AI Disclaimer

Last updated 25 August 2026 · Soulvaya Labs, Inc. · bellcowgm.com

Written in plain English. This document describes, accurately, how the product actually works. The corporate details — legal entity, state of incorporation, registered address and governing law — are settled and stated in the Terms of Service. Where anything here conflicts with the law that applies to you, the law wins. Questions or corrections: support@bellcowgm.com.

Bellcow is built on models and on AI. Both are useful. Both are wrong sometimes. This page is the honest account of what they can and cannot do, so that nothing in the product surprises you later.

Bellcow advises. You decide.

Every number Bellcow shows you — a projection, a win probability, championship odds, a confidence score, an acceptance estimate on a trade — is a modeled estimate, not a fact about the future. It is informational. You retain full and final authority over every roster move you make, and you make each one yourself.

1. This is information, not instruction

Bellcow's output is provided for informational purposes only. It is not financial advice, not investment advice, not betting advice, not legal or tax advice, and not a professional recommendation of any kind. Treat it the way you would treat a well-informed friend who has read every box score: worth listening to, not worth following blindly.

2. Everything here is probabilistic

Bellcow does not predict what will happen. It estimates a distribution over what might happen, and shows you a summary of that distribution.

A weekly projection is a central estimate. Real outcomes scatter widely around it, in both directions.
Rest-of-season figures are published as median, floor and ceiling precisely because a single number would be misleading.
Playoff and championship odds come from running your remaining schedule many times over. They are simulation output with their own margin of error, and Bellcow will tell you when a change is smaller than that margin rather than dress up noise as a signal.
A "68% chance" means that in about a third of those worlds, the other thing happens. When it does, the model was not necessarily wrong.

3. No guaranteed outcomes. Ever.

Using Bellcow will not guarantee that you win a matchup, make the playoffs, win your league, or beat anyone. Nobody can guarantee those things, and we will never claim to. If you ever see marketing — ours or anyone's — promising a championship, treat it as a reason to distrust the product. Championship odds displayed in the app are a modeled probability, not a forecast of your season and not a promise.

4. The AI can be wrong, including confidently

The in-app assistant is powered by a third-party large language model (Anthropic's Claude), given a compact summary of your league. Large language models can:

State something incorrect in a fluent, confident tone.
Mix up similarly named players, or attribute a stat to the wrong person.
Reason from a stale or incomplete summary without flagging that it is doing so.
Miss a league rule that is unusual, or misread a scoring setting.
Give a different answer to the same question asked twice.

Confidence of tone is not evidence of accuracy. If the assistant tells you something surprising about a player, check it against the player page, the projection, and your platform before acting on it. Where no AI key is configured, the assistant answers from a local rule-based fallback instead, and the app tells you which mode replied.

5. The models can be wrong too

The projections, expected-points, value, breakout and simulation models are ours, built from public NFL data and calibrated on historical seasons. They still carry real limitations: a rookie or a returning player with no current-season usage has to be cold-started from priors; a role change in the middle of a week takes time to show up in the data; a coordinator's plan is not in any dataset; and a model calibrated on past seasons assumes the next one rhymes with them.

6. Predictions change, and that is correct behaviour

A recommendation is a snapshot of what the model believed at the moment it ran. New information — a Friday practice report, a Saturday inactive, a snap-count shift, a line move — changes it. If Bellcow tells you one thing on Tuesday and something different on Sunday morning, that is the system updating on evidence, not a system contradicting itself. Re-check before kickoff; that is when the last real information arrives.

7. Data can be delayed or wrong at the worst moment

Bellcow depends on upstream feeds and on scheduled jobs that run on a cadence set by what free data sources allow. Data can be late, briefly inconsistent between sources, or simply unavailable. The Sunday-morning inactive list is the single hardest thing to be current on, and it is also the moment where being current matters most. When sources disagree, Bellcow is designed to surface the conflict rather than silently pick a winner. Read the full source picture on Data Sources.

8. What a confidence score actually means

Every player and recommendation carries a HIGH / MEDIUM / LOW rating and a 0–100 score. That score is a modeled estimate of how much Bellcow trusts its own inputs — how fresh the stats, injury reports, depth charts and league sync are, how many sources agree, whether a projection exists, and whether the sources conflict. It is not a probability that the recommendation is correct, and it is not a guarantee attached to the advice. Low confidence is always shown, never hidden, because a quiet answer is more useful than a falsely certain one.

9. You retain final authority

Bellcow is read-only. It never submits a lineup, a waiver claim or a trade. You execute every move yourself on your own platform.
You know things the model does not: your league's politics, who is checked out, who never accepts a trade, what your keeper rules reward.
You are responsible for every decision you make, whether or not it matches what Bellcow suggested.
Overriding the model is a legitimate move, not a misuse of the product.

10. What is sent to the AI

Your question and a compact league summary go to Anthropic's API to be answered. Your email address, session token, payment information and Yahoo tokens are not included. Your chat history lives in your browser tab, not in our storage. Your league data is not used to train third-party AI models. The exact contents of the summary are itemised in the Privacy Policy.

11. Not for betting

Bellcow is built for season-long fantasy football team management. It is not a betting product, not a daily-fantasy optimiser, and not a source of picks. Nothing in it is an invitation to wager, and it should not be used as the basis for any bet. See the Fantasy Sports Disclaimer.

12. Liability

To the fullest extent permitted by law, Soulvaya Labs, Inc. accepts no liability for decisions you make based on AI or model output, or for losses of any kind — including lost league winnings or entry fees — arising from them. The full terms are in section 20 of the Terms.

Contact

If Bellcow told you something that turned out to be wrong, we genuinely want to hear about it — that is how the models get better.
support@bellcowgm.com · Contact Support · Help Center