rocky

rocky is not a crisis service and cannot recognise one. If you are in danger or thinking about harming yourself, please contact a crisis line or your local emergency number.

Shown to everyone, every time. Not because of anything rocky thinks about you.

One sentence, in your words, kept exactly as you write it and handed back when the condition you pick is met. Optional. A gap is the time between one entry and the next; your usual gap is the average of your own.

An if-then plan tied to a specific cue is the most studied way of deciding in advance, which is why it has this shape.

One person, both ways. You send them ten to twenty minutes of text or voice each week, and they do the same for you. Rocky holds the promise. The message goes to them, not here.

One account, and no email, because pairing does not need one. It exists so a person you paired with can be told what you asked them to be told, and so a record can come back to you.

Rocky ships pointed at its own project. If you run your own, name it here and nothing will be sent to Rocky's.

An agreement, not a setting. The switch is here, with the words it agrees to.

A separate agreement from storing your record. Off by default, and off does not affect anything else.

A few people who already know each other. Everyone in one can see who else is in it, and nothing else about them.

You wrote down when you would want someone told. This is the part that tells them. It sends the same words to everyone, every time, and never anything you wrote.

It counts how many registered people you named, never which one and never what you said about them.

All of it, oldest first, with the gap between entries.

Your entries are kept in this browser, on this device, always. If you pair with someone, a server holds who you are paired with and which weeks were kept. If you turn keeping on, that server holds your entries too, readable by whoever operates Rocky and used for research: see "storing your record on the server" for what that does. Clearing browser data deletes what is here regardless.

an entry

One line you wrote, and when. Rocky keeps the line for you. It does not read it.

a gap

The time between one entry and the next. Treated as information, not as missing data. This is what "gone quiet" measures.

your usual gap

The average gap across your own record. Yours only, never compared to anyone.

the multiple, written 3.4x

The gap you are in now divided by your usual gap. A ratio, not a score, not a risk level, not a measure of how you are doing.

what you decided

A sentence you wrote in advance, handed back when your condition is met. Rocky holds it and does not write one.

• Know how you feel. No way to tell, and it does not guess.
• Read what you write. Only a count of how many people you named.
• Compare you to anyone. Nothing here knows what anyone else does.
• Diagnose, score, or rate you.
• Detect a crisis. It cannot, and does not try.
• Decide anything you did not decide first.
• Send your entries to the person you paired with. They see whether your weekly message went, and nothing else. What reaches them if your rule fires is a fixed message saying you asked to be checked on, the same for everyone. If you switch on keeping, your record is kept on a server and people who operate Rocky can read it. That is a change from what this said before 2026-08-15, and it is off until you turn it on.
• Talk to you. No voice, no opinions. It hands back your own words and stops.

A list of limits, not of features.

None of the ideas here are original. Three are worth knowing, for what they support and where they stop.

• Deciding in advance, in if-then form. Well studied, with a medium to large effect on whether the intended thing happens. Effects were larger with if-then phrasing and with rehearsal, which is why the sentence has this shape and is shown back on ordinary days.
• Naming people in advance. One component of safety planning, which has solid clinical evidence. Rocky is not a safety plan and that evidence does not transfer. A safety plan is made with a clinician and covers warning signs and access to means. If that is what you need, ask a clinician for one rather than using this.
• Writing things down. Studied for decades, small to moderate benefits, using long reflective sessions. A one line entry is not that, so treat this as a record and not as a treatment.

Rocky has not been studied. The above is evidence for the ideas it borrows, which is a weaker claim.

Five methods and a fixed-rule baseline, on simulated records, scored causally, each compared against firing at random at the same volume. Precision is how often an alert landed inside a genuinely hard stretch.