Does Signal Syndicate sell picks?

No. Signal is a sports intelligence platform. We publish research, model validation, and educational explainers — not tout-style picks or locks.

What is a holdout test?

We train or tune on one time window and evaluate on a separate window the model never saw during development. Past holdout results are estimates, not guarantees.

What happens when a model fails out-of-sample?

We RETIRE or keep it in shadow mode. We document the failure publicly in our research library when founder-approved.

What is the courtroom gate?

Multiple local AI models (DeepSeek, Qwen, Llama) review quant phases independently. Promotion requires consensus and founder sign-off.

How is this different from picks sites?

Picks sites sell outcomes. Signal sells process transparency — methodology, drawdowns, and rejected lanes included.

How Signal Syndicate Researches Sports Markets

Our holdout testing, sample-size gates, multi-AI courtroom review, and why we publish failed models — not picks.

How Signal Syndicate Researches Sports Markets

Signal Syndicate is built on a simple premise: trust compounds when you show your work — including models that failed.

This page outlines how we research, validate, and publish — without picks-site framing.

---

1. How Signal researches

Our quant lane produces structured reports: backtests, out-of-sample (OOS) windows, walk-forward folds, and shadow trackers. Content and research lanes read quant output; they do not change production model parameters without founder approval.

Inputs: Historical odds caches, graded results, as-of feature builds, courtroom panel reviews.

Outputs: Phase reports, RETIRE/MONITOR/SHADOW verdicts, research library documents, and educational blog posts.

---

2. Holdout testing and sample size

We separate development from evaluation:

| Concept | Plain English |
|---------|----------------|
| Holdout window | Data the model never saw while being tuned |
| OOS forward test | Real forward window (e.g. 2026 May–Jul) with frozen rules |
| Sample size (n=) | Number of graded bets or slates — small n = low confidence |

We do not promote lanes with insufficient OOS sample or unanimous negative courtroom review.

---

3. Model rejection (why we publish failures)

When a lane backtests well but fails forward — we RETIRE it and document why. Example pattern: strong 2025 backtest, negative 2026 OOS ROI at n=34 → RETIRE, shadow closed.

Publishing failures is the moat. Picks sites hide them.

---

4. Product surfaces (where research shows up)

| Surface | Role |
|---------|------|
| Signal Board | Model lean, edge estimate, confidence — estimates only |
| Daily Intel | What changed today — intelligence framing, not locks |
| Ask Signal | Interactive Q&A over knowledge base + session context |
| Research Library | Deep quant-derived reports (demo; freemium gate) |
| Public Blog | Evergreen education + methodology cross-links |

---

5. Why trust Signal?

| Picks sites | Signal Syndicate |
|-------------|------------------|
| Sell outcomes | Sell methodology |
| Hide drawdowns | Show calibration + failures |
| Aggregate headlines | Original validation pipeline |
| Auto-promote hype | Founder + courtroom gates |

We are not a news aggregator. External sources may provide context; Signal analysis is always primary (see `docs/SEO_CITATION_FRAMEWORK.md`).

---

6. What we do not do

- Guarantee profits or "locks"
- Auto-publish without founder review
- Present past backtest as future promise
- Scrape private content or burn live API credits on public SEO pages

---

Go deeper

- Try the demo — Research → Library
- Read our Closing Line Value explainer
- Read Sports Intelligence explained

---

Research only · estimates only · not betting advice · past results ≠ future performance.

Go deeper in Signal Syndicate

Blog posts are public education. The live product demo has Research, signals, and Ask Signal.

Try the demo Open Research → Library

All figures are estimates. Past analysis is not a guarantee of future results. Not betting advice.