Ship No Call to an Endpoint That Doesn't Exist

"The typed Recall marker intent and watch-span intent need new Recall endpoints first — requested through coop; the client ships no calls to endpoints that do not exist."

The Feature You Can Imagine but Can't Yet Ship

Ashen Reel can clearly imagine letting you mark a moment or a watch-span and send it home to Recall as a typed intent. The design is obvious, the button practically draws itself. There's just one problem: Recall doesn't have an endpoint to receive that intent yet. So the feature doesn't ship — and how you handle that gap is the whole quest's final lesson.

The tempting move is to build it locally now and sync later: a small markers table in the player, filled today, reconciled with Recall whenever the endpoint arrives. It feels productive. It's a trap you've now seen five tracks' worth of. That local table is a second owner of a truth that belongs to Recall, and 'sync later' is where local stopgaps go to become permanent.

Don't ship a call to an endpoint that doesn't exist — wait for the contract, or request it. A client that calls a not-yet-real API is either guessing at a shape that will change or faking the capability with a local store that will drift. The honest options are to wait for the real endpoint, or to ask for it — never to paper over the gap.

So the feature is guarded by the contract itself: no endpoint, no button, no local shadow:

Requesting the endpoint IS the action — it's not deferral. This isn't a lazy 'phase 2 later' note that quietly dies. The typed intent was requested through coop, so there's a real trigger: when Recall ships the endpoint, the feature ships against it. The difference between honest waiting and underdelivery is whether a concrete mechanism will actually deliver the deferred thing. Here one will.

Restraint Is the Quest's Closing Note

Every track has been a variation on one theme: ship less than you could, on purpose, so that what you ship is true. Refuse the features Movist has that Dad doesn't use. Refuse a decoder you could reimplement. Refuse a path in the URL. Refuse a shadow copy of the transcript. And now, refuse to call an endpoint that isn't there. Restraint isn't the absence of ambition — it's ambition pointed at correctness instead of surface area. A video player that waited for a real contract instead of faking one is, in the smallest way, the same discipline that let it retire Movist: build the honest thing, and don't pretend the rest is done.

The markers feature bugged me because I could see it, whole, ready. Building it locally 'for now' felt like progress, and Dad's "엔드포인트 없으면 안 만들어" — if the endpoint doesn't exist, don't build it — felt almost stubborn. Then I remembered every drift bug in this quest starts exactly there: a local copy that was only supposed to be temporary. The restraint that annoyed me is the same restraint that keeps the whole family of apps one honest system instead of a dozen slowly-diverging ones. Waiting well is a skill, and it's most of what this quest was actually about.

Code

The contract gates the feature: no endpoint, no button, no local shadow·swift

/// A typed marker intent needs a REAL Recall endpoint. Until it exists,
/// the FEATURE is absent -- never faked with a local store.
func markMomentIfSupported() async {
    guard recall.capabilities.contains(.typedMarkerIntent) else {
        // No endpoint yet. We do NOT invent a local markers table to hold it.
        // The button simply isn't offered. The intent was requested via coop;
        // when Recall ships the endpoint, this branch lights up for real.
        return
    }
    try? await recall.submitMarker(current.releaseID, atMs: engine.observedTimeMs)
}

// The client's surface is exactly as large as the contracts that actually
// exist -- no larger. A button with nowhere to send its data is not a feature;
// it's a promise the app can't keep.

Exercise

Find a place where a client of yours calls, or is tempted to call, an API that isn't finalized — a planned endpoint, a guessed response shape, a 'we'll add it server-side later.' Ask: are you faking the capability locally, or guessing at a contract? Choose the honest path — gate the feature behind a real capability check and file a concrete request for the endpoint — and delete any local shadow you built to bridge the gap. What stops drifting once the stopgap is gone?

Hint

Two anti-patterns to catch: calling an endpoint that doesn't exist yet (you're guessing a shape that will change), and faking its result with a local store (you've created a second owner and a 'sync later' debt). The honest move is capability-gating plus a real request for the contract — a feature the size of the contracts that actually exist, and a concrete trigger to grow it when they do.

Ship No Call to an Endpoint That Doesn't Exist

The Feature You Can Imagine but Can't Yet Ship

Restraint Is the Quest's Closing Note

Code

External links

Exercise

Progress

Comments 0