S&P 500 · deterministic valuation · adversarial review
Divergia ranks every company against the whole index on quality, value, growth and risk, values it with the model that fits its kind of business, and lets an AI committee critique the numbers without being able to change them. When the data runs out, the engine abstains instead of guessing.
No card needed. Free to analyze.
GDDY
Technology
Free-cash-flow DCF · discount rate 8.8% · 10 years projected
A model's reading, not a buy recommendation.
Terminal
The first five companies in today's screener, with the numbers exactly as the engine computes them. This page recalculates nothing: it only shows them.
| Verdict | ||||||
|---|---|---|---|---|---|---|
| GDDYTechnology | +0.38 | +1.04 | +0.11 | +0.05 | ≥ +150% | MODERATE SIGNAL |
| VRSKIndustrials | +1.00 | −0.79 | −0.84 | +0.27 | +24% | MODERATE SIGNAL |
| BKNGConsumer cyclical | +1.08 | −0.30 | +1.14 | −0.53 | +39% | MODERATE SIGNAL |
| EXPEConsumer cyclical | +0.50 | +0.03 | −0.85 | −0.83 | +39% | MODERATE SIGNAL |
| ROLIndustrials | +0.41 | −0.95 | −0.36 | −0.31 | +11% | MODERATE SIGNAL |
Z-scores against S&P 500 companies: 0 is the average and +1 is one standard deviation above it. For risk, a higher number means more risk, so it counts against. Margin = (intrinsic value − price) / price.
Methodology
What the engine does, and what it deliberately does not do.
Factor engine
We reject fixed cut-offs like a maximum P/E. Every company is measured against the rest of the S&P 500 with z-scores, and the score combines four factors with fixed weights:
Sector-aware valuation
The engine detects what kind of company it is and applies the model that fits. The discount rate starts from each company's beta and the 10-year Treasury yield.
Operating companies
Discounted free cash flow, with a WACC specific to each company
Banks and insurers
Residual income on book value, blended with earnings
REITs
Discounted cash flow on AFFO
Backtest
On each date it uses the companies in the index that day, including the ones that later left, and only data that was already available 90 days earlier.
Rebalance on date Tdata available at T − 90 days
E.g. the Jan 1, 2020 rebalance uses data available on Oct 2, 2019
Abstention
When the data is missing or contradicts itself, the engine publishes no value and says why. An abstention is a result, not an error.
AI committee
A single language model tends to agree with you. Divergia splits the review across agents with different mandates, and all of them work on numbers the engine already computed.
The engine computes
Intrinsic value, factors, scenarios and risks, with deterministic math.
The committee reviews
Eight agents examine those numbers from different angles, each with its own mandate.
The synthesis concludes
It says whether it agrees or disagrees with the engine's verdict, and why.
The AI can disagree with the verdict, but it cannot change the figures.
Read the numbers
Quant
Extracts the price and valuation signals, always with concrete figures.
Delivers
Quality
Looks for competitive advantages and business quality with metrics such as ROIC and cash-flow stability.
Delivers
Risk
Names the main risk and lays out quantified downside scenarios.
Delivers
Read the context
Macro
Judges whether the market regime and interest rates help or hurt the company.
Delivers
News
Reviews earnings, guidance, ratings, regulatory issues and management changes.
Delivers
Research
Extracts what the 10-K (risk factors and MD&A) and the news actually say, without interpreting.
Delivers
Attacks the thesis
Devil's advocate
Looks for the concrete scenarios that would break the bull thesis and estimates how likely it is to fail.
Delivers
Concludes
Synthesis
Combines everyone's evidence and says whether it agrees or disagrees with the engine's verdict.
Delivers
Free plan: 2 full committee verdicts a day.
Why Divergia
Divergia is not a generic bot fed with superficial prompts. The math is computed first; AI critique comes after.
Our back-end logic does not make generic API calls to show empty dashboards. We process raw data right from financial statements, run the modeling, and generate the signal before you interact with the screen.
Our system simulates a 90-day lag on SEC data availability, so reported returns do not use information the investor did not have at the time. And we show all four configurations, including the one that trails SPY.
If you are looking for magic alerts or intraday day-trading strategies, Divergia is not for you. This tool is built entirely to consolidate algorithmic value investing and structural review of solid businesses over the medium to long term.
Backtest
January 2019 to January 2025, using the S&P 500 membership on each date (398 distinct companies over the period) and returns rebuilt day by day. We show all four configurations, including the one that trails SPY.
Four engine configurations against SPY. Hover a bar for the detail.
Score ranking, equal weight
Score ranking, inverse-volatility weight
The above, with less exposure under market stress
Decision engine (DE_ALPHA)
| Configuration | Annual | vs SPY | Sharpe | Max drawdown |
|---|---|---|---|---|
| Score ranking, equal weight | 17.3% | +0.20pp | 0.62 | 43.1% |
| Score ranking, inverse-volatility weight | 17.5% | +0.37pp | 0.63 | 42.6% |
| The above, with less exposure under market stress | 15.6% | −1.48pp | 0.61 | 38.4% |
| Decision engine (DE_ALPHA) | 23.8% | +6.70pp | 0.88 | 38.4% |
SPY over the same period: 17.1% a year. Sharpe and max drawdown from daily data.
Why +6.70pp is not a promise
All four figures changed in the latest official run because of data repairs (prices, a historical risk-free rate and SEC cash flows), not engine improvements. For DE_ALPHA, the gap to SPY went from +3.91pp to +6.70pp. In the statistical tests of those repairs, every 95% confidence interval crosses zero. Past performance does not guarantee future results.
Official run of August 10, 2026.
Start free, no card. Pro costs 29 USD a month and opens the full screener.