but doesn't reorder them. Only non-uniform adjustments (different per sector) can change the ranking.
The empirical result: The 2022 tightening scenario (n=2,142) shows fundamental ρ = +0.379 → corrected ρ = +0.436 (+5.7pp). This is the only scenario with a full OOS validation. Other scenarios use the same methodology with their historical analogs, but have not been separately validated — their sector corrections are directionally correct but not independently confirmed.
Uncalibrated scenarios: Black swan events (pandemic, GFC, major war) have no close historical analog in the 2014–2024 dataset. Rankings for these scenarios show estimated relative magnitude only — the directional ranking is based on which companies have the most defensive fundamentals, not a validated ρ.
When a user specifies a macro scenario (e.g. "similar to 2022 tightening"), VYRA applies sector-rotation corrections — historically calibrated alpha shifts for each sector during that regime type. A critical mathematical point: uniform macro shocks cannot improve Spearman ρ. Spearman measures rank order, which is invariant to any monotone transformation. Adding the same shock to all 500 stocks shifts returns but doesn't reorder them. Only non-uniform adjustments (different per sector) can change the ranking.
The empirical result: The 2022 tightening scenario (n=2,142) shows fundamental ρ = +0.379 → corrected ρ = +0.436 (+5.7pp). This is the only scenario with a full OOS validation. Other scenarios use the same methodology with their historical analogs, but have not been separately validated — their sector corrections are directionally correct but not independently confirmed.
Uncalibrated scenarios: Black swan events (pandemic, GFC, major war) have no close historical analog in the 2014–2024 dataset. Rankings for these scenarios show estimated relative magnitude only — the directional ranking is based on which companies have the most defensive fundamentals, not a validated ρ.
How well does VYRA rank stocks when a specific macro scenario is applied? A uniform macro shock cannot improve ranking (Spearman rho is invariant to monotone transforms). Sector rotation correction — applying empirically calibrated per-sector alpha shifts from historical analog periods — produces a non-monotone correction that genuinely improves cross-sectional ranking.
Validation note: The 2022 tightening scenario has a true OOS ρ measurement (n=2,142, ρ +0.436 corrected). All other scenarios use the same sector-rotation methodology but have not been separately validated with an independent OOS ρ run — their performance is inferred from methodology, not a distinct measurement.
•••
Per-Company Projection Accuracy — Full S&P 500 (453 companies, 10 years)
What VYRA is — and is not — predicting
VYRA does not try to predict an exact return at an exact future date. It is a directional ranking and relative-magnitude signal: which companies are likely to outperform their sector peers, and by roughly how much. Two things follow from this design intent:
1. Direction is what matters, and direction is where VYRA excels. A projection of +18% on a stock that returns +35% is not a miss — it is a correct call. VYRA correctly called the direction in 63–75% of all company-years across the full S&P 500 over 10 years. Crucially, every single year is above 50% (random), including the worst macro year in the dataset (2021, the sharpest tightening cycle in 40 years). No sustained regime has ever driven directional accuracy below random.
2. VYRA is intentionally conservative on magnitude — this is a feature, not a flaw. The model is built on proprietary trailing fundamentals without incorporating analyst price targets, forward guidance, or sentiment. This produces projections that are almost always lower than the actual realized return in bull markets — that is by design. The measured bias is −15pp (fundamental model) / −6pp (ML-enhanced). A signal that under-promises and whose actual portfolio performance exceeds the forecast is far safer than an overconfident model that fails when conditions change. The wide vertical spread you see in the scatter chart is entirely explained by this conservative floor design — not by incorrect directional calls.
3. The right scorecard is portfolio performance, not point-forecast error. The proof is in what happens when you trade on the signal. The ML-enhanced model's monthly rebalance synthetic fund backtest (equal-weight top quartile, 2/20 fee simulation, 102 monthly periods 2016–2024) produced: net Sharpe 1.17 · 0/102 eval. windows negative · monthly IC 0.1146 (t=7.24, 60 independent folds) · 200× LP cumulative multiple. The June annual strip (10 non-overlapping periods, 2014–2023) is the primary fund-level claim: net Sharpe 1.14, MaxDD 0% (0/10 annual periods negative), LP× ~27× — the foundation the monthly signal is built on. That is the scorecard for a ranking signal — not how close any individual projection was to the actual number.
VYRA does not try to predict an exact return at an exact future date. It is a directional ranking and relative-magnitude signal: which companies are likely to outperform their sector peers, and by roughly how much. Two things follow from this design intent:
1. Direction is what matters, and direction is where VYRA excels. A projection of +18% on a stock that returns +35% is not a miss — it is a correct call. VYRA correctly called the direction in 63–75% of all company-years across the full S&P 500 over 10 years. Crucially, every single year is above 50% (random), including the worst macro year in the dataset (2021, the sharpest tightening cycle in 40 years). No sustained regime has ever driven directional accuracy below random.
2. VYRA is intentionally conservative on magnitude — this is a feature, not a flaw. The model is built on proprietary trailing fundamentals without incorporating analyst price targets, forward guidance, or sentiment. This produces projections that are almost always lower than the actual realized return in bull markets — that is by design. The measured bias is −15pp (fundamental model) / −6pp (ML-enhanced). A signal that under-promises and whose actual portfolio performance exceeds the forecast is far safer than an overconfident model that fails when conditions change. The wide vertical spread you see in the scatter chart is entirely explained by this conservative floor design — not by incorrect directional calls.
3. The right scorecard is portfolio performance, not point-forecast error. The proof is in what happens when you trade on the signal. The ML-enhanced model's monthly rebalance synthetic fund backtest (equal-weight top quartile, 2/20 fee simulation, 102 monthly periods 2016–2024) produced: net Sharpe 1.17 · 0/102 eval. windows negative · monthly IC 0.1146 (t=7.24, 60 independent folds) · 200× LP cumulative multiple. The June annual strip (10 non-overlapping periods, 2014–2023) is the primary fund-level claim: net Sharpe 1.14, MaxDD 0% (0/10 annual periods negative), LP× ~27× — the foundation the monthly signal is built on. That is the scorecard for a ranking signal — not how close any individual projection was to the actual number.
How to read the charts below:
The directional accuracy chart shows the percentage of company-year calls that were correct — consistently 60–75%, well above the 50% random baseline across all years and all three domains.
The scatter chart shows projected vs actual returns — the wide spread is the conservative bias described above. Focus on the positive slope of the fit line (ρ = 0.41, ML-enhanced), which means higher projections reliably rank companies with higher actual outcomes.
The trend chart shows how directional accuracy evolves across years and macro regimes — note every year and every domain is above 50%, and the ML model consistently exceeds the fundamentals baseline.
TSR Directional Accuracy
67%
ML-enhanced · 57% fundamentals
SP Directional Accuracy
73%
ML-enhanced · 57% fundamentals
EQ Directional Accuracy
73%
ML-enhanced · 69% fundamentals
The right way to evaluate a ranking signal:
•••
Loading chart…🔬 New Research — May 2026: Magnitude, Short-Horizon & Tactical Fund
Three new validated findings (May 2026)
(1) VYRA rank predicts not only which companies outperform but by how much — both in earnings surprise magnitude and price return magnitude. (2) The signal carries statistically significant alpha at the monthly horizon (21-day IC = +0.096, 3-month IC = +0.130). (3) A 3-sleeve monthly tactical fund combining Market Neutral Core + PEAD + Recognition Breakout achieves Sharpe=1.98, MaxDD=−3.8%.
(1) VYRA rank predicts not only which companies outperform but by how much — both in earnings surprise magnitude and price return magnitude. (2) The signal carries statistically significant alpha at the monthly horizon (21-day IC = +0.096, 3-month IC = +0.130). (3) A 3-sleeve monthly tactical fund combining Market Neutral Core + PEAD + Recognition Breakout achieves Sharpe=1.98, MaxDD=−3.8%.
•••
🎯 Optimal Portfolio Construction — ML Signal Concentration
The concentration discovery
VYRA's ML-enhanced ranking signal becomes dramatically more powerful when concentrated into fewer names. A systematic walk-forward OOS analysis (6 periods, 2018-2024) revealed that portfolios of the top 20-30 ML-ranked S&P 500 names delivered over 50% mean annual return — while beating SPY in every single measured period. This is the same signal that drives the published fund, deployed at higher conviction.
VYRA's ML-enhanced ranking signal becomes dramatically more powerful when concentrated into fewer names. A systematic walk-forward OOS analysis (6 periods, 2018-2024) revealed that portfolios of the top 20-30 ML-ranked S&P 500 names delivered over 50% mean annual return — while beating SPY in every single measured period. This is the same signal that drives the published fund, deployed at higher conviction.
•••
🛡 Values-Based Policy Screening — Universe Coverage & Performance
What the policy screener does
VYRA can apply investment policy screens — such as values-based, religious, or ESG mandates — to any universe of stocks before selecting a portfolio. Each company is classified as Green (fully compliant), Red (excluded by at least one hard screen), or Amber (insufficient evidence to classify).
The key question is: does applying a policy filter hurt performance? The answer across all six preset policies is consistently no — constrained compliant portfolios match or exceed the S&P 500 benchmark. This is explained by selection effect: tighter screens force the portfolio to concentrate in VYRA's highest-conviction names rather than diluting across near-benchmark positions.
Screens are applied across two universes: S&P 500 (503 large-cap names) and Russell 2000 (1,938 small-cap names). Evidence is sourced from SEC 10-K filings (for revenue-based screens), public SIC/GICS codes (for sector screens), financial ratios in the database (for Shariah D/E screens), and AI-assisted controversy analysis (for ESG screens).
VYRA can apply investment policy screens — such as values-based, religious, or ESG mandates — to any universe of stocks before selecting a portfolio. Each company is classified as Green (fully compliant), Red (excluded by at least one hard screen), or Amber (insufficient evidence to classify).
The key question is: does applying a policy filter hurt performance? The answer across all six preset policies is consistently no — constrained compliant portfolios match or exceed the S&P 500 benchmark. This is explained by selection effect: tighter screens force the portfolio to concentrate in VYRA's highest-conviction names rather than diluting across near-benchmark positions.
Screens are applied across two universes: S&P 500 (503 large-cap names) and Russell 2000 (1,938 small-cap names). Evidence is sourced from SEC 10-K filings (for revenue-based screens), public SIC/GICS codes (for sector screens), financial ratios in the database (for Shariah D/E screens), and AI-assisted controversy analysis (for ESG screens).
•••
🔔 Alerts & Tracking
Track companies you care about.
VYRA monitors key fundamental metrics weekly and alerts you when something moves significantly. Add tickers to the list below or click Track on any report.
Current Macro Regime
Loading…
Tracked Companies
Loading…
Recent Driver Alerts
Alerts fire when a tracked company's key fundamental metrics move beyond proprietary thresholds. Updated weekly.
Loading…
📡 Signal Dashboard
Early-signal long candidates.
Companies where VYRA's forward-looking view is meaningfully ahead of Wall Street consensus — ranked by Signal vs Consensus Gap (VYRA Outlook score minus analyst consensus score). A large positive gap means VYRA sees something the street doesn't yet.
•••
Loading signal data…🛡 Policy Screen
Mandate screening.
Evaluate holdings against a published policy (green / red / amber). This is separate from Signals, which ranks consensus gaps and forward outlook.
| Ticker | Company | Class | Sector | Mkt Cap | Vyra Score | TSR Proj | Summary |
|---|---|---|---|---|---|---|---|
| Open this tab to load policies… | |||||||
🎯 Activist Target Screener
Activist opportunity intelligence.
Companies where VYRA projects meaningful upside on its recommended path while management is under-executing — ranked by catalyst probability (TSR opportunity × governance receptivity × management execution gap × economic activatability). Credible view: sort by Credible Opportunity Score (constraint-adjusted). G/D signals: each row includes the G/D ratio (ROIC/WACC capital efficiency), G/D-adjusted catalyst probability, and entropy-adjusted score. A macro entropy regime badge shows the current market-wide activist climate. SOTP and governance scores are model estimates, not legal advice.
•••
Loading screener…🌱 Nonprofit Screener
VYRA Philanthropic Intelligence.
Nonprofits ranked by Philanthropic Score — a composite signal across 8 components: program efficiency, revenue health, financial resilience, mission quality, program reach, governance, executive compensation, and fundraising efficiency. Sources: Form 990 (IRS / ProPublica) + NCCS parsed data. Scores update weekly. Not legal or financial advice.
•••
Loading screener…📈 Market Intelligence
Showing: VYRA's estimated TSR (stock return + dividends) over the next 4 quarters, derived from current fundamentals only. ~61% directionally accurate. Per-stock uncertainty is wide — use for ranking, not as precise price targets. Not a recent return or price.
Universe:
Domain:
Current Macro Regime
Sector Rotation
What this shows: VYRA's estimated Total Shareholder Return (stock price + dividends) over the next 4 quarters (~12 months), derived from proprietary trailing fundamentals relative to each company's sector median. These are genuine TSR estimates, not arbitrary scores. Validated at ~61% directional accuracy across 11 years: the model correctly calls the direction (up or down) roughly 6 in 10 times. Important: these are fundamental-only estimates — analyst consensus, forward guidance, and recent news are not included. Per-stock uncertainty is wide (typical 80% CI spans ±30pp). The signal is stronger for ranking companies against each other than for predicting exact return levels. Click any sector row to see all companies ranked best to worst.
•••
Loading sector rankings...Scenario Filter — applies to Sector Rotation & All-Weather below
All-Weather Picks & Macro-Vulnerable Companies
What this shows: Every S&P 500 company ranked by average estimated TSR across all 33 macro scenarios. These are genuine TSR estimates (stock price + dividends, 4 quarters) derived from current fundamentals. Averaging across scenarios reduces noise — validated at Spearman ρ +0.289 across 11 out-of-sample periods (2014–2024). All-weather picks are TSR-positive in 70%+ of scenarios. Macro-vulnerable companies are negative across most scenarios. Read the numbers as relative rankings, not precise targets — per-stock uncertainty is wide. Analyst consensus, forward guidance, and recent news are not included. · Calibrated scenarios use sector rotation for improved ranking. For scenarios with historical analogs, VYRA applies sector rotation corrections improving Spearman ρ from +0.379 → +0.436 (+5.7pp, 2022 OOS). Novel scenarios (black swan, war) show magnitude estimates only. Methodology →
•••
Loading...Data Freshness
🌐 VYRA Worldview
DWV slot:
Market-wide context updated on a fixed schedule. This is background narrative for how VYRA is reading macro and markets — not a stock pick list and not synchronized to any single company report.
•••
Open this tab to load…📐 Vyra Scores
Vyra Score (0–100) measures fundamentals quality relative to peers — not whether the stock will go up.
It combines fundamentals quality, growth trajectory, valuation, analyst consensus, and VYRA's TSR outlook (each 20%).
The strongest signal is when both Vyra Score and TSR projection are high.
They can diverge: a company with a strong TSR projection may score lower when leverage is high, analyst consensus is weak, or growth is commodity-driven rather than fundamental.
Click any ticker to see the full score breakdown by dimension.
Leaderboards use the latest batch in scorecard_cache. Use ↻ Refresh to rebuild.
Company leaderboard
Open this tab to load…
Sectors (median Vyra Score)
Loading…
Rating distribution
Most improved (30d)
Biggest drops (30d)
Search by Vyra Score
Portfolio score alerts
Notify when a holding’s Vyra Score moves beyond a threshold (requires Reports tier). Unread notifications appear below.
Recent score notifications
💼 Portfolio
What these numbers mean: Estimated TSR (stock return + dividends) across all 33 macro scenarios — genuine fundamental model estimates, not arbitrary scores. 4-quarter (~12-month) horizon. ~61% directionally accurate, validated 11 years. Per-stock uncertainty is wide; use for relative ranking and resilience comparison, not as precise targets.
Domain:
Holdings
VYRA will automatically fetch financial data for any company not yet in its database.
No holdings yet — add tickers above