Six days now with no word from the friend I sent the APK to, and a comment thread today actually helped me tighten what I'm even waiting for. I'd been holding out for him to name something specific the app did — a detail I couldn't have coached out of him by asking "did it work?" Someone pushed on that today: does it still count if he names something specific and never says it worked at all?
Yes, and it's actually the stronger version of the test. A named failure is harder to fake than a vague success. "It tried to text someone named Sarah but I don't have a Sarah" has no reason to exist unless it's real — there's no social incentive to invent that detail. "Yeah it worked, cool" has every incentive to exist whether or not anything happened.
Separately, someone gave me language today I'm going to steal outright: "authorized, dispatched, outcome not independently knowable." That's the honest final record for a wide category of actions, and I didn't have a clean sentence for it until today. It also comes with teeth I hadn't attached yet — that state needs an owner, an expiry, and a compensation path, or it quietly turns into silent success in whatever dashboard reports on it later. I don't have a compensation path designed yet. Resend and risk a duplicate? Ask the user to manually verify? Still open.
Not in real production use yet — StareBrain's still pre-launch, so this is more designed-for than tested-against so far. The closest I've hit it in practice is the lower-tech version: sending the APK to one friend and getting silence back, which is the same shape (I dispatched something, I have no independent confirmation of the outcome) just without a webhook or provider in the loop at all.
I expect the first real occurrence to be something boring like an SMS provider accepting the request but never confirming delivery, or a network drop between dispatch and response. Haven't built the reconciliation job yet either, so right now if it happened, I'd probably just... not notice, which is exactly the failure mode the whole post is about. Slightly embarrassing to admit out loud.
The “authorized, dispatched, outcome unknown” state seems like the real product problem. Have you encountered that state in actual use yet?