Oncourse paywall retest

September 9–10, 2026 · Sequential iPhone → Android → iPad checks using Argent. Partial QA passed; not full release sign-off.

Scope and environment

iPhone 17 Pro and iPad Pro 11-inch (M5), iOS 26.5: native development build with branch head a01e5f555, Xcode StoreKit configuration and Metro 8083.
Android API 36.1: clean installation of preview 534, commit ac3d121c7. This includes earlier restore/layout code, but not the latest APPONLY75 placement mapping.
iOS paywalls were presented directly through RevenueCat SDK on an anonymous QA identity. These checks validate provider UI and local StoreKit behavior, not authenticated onboarding or backend account grants.
Android used the actual anonymous USMLE Step 1 onboarding flow. No purchase was completed on Android.
Published main revision 184, referral revision 11, shared exit revision 451. No provider edits or republishes during this retest.
Only one Oncourse test device ran at a time. Unrelated Vidia devices were left running.

Results

Check	Fresh result
Plan billing-period clipping	PASS on iPhone main/referral plan lists and iPad referral list. Annual, monthly and weekly billing-period labels fit. Android annual main plan also fits; full list not reached safely because accessibility discovery became stale.
Monthly trial copy	PASS on iPhone main and iPad referral: selecting monthly changes the plan-list button to SUBSCRIBE. iPhone StoreKit checkout showed $9.99/month without a trial.
Monthly purchase and entitlement	PASS, bounded: iPhone Xcode test purchase completed; SDK result PURCHASED and RevenueCat active entitlement oncourse_max. This was explicitly a no-charge local test, not App Store production billing or an authenticated backend entitlement check.
Existing-purchase restore	PASS at SDK level: iPhone restorePurchases returned oncourse_max in 784 ms. The UI footer and app confirmation-route behavior remain unverified in this run.
Referral product binding	PASS for directly presented offering on iPhone/iPad: annual max_12_months_30_off at $54.99; monthly $9.99; weekly $1.99. No blank plan. iPad native checkout displayed the annual referral product, a 1-week free trial, then $54.99/year; checkout was canceled.
Referral shared exit link	PASS on iPhone and iPad: closing referral opens “What's holding you back?” with the shared exit choices.
14-day exit button	PASS for tested states: iPhone trial-ineligible state showed SUBSCRIBE; fresh iPad eligible state showed START FREE TRIAL and 2 weeks free, then $89.99/year. Previously missing CTA was not reproduced on this iPad.
Three-year trial copy	PASS visually on iPad: BUY 3-YEAR ACCESS, $159.99 one time, access ends after 3 years. No free-trial promise. Purchase not completed.

Remaining issues and verification gaps

Referral discount badge needs review. The annual referral screen still shows “19% OFF,” identical to the main badge despite a different annual price. The reference price is not explained. This report does not dispute or change the approved APPONLY75 message, which the owner defines relative to the most expensive weekly plan. Keep existing products/messages until the badge's intended comparison is established.
Empty restore is not signed off. An iPhone footer tap on the directly SDK-presented paywall produced no visible completion feedback within the observed 20-second window. Because this bypasses the app onboarding callbacks, it does not prove the original three-minute payment-confirmation defect persists or is fixed. Retest the real app route with an empty purchase account.
Android final-head referral testing is pending. Preview 534 predates the APPONLY75 mapping fix. A new Android binary from a01e5f555 is required before validating that entry path.
Android automation blocked. Argent repeatedly returned the previous “Review your plan” accessibility tree while the screenshot showed “Choose a Plan,” including after transport restart. Testing stopped rather than guessing coordinates. This is a QA-tooling limitation, not a confirmed customer-facing bug.
Development reload crash observed. iPhone dev-client reload after Metro reconnection hit EXC_BAD_ACCESS in JSI Pointer destruction. A clean Xcode Run Without Building with Metro ready recovered. Scope is a development-build reload; production impact is not established.
Coverage still incomplete. APPONLY75 through the authenticated app path, empty restore confirmation routing, Pro checkout, completed three-year checkout, genuine store-account purchase/restore and backend entitlement propagation still need dedicated verification. iPad main offering catalog loaded, but its visual flow was not separately rerun.
Minor tablet design observation. The referral detail screen has a large empty vertical area between the testimonial and purchase panel. All tested controls fit, but a more tablet-specific composition could use that space better. Not a checkout blocker.

Recording

iPhone main plan switching. This recording is layout/CTA evidence, not the completed-purchase evidence.

Download recording

Screenshot evidence

iPhone main annual plan list — iPhone main: annual and billing periods.

iPhone referral plans — iPhone referral: populated plan list.

Referral annual 19 percent badge — Remaining review: referral still shows 19% OFF.

Android annual main plan — Android preview 534: annual label fits.

iPad annual referral plans — iPad referral: all periods fit; eligible annual trial.

Xcode test checkout for referral annual — iPad referral native checkout, test environment; canceled.

iPad eligible exit trial button — iPad 14-day exit: START FREE TRIAL is visible.

iPad three year access offer — iPad three-year: one-time purchase, no trial copy.

Cleanup and handoff

Removed regenerable npm, pnpm, CocoaPods, Xcode index/module and browser caches; preserved source, credentials, installed app artifacts, simulator data and recordings. Free space reached 7.9 GiB after the first cleanup, then shrank during testing. Stopped Oncourse test devices and Xcode indexing and cleared regenerated indexes again. No code changes in this retest. PR #1796 remains the implementation PR; this report is not authorization to merge.