Pre-Launch Validation: What We're Testing Before Veilsort Ships
Veilsort is in pre-launch validation on TestFlight. Before the public App Store launch, we are testing the things that actually matter to users — and we will not publish accuracy numbers, tester counts, or testimonials until that validation is complete. Here is what we are testing, and what we will and will not promise.
Veilsort is a privacy-first email analysis tool for iOS that runs entirely on-device using Apple Intelligence. The core pipeline is built: share an email from Mail.app (or paste it), PII is masked, Apple Foundation Models analyze it locally, and you get a summary, risk flags, deadlines, and a cautious suggested reply. None of that touches a server. That is the bet, and it is the foundation we are now validating with real people on real emails before we ship to the App Store.
What We're Validating
Pre-launch validation is not a marketing exercise. We are running it because the only honest way to claim Veilsort works is to watch real users complete the primary workflow on their own devices. Specifically, we are validating:
- The share-from-Mail flow. Open Mail, tap Share, select Veilsort, and reach a complete analysis result within a few seconds — on supported iPhone hardware running iOS 26 with Apple Intelligence enabled.
- The four result sections. Summary, risk flags (with domain, severity, and a verbatim supporting quote), deadlines (explicit and inferred, with source quotes), and a cautious suggested reply that is copyable and non-committal.
- The unavailable state. On a device without Apple Intelligence, Veilsort shows a graceful locked screen. No crash, no network fallback, no silent failure.
- Zero network behavior. A proxy capture during the full share → analysis → result flow must show zero outbound requests. This is a hard invariant of the app, and validation is the only way to confirm it holds in the field.
- Accessibility. VoiceOver traversal, Dynamic Type at AX5, Reduce Motion, and contrast ≥ 4.5:1.
What We're Not Promising Yet
Because validation is not yet complete, there are several things we will not say until we have the evidence to back them up:
No accuracy percentage. No phishing-detection rate. No tester count. No testimonials. No demographic breakdown. No claims about how Veilsort compares to specific cloud services. Until the validation is finished and the results are reproducible, those numbers do not exist — and presenting them would be marketing, not engineering.
This is also why Veilsort's in-app safety acknowledgement screen (shown once after onboarding) reminds users that AI analysis may be incomplete and that Veilsort is not legal advice. On-device Foundation Models are good, but they are not infallible, and the right thing to do before launch is to measure that honestly rather than assert it.
What's Intentionally Out of Scope
Some things testers and readers have asked about are not on the roadmap for v1.0, and we want to be clear about that here rather than imply otherwise:
- Bulk or batch analysis. One email at a time, by design. v1.0 is single-email analysis.
- Header, SPF/DKIM, URL-destination, or attachment analysis. v1.0 is text-only. Attachment analysis is explicitly out of scope.
- Personalized inbox-pattern learning. Veilsort does not learn from your inbox. Each analysis is independent.
- Cloud sync or accounts. Not planned. The zero-network architecture is the product, not a limitation we will quietly remove later.
What Happens Next
When pre-launch validation is complete, we will publish what we measured and how, with the methodology and the limits of the measurement stated plainly. If the validation surfaces things we need to fix, we fix them before launch, not after. The App Store launch happens when the validation evidence is in and the bar is met — not on a marketing calendar. Until then, Veilsort is available on TestFlight for a limited group of testers, and the public App Store listing shows "Coming Soon."