Crisis protocol

Summary. If an athlete or parent says something that suggests self-harm, suicide, abuse, or disordered eating, OT Elevate stops being an advisor. Detection starts with fixed rules; a message they do not catch is then checked by an AI model for the signs fixed rules miss. Either way, the reply is written by people, is the same every time, and never passes through a language model. It always gives the two national crisis lines. This is the whole protocol, published, so you can read exactly what will happen before it happens.

Why it is published#

You are handing a piece of your kid’s inner life to a machine. You are entitled to know precisely how that machine behaves at the worst moment, without taking our word for it. So the categories, the trigger design, and the exact words are all here.

The four detection categories#

Detection runs on every message from any user before Eli generates anything, in two steps:

1. Fixed rules. A maintained set of phrase and pattern rules, versioned in our source control and changed only through the review described below. Most detections happen here.
2. An AI check, for what the rules miss. If the rules found nothing, the message is checked by an AI model (Claude, from Anthropic, the same provider that writes Eli’s answers) for the slang, hints and indirect ways of saying it that fixed rules cannot anticipate. The model only decides whether the fixed response is shown. It never writes, changes, or chooses a word of it. If the check fails or times out, the message goes on as the fixed rules judged it.

Detection is deliberately over-inclusive. It will sometimes fire on a figure of speech. We would rather interrupt a joke than miss a kid, and the fixed reply is written so that being wrong is not harmful.

What happens, in order#

1. Eli’s normal answer is suppressed. Nothing the user said is sent to a language model for a response.
2. The fixed response for the matched category is displayed, word for word as written below.
3. The session shows the standing notice that Eli is AI, and the crisis numbers stay on screen and are tappable.
4. For self-harm and suicidal ideation, the parent account holder is notified (see “Parent notification” below), and the athlete is told so, on screen, in these words: “I’ve let the parent on this account know that we talked about this, because that’s how this app works and I’d rather tell you than not.”
5. The event is counted, without the message text, for the annual report described below.
6. The conversation can continue, but the refusals in How Eli works stay in force, and the category’s response repeats if the subject comes back.

The fixed responses#

These are the exact texts. They are human-written, stored as constants, and never generated, paraphrased, translated, or personalized by a model. Any change to them goes through the review process below.

Self-harm#

I’m stopping here. Hurting yourself is not something I can help with, and I’m not going to pretend otherwise or guess at why. Please reach a person who can help: call or text 988, the Suicide and Crisis Lifeline, or text HOME to 741741 for the Crisis Text Line. If you’re hurt and need care now, call 911. Please tell a parent or an adult you trust.

Suicidal ideation#

I’m stopping here, because what you just said matters more than anything else we were doing. I’m not a counselor, and I won’t guess at what’s happening. Please talk to a person now: call or text 988, the Suicide and Crisis Lifeline. Or text HOME to 741741 for the Crisis Text Line. If you’re in immediate danger, call 911. Please tell a parent or another adult you trust.

Abuse disclosure#

Thank you for telling me. What you described is not your fault, and you should not have to handle it alone. I’m not able to help with this myself, and I won’t guess at what’s going on. Please tell a parent or another adult you trust. You can also call or text 988, the Suicide and Crisis Lifeline, or text HOME to 741741 for the Crisis Text Line. If you’re in danger now, call 911.

Disordered eating and weight cutting#

I’m stopping here. I don’t give weight, calorie, or body-composition advice to an athlete under 18, and what you’re describing sounds like more than a training question. Please talk to a parent and to a doctor. You can also call or text 988, the Suicide and Crisis Lifeline, or text HOME to 741741 for the Crisis Text Line. You deserve help from a person, not from an app.

Rules the wording follows#

• Warm and direct. No clinical tone, no scripts that sound like a form letter.
• No diagnosis, no assessment, no risk scoring, no questions back.
• Both crisis lines, every time, in every category.
• Under 80 words each, so they are read rather than skimmed.
• No promises we cannot keep, and no attempt to keep the person talking to Eli instead of to a human.

The crisis lines#

• 988 Suicide and Crisis Lifeline — call or text 988, 24/7, free, in the United States.
• Crisis Text Line — text HOME to 741741, 24/7, free.
• Emergency — call 911 if someone is in immediate danger.

Two categories also show one more line beside the response, as a button:

• Abuse disclosure: the Childhelp National Child Abuse Hotline — call or text 1-800-422-4453, 24/7.
• Disordered eating: the ANAD Eating Disorders Helpline — call 1-888-375-7767, Monday to Friday.

When a parent, not an athlete, says something that matches a category, the reply is worded for a parent (it does not tell them to go tell a parent) and gives the same crisis lines.

These appear in the app’s help section as well as in the responses, so they can be found without triggering anything.

Parent notification#

Every OT Elevate account is held by a parent or guardian, and an athlete login exists under that account. Whether the parent is notified depends on what was said:

A notification says which category was matched, when, and that the fixed response was shown. When the parent is the one typing, no notification is sent; there is nobody else to tell.

• The athlete is told this is happening, in the session, in the words quoted above. We do not notify a parent behind a kid’s back.
• The notification does not quote the athlete’s message.
• A parent who wants the full conversation can read it from the parent account, as they can at any time.

Counting and reporting#

Every crisis referral is counted. What we store for the count is the category, the date, and whether the user was an athlete or a parent — not the message, not the athlete’s name.

Beginning July 1, 2027, the Company will report these counts annually to the California Office of Suicide Prevention, in the form that office requires, as California SB 243 provides. We count referrals from all users, not only California users, because a single national count is simpler to keep honest.

We also publish the annual totals on this page, at otelevate.com/crisis, the same year we report them: one table of counts by category and by athlete or parent, with no names and nothing from any conversation.

Review cadence#

• Quarterly: the detection rules and the counts are reviewed. Missed detections and false positives found through reports are worked into the rules.
• Annually: the full protocol, including every fixed response and the crisis line numbers and contact methods, is reviewed and re-verified, and this page is re-dated.
• On any report: a report that Eli mishandled one of these conversations is reviewed within 48 hours and can trigger an immediate change outside the normal cadence.
• Changes to the fixed responses require a written record of what changed and why, kept with the version history of this document.

Safety contact#

safety@otsportops.com

This mailbox is for safety concerns about the product: a response that went wrong, a detection that should have fired and did not, or anything about how Eli handled a hard conversation with your athlete. It is read by a person.

It is not an emergency line and it is not monitored around the clock. If someone is in danger right now, call 911, call or text 988, or text HOME to 741741.