Weekly equity picks, governed by a methodology you can audit.
Same first-principles framework. Now run live, every Monday — with the integrity check published alongside.
The Signal Live is the operational deployment of our GPT-Augmented Momentum Strategy. Every weekly pick traces to an exact model version, an immutable feature snapshot, and a calibration check against the canonical benchmark. If the calibration fails, no picks ship. That's the promise.
Get Premium picks free for 30 days.
Email-only signup. No password. No credit card. First Premium picks land Monday at 7am ET.
Both tiers receive the weekly governance discipline.
Premium adds the picks. Free is the methodology — permanently.
The Research Letter
- Weekly Research Letter (Mondays 7am ET): regime read + governance block + lagged 1-week realized return
- Monthly methodology deep-dive (one essay per month on walk-forward, leakage detection, regime change)
- Quarterly performance attribution report (PDF)
- Full access to the Research Kit (methodology + results PDFs + Python template)
- Subscriber-only updates as Signal expands to crypto, treasuries, and other asset classes
The Signal Live · Premium
- Weekly Up/Down equity picks delivered Monday 7am ET — top-k Long, bottom-k Short (optional), rest Cash
- Full integrity block per pick set: model version, training window, embargo, gates passed, calibration result
- Live equity curve, week-over-week attribution
- Subscriber portal page with full pick history
- Early access to subscriber community (launching Q3 2026)
- First-access to Crypto Momentum, Treasury Carry, and other Signal cells when they launch
30-day free trial. No credit card required at signup. Cancel anytime before day 30 and owe nothing.
How a weekly run actually works.
A single disciplined walk-forward pipeline produces both the historical backtest and this week's live picks.
Calibrate
Sunday evening: re-run the canonical 2019–2024 backtest. If results don't match exactly, the run aborts and no picks ship.
Audit
Four leakage gates run before any model fit: future-data check, correlation cap, target-column scan, time-split integrity. All four must pass.
Predict
Walk-forward retrain on the latest 104 weeks with 1-week embargo. Model scores all tickers. Top-k go to Up, bottom-k optional Short, rest Cash.
Ship
Monday 7am ET: picks + integrity block delivered to Premium. Research Letter (no picks) goes to Free. Both archived to subscriber portal.
If the math doesn't reproduce, you get a Research Letter, not picks.
Every Sunday evening, the runner reproduces the canonical 2019–2024 backtest on frozen reference data. The numbers below are what we expect to see, exactly. Any drift aborts the run.
| Strategy | CAGR | Sharpe | Max DD |
|---|---|---|---|
| SPY Benchmark | 14.9% | 0.59 | -32.2% |
| Gradient Boosting (Long/Cash) | 15.4% | 0.717 | -22.6% |
Better risk-adjusted returns and dramatically lower max drawdown than the benchmark, on five years of out-of-sample walk-forward data. These exact numbers must reproduce every Sunday or no picks ship Monday.
What makes The Signal Live different.
Not "we use AI." Not "trust the algo." A different operating model — one you can audit.
| Most "AI Picks" Services | The Signal Live |
|---|---|
| Performance numbers from in-sample backtests | All numbers from walk-forward out-of-sample with embargo |
| Methodology described in 2 paragraphs | Full methodology PDF + runnable Python template free |
| "Trust the model" | Research Letter publishes alongside every pick. Calibration check publicly reproducible. |
| Picks ship regardless of model state | If calibration fails, picks don't ship. Letter explains why. |
| No leakage testing | Four leakage gates as hard assertions before every model fit |
| Black-box model versions | Every pick traces to exact model version, training window, feature snapshot |
The discipline is the product. The picks are the byproduct.
The economics of careful research just shifted.
In 2024-25, the technical bar to "build a momentum strategy with ML" dropped to a few hours of GPT-assisted code. Anyone can produce a model. The hard part isn't building it. The hard part is knowing whether what you built is real.
We built The Signal Live around that hard part. Walk-forward validation. Embargo periods. Hard-assertion leakage gates. Weekly calibration against canonical reference data. The model is gradient boosting because that's what survived the audit — not because it's flashy.
You can build something similar yourself in the Course. Or you can subscribe to the Live and have it ship to you every Monday with the audit trail attached. Either path leads to the same place: equity research you can defend.
Mehrzad Mahdavi, PhD
I've spent 30 years in quantitative research and financial technology — most of it building the data infrastructure and analytics platforms that other people's models ran on. The vantage point that gave me wasn't "I built the best model." It was watching a lot of models, built by smart people, fail in production in instructive ways.
The pattern was almost always the same: the model looked excellent in backtest, got deployed, performed adequately for a while, then a regime changed and stopped working in a way nobody anticipated. The post-mortem usually traced back not to the model but to the validation framework one level up — subtle leakage, insufficient regime variation in the test set, embargo periods that were too short.
The Signal Live is the operating model I wish I'd had access to thirty years ago. Walk-forward, audited, calibrated weekly, and shipped with the audit trail attached. If you build ML systems for equity research — or just want institutional-grade research without the institutional fee structure — I'd be honored to have you spend the first 30 days with us, no credit card, no commitment.
Get the methodology, no commitment.
Weekly governance updates + monthly methodology deep-dives + the full Research Kit. Free forever.