Short answer: open any piece of advice and ask five questions. 1. Does it cite a specific publication, or "studies"? 2. Is an evidence grade shown per item? 3. Does the app say anywhere that something doesn't work? 4. What happens to what you enter? 5. When a practice is contraindicated for you, does the app remove it or warn you?

1. Traceability: "studies" is not a source

The first test is the fastest. Open one piece of advice and look at what backs it. Three levels separate immediately:

Only the third level can be checked. And it genuinely can: paste the DOI into a search engine and confirm the title matches. A reference whose last character is wrong resolves to nothing — a silent, common error, the kind only an automated check catches.

2. Evidence grades, shown per item

Not every health recommendation is equally solid, and an honest app displays that rather than presenting everything on the same level. The scale used here:

Grade What it means Example
AMultiple human randomised trials, or a meta-analysis, pointing the same wayBehavioural therapy for insomnia
BOne well-conducted trial, or strong human data plus an established mechanismMelatonin for shifting the clock
CEmerging: small samples, short duration, or mechanistic dataSeveral fashionable recovery practices

An app that never separates A from C is asking you to trust its curation without giving you the means to judge it. Where a meta-analysis exists, it should outrank any single group's work — that is the rule that protects you from fashions.

3. The refusal test

This one is the most revealing, and nobody runs it. Look, inside the app, for a place where it says something does not work.

Content built from the literature necessarily contains some: meta-analyses regularly contradict popular practices. Content built from what shares well contains none, because closing a door doesn't sell.

Two easy checks. Melatonin: an analysis published in JAMA in 2023 measured 25 gummy products, of which 22 were inaccurately labelled, with actual content running from 74% to 347% of the claim. An app recommending melatonin without mentioning that has read a trial abstract, not the literature. And the sleep score on wearables: a clinical case series published in the Journal of Clinical Sleep Medicine in 2017 described patients whose insomnia was sustained by the pursuit of a perfect score — the monitoring producing exactly the arousal it claimed to measure. The phenomenon is called orthosomnia. A sleep app that says nothing about it is selling the problem along with the solution.

4. What happens to what you enter

Health data is among the most sensitive there is. Three concrete questions, in order of what they guarantee:

The strongest degree is rarely offered: an app that contains no networking code. That is no longer a promise, it is a structural property — with no networking layer, no path exists for data to leave, independently of what the legal text says.

5. Remove, rather than warn

The most discriminating design test. When a practice isn't for you — a contraindication, a particular situation — does the app show it with a warning attached, or not show it at all?

A warning transfers the decision to you at a moment when you have neither the context nor the appetite to make it. Removal handles it upstream. Sleep window compression, for instance, is effective and well documented — and it is not for people with bipolar disorder or epilepsy. An app that knows your situation and offers it anyway inside a red box has done half the job.

The same reasoning applies to what is offered. A protocol written for night work, or for the first months with a newborn, should not be a general version you are expected to adapt yourself: either it is written for that situation or it isn't.

The regulatory line, in one sentence

A wellbeing app is not a medical device: it does not diagnose, treat or prevent any condition. The distinction is useful rather than a formality, because it defines what you can legitimately expect. An app that in practice promises to treat something while presenting itself as a wellbeing tool is contradicting itself — and that is a warning sign in its own right.

The grid applied to Sillon

It would be dishonest to publish this list without submitting to it. Here are the answers, all checkable.

Question Sillon
Sources153 primary references, each checked against the Crossref registry — title, journal, year. A practice without a valid citation does not ship.
Evidence gradeEach of the 109 practices carries an A, B or C grade. Grade C is flagged in the interface.
RefusalsYes — high-dose melatonin, wearable sleep scores, supplements whose labels aren't verified.
DataNo account, no advertising, and no networking code in the app. Apple Health is optional and read-only.
ContraindicationsThey remove the practice from the engine rather than showing it with a warning.
StatusA wellbeing tool. Diagnoses nothing, replaces no medical advice. Some cards exist only to send you to a professional.

The honest limit. An app, however well sourced, does not know your file, examines nobody and does not replace an appointment. For insomnia lasting more than three months, persistent pain or a mood disorder, the best it can do is support what a professional oversees. An app that lets you believe otherwise fails the most important test.

An app willing to be judged on this grid

109 graded practices, 153 verified sources, no networking code. Free, offline, no account.