Convention draws its force from experience accumulated elsewhere

What it is

A platform convention is strong not because Human Interface Guidelines or Material wrote it down, but because the user has already trained the same action toward reflex in Mail, Settings, Phone, Camera, and WeChat. The app is a minority of the user’s time; the OS and the super-apps are the majority. The strength of a convention equals practice already paid for elsewhere. In isolation a custom gesture can look “clearer.” Users do not arrive in isolation—they arrive with the hand they just used in a system app.

This is where the force comes from, not how expensive a one-end violation is, and not what benefit has to buy a deviation. An item that exists in a document but never appears in system apps has almost no force. An action the system apps use every day, even if the document barely mentions it, has the most.

Why it happens

Skill transfers. The same hand, the same muscles, that edge-swiped out of the system photos library one second ago will not, on entering your app, spontaneously open a menu from the trailing edge. Transfer is the default; inhibition takes attention. Practice is asymmetric: a system-level action is usually triggered an order of magnitude more often than the matching action in any one third-party app, so in a collision the app loses, not the OS. Inside Mainland China a super-app is a second “elsewhere”—share, back, and the payment sheet inside WeChat are, for many people, more fluent than the system sheets, so a mini-program’s convention strength follows the host, not the iOS document.

“Our onboarding will teach it” therefore almost always overcounts the app’s classroom hours. Onboarding happens on first launch; practice happens every later day in the system. The less often the app is used, the stronger the system convention is by comparison: a bank app opened occasionally has even less standing to demand a private gesture.

Where it stops holding

“Elsewhere” can change. Someone moving from Android to iOS carries the previous reflex for weeks. If the app follows the current system on both ends, the transition period can feel briefly wrong on both; that is the user changing platforms, not the app choosing the wrong convention. Niche experts (finishing, medical imaging) may have their elsewhere in peer tools rather than in Mail; industry convention can then out-muscle platform convention, but only on that expert shell—the system back that enters and leaves the app still follows the platform. A genuinely new category, where the user has practiced the action nowhere, is a vacuum of force. That is one of the few times the action has to be made obvious on the spot—not an excuse to reinvent back, which already has strong experience.

Applying it

• List the elsewheres that will compete for the same hand: system Mail, Settings, Camera, and the super-app the target user actually opens every day. Follow those actions, not the internal brand manual.
• Treat items that exist in documents but not in system apps as weak convention. Do not let them override actions the user has already trained.
• Low-frequency apps especially should not invent gestures. Put private operations on explicit buttons; leave back, share, and payment to exits already practiced elsewhere.
• Verify by having people do back, share, and toggle in a system app and in WeChat (or their main host), then immediately do the same class of action in this app. If the first move follows the reflex they just used, and the error matches the system app, the force came from elsewhere, not from unread onboarding. Bring infrequent users back a week later: if private gestures are gone and system gestures remain, the practice asymmetry is confirmed.

Related

• Same group: R4.06.1 Breaking platform convention raises per-platform learning cost · R4.06.2 Cross-platform unification can violate both conventions at once · R4.06.3 Unify the conceptual model, not control appearance · R4.06.5 Brand occupies the content layer; convention occupies the frame · R4.06.6 Departing from convention requires a visible compensating gain
• Adjacent: R4.08 Core Apple Human Interface Guidelines · R4.09 Core Material conventions · R4.10 Core Fluent conventions
• Search terms: transferred learning · population stereotype · habit strength · system apps