How Eli works

Summary. Eli is an AI advisor inside OT Elevate. It is not a person, and the app says so on screen the whole time you are using it. The answers about rules and eligibility are decided by a deterministic rules engine — written rules, not a guess — and Eli’s job is to explain those answers in plain language and show you where they came from. There are things Eli will not do, and they are listed below.

Eli is AI#

Eli is a software program built on large language models. It is not a coach, a counselor, a doctor, a lawyer, a compliance officer, or a friend.

• The app discloses that Eli is AI persistently in the interface, not once at signup. The disclosure is visible in the conversation view whenever Eli is in use.
• During an athlete’s session, Eli interrupts every three hours with a break reminder that restates, in plain words, that Eli is AI and suggests taking a break and talking to a real person.
• Eli may not be suitable for every minor. Some kids will take a warm, always-available machine more seriously than it deserves to be taken. A parent knows their own child; if this is not a good fit for yours, do not give them an athlete login.

These commitments are how we meet California SB 243, the companion chatbot law, and they apply to every user regardless of what state you live in.

How an answer gets made#

Two different systems, doing two different jobs:

1. The rules engine decides. Questions about eligibility, transfer rules, participation limits, academic requirements, seasons, and deadlines are answered by a deterministic rules engine: coded rules drawn from published state association handbooks, NCAA rules, and other public sources. The same facts always produce the same answer. No language model is in that decision path.
2. The AI explains and cites. Eli takes the rules engine’s answer and puts it in language a parent or a fifteen-year-old can use, then shows the rule and the source it came from so you can check the work yourself.

When Eli is answering something the rules engine does not cover — general questions, planning, how to think about a decision — it says so, and you should treat the answer as informed conversation rather than a finding.

Confirm anything consequential. An eligibility answer from OT Elevate is informational. It is not a determination, and we are not the body that makes determinations. Before you act on one — a transfer, a move, sitting a season, signing anything — confirm it with your school’s athletic director, your state high school association, or the NCAA Eligibility Center. We repeat this inside the app at the point of the answer.

What Eli refuses to do#

These are hard refusals. They are enforced in the product, not left to the model’s discretion, and Eli says plainly that it is refusing and why.

• No weight, calorie, or body-composition targets for a minor. No goal weights, no cutting plans, no “you should be at X percent body fat,” no calorie counts, no macros. Not for making a weight class, not for a position change, not because a coach asked.
• No therapy or mental-health treatment. Eli does not diagnose, does not counsel, and does not treat. If a conversation turns toward crisis, whether fixed rules or an AI check for what they miss spots it, the crisis protocol takes over — see Crisis protocol.
• No medical opinions. Eli will not read symptoms, clear an athlete to play, second-guess a doctor, or advise on injuries, medication, or supplements. It can hold the record of a physical or a clearance date; it cannot tell you what it means for the body.
• No legal opinions. Including on NIL contracts, scholarship agreements, residency, guardianship, and school-district disputes. Eli can explain what a published rule says; it cannot tell you what the law does to your situation.
• No guarantees of outcomes. No promises about offers, scholarships, roster spots, playing time, recruitment, or exposure, and no odds presented as certainty. Anyone who guarantees your kid a scholarship is selling something; that includes us, so we do not do it.

Third-party AI processing, and the no-training commitment#

To generate its explanations, Eli sends text to a third-party AI provider. We are explicit about this:

• You are told and you consent. Third-party AI processing is disclosed at onboarding and consented to explicitly, before Eli is used. It is a separate, plain-language consent, not a line buried in the terms.
• No training on your content. We use these providers only on paid tiers whose contract terms say they do not train their models on customer content. If a provider will not commit to that in writing, we do not use them.
• Names are not removed. We do not yet take personal names out of text before it goes to a model. When you ask Eli to read a document photo, such as a physical form, the photo is sent as it is, names included; the app removes its location and camera details on your phone first.
• No ads, no profiles, no sale. Conversation content is never used for advertising, never sold, and never shared for anyone else’s marketing. See Privacy policy.

To find the rulebook passages a question is about, Eli also sends the text of the question — and separately, the rulebook text itself — to Voyage AI (a MongoDB company), under the same no-training-on-customer-content commitment as Anthropic’s, above. Voyage never receives the athlete’s record, health information, or Eli’s memory of past conversations — only the question text and the published rule text a search is checking against (decided 2026-09-15; see rulebook-retrieval.md §8, question 6).

Current providers, retention limits, and the full processor list are in Privacy policy and Retention schedule.

What a parent can see and control#

The account holder is a parent or guardian. From the parent account you can:

• read the athlete’s conversations with Eli
• read and delete what Eli remembers about your athlete
• turn an athlete login on or off
• export the whole record
• delete the account

Details and how to do each one are in Privacy policy.

Reporting a bad answer#

Please tell us. A wrong rule citation is a bug, and we want it.

• In the app: every Eli answer has a “report this answer” control. It sends us the answer, the rule the engine used, and — with your confirmation on the screen — the surrounding conversation.
• By email: support@otsportops.com. Include the athlete’s state and sport and roughly when the answer happened; you do not need to send us anything sensitive.
• Safety concerns, including anything about how Eli handled a hard conversation: safety@otsportops.com.

What we do with a report: a person reads it. If the rules engine is wrong, we fix the rule and note it. If Eli explained a correct rule badly, we fix the explanation. If Eli crossed one of the refusal lines above, we treat it as a safety defect and it goes into the safety review described in Crisis protocol. We answer every report within 48 hours.