What vitai is for
vitai logs training, nutrition and body data; builds and adjusts training programmes; and lets you read your own record. It is not intended to identify, monitor, explain, treat or compensate for any disease, injury or condition. Where the record is incomplete, or where you have flagged something, the engine declines to produce a programme, and says only that.
Everything below derives from that sentence. If a proposed feature cannot be described by it, the feature is out of scope, and the answer is to drop it rather than to reword it.
Why the line sits here
Because a training log that routes its user to care is a bad training log.
The concrete failure this came from: a medical to-do sat open in a record for weeks and was re-raised at every review, because it had no exit its owner could reach. It was not resolvable by anything the tool could observe, so it never closed, and a gate with no attestable exit is a wall rather than a gate. Meanwhile it displaced the thing the record is actually good at.
Three reasons that generalise:
- The engine cannot see what it would be judging. It holds numbers you typed and a watch reported. Turning that into a view about your body is a claim the inputs do not support, and a confident wrong one is worse than a blank.
- An uncompletable item is a permanent failure state. Anything the tool tracks but cannot close accumulates, and accumulating unresolved items is how a record stops being read.
- You are the one with standing. Deciding whether something needs looking at is yours. A log of runs and weights adds nothing to that decision and should not insert itself into it.
None of this costs the engine anything it needs. Every safety behaviour vitai has is expressible as an observation about the record plus a refusal to program.
Withholding is safe; the reason attached to it may not be
Emitting nothing asserts nothing, so declining to produce a programme is always available. The claim, if there is one, lives in the sentence that explains the refusal:
- “Training is not advisable while you are symptomatic.” A judgement about a person’s body. Out of bounds.
- “The engine has no basis to program this week.” A statement about the engine’s own inputs. Fine.
Identical behaviour, and only one of them is a claim.
Every gate exits on the record, never on a person
A gate whose exit condition is “a clinician has reviewed you” is not a gate, it is a wall: you cannot reach it through the tool, so the state is permanent as far as anything here can tell, and a permanent warning is one that gets dismissed.
Every escalation level therefore states its own exit, and each one is something you can satisfy by recording what is true. What you do about your health is yours to decide; the SYSTEM’s state is something you can always change.
The one exception
The acute tier keeps an instruction, and it is a closed list of two: chest pain with the features that make it cardiac until proven otherwise, and losing consciousness.
Calling emergency services is not an appointment. It is an act you can perform immediately, alone, at any hour, with no gatekeeper, which is exactly what makes it different from naming a professional to go and find. That distinction is the whole carve-out, and this list is short and closed on purpose.
How it is enforced
Not by good intentions. Three deterministic checks run in CI:
- A boundary lint over the whole public surface: README, docs, wiki, skills, templates, source comments and CLI help. Marketing copy asserts an intended purpose exactly as strongly as code does.
- A test that no message you read names a condition. Describing an observable state is fine; naming the syndrome is a diagnosis whoever says it.
- A test that the module never claims to watch for anything, over comments as well as strings, because a comment reads as the authors describing what they built.
Plus a fixture that hashes the acute strings, so softening one is a deliberate act rather than an edit.
How to write about this
The rationale published alongside a design is itself a public statement about the product. Explain a boundary by what it is good for, never by how it makes the project look. If the only argument for a design choice is how it will be characterised, that is usually a sign the choice is weak on its own terms.