Limitations & Known Issues

Deferred linking depends on the signals the platforms expose, and those signals can be masked, stripped or absent. The sections below describe the edge cases seen in real traffic, their cause, and their effect on attribution.

---

Private Relay and VPNs

Several kinds of service change the client’s IP address before it reaches Detour:

• iCloud Private Relay
• consumer VPNs
• corporate proxies

Private Relay is the most common. It masks the real IP and presents a relay IP instead, sometimes with a coarse region attached.

Impact

• IP equality is one of the strongest signals in probabilistic matching. A missing or altered IP lowers the total score and produces false negatives, that is matches that should have happened and did not.
• When the click carried one IP and the first open carries a relay IP, the equality check fails outright. Combined with otherwise weak signals, the candidate drops below the threshold.

How Private Relay works

iCloud Private Relay is an opt-in iCloud+ service that proxies and encrypts traffic from Apple devices, so that no single party sees both who the user is and what sites they visit. Traffic takes two hops, an Apple-operated ingress relay and then a third-party egress relay, which masks the real IP address and DNS lookups while preserving the user’s approximate region.

• Who has it — iCloud+ subscribers who turn it on in iCloud settings.
• What it covers — web browsing in Safari, plus other unencrypted app traffic the system can proxy. It runs in the OS networking stack and is not visible to the user once enabled.

Improve iOS match rates

Enable the copy-link feature for iOS users. The clipboard signal is not affected by Private Relay and raises the chance of a successful match.

---

iOS pasteboard permission

Clipboard contents are a strong probabilistic signal, but iOS restricts access to them. Reading the clipboard triggers a system prompt in some contexts, and users can decline it. In other flows the prompt never appears at all.

Impact

• Without paste access, the pastedLink signal is missing and the score drops accordingly.
• The copy-to-install flow depends on the user actually copying the link. Users who skip that step gain nothing from it.

---

Redirect chains and referrer loss

Any service between the tap and Detour can rewrite the URL. Ad networks, trackers and link shorteners routinely strip query parameters or referrer information, and on Android click_id is carried in those parameters.

Impact

• A missing click_id at install time rules out deterministic matching for that click. Detour falls back to probabilistic matching, which is less reliable.
• A redirect can also change the landing domain, in which case assetlinks verification runs against a different host.

---

Play Store referrer missing

The Android Install Referrer usually delivers referrer data. It does not in these cases:

• The device has an enterprise account configured.

  Referrer data is reset automatically, even when a non-enterprise account is currently selected. This is a known Android platform limitation, tracked in the Google Issue Tracker.

• The link opened the Play Store website rather than the Play Store app.

• The install did not come from a normal store flow: pre-installs, sideloads, some Play Store variants, and internal or external test tracks.

• The Play Store and the app were not interacted with in the expected order, which can delay or lose the referrer.

Impact

Deterministic matching fails when the referrer is missing or malformed, and the install falls back to probabilistic matching.

---

In-app browsers block the App Store

Instagram, Threads, Facebook and TikTok open links in their own webview on iOS, and that webview refuses App Store URLs. A link placed in a bio or a caption is opened there by default, so a deferred link that redirects to the App Store stops at that point. The Meta webviews end on a blank page. TikTok shows an error toast.

Only iOS App Store destinations are affected. Android in-app browsers follow Play Store URLs normally, and a link whose destination is a web page loads in all of these webviews.

Detour treats the two families differently:

• Instagram, Threads and Facebook — Detour hands the link to the system browser, which opens the App Store from there. If no hand-off goes through, the visitor stays on the page with a button to continue manually.
• TikTok — no hand-off works. Measured on 2026-09-10, TikTok refused a plain navigation to the store, a redirect through the Detour host, and the itms-apps, x-safari and googlechromes schemes alike, because it checks where a navigation ends rather than which scheme starts it. Detour skips the attempts and tells the visitor to open the page in their own browser from the ••• menu.

Impact

• A visitor who stays inside the webview never reaches the store, so that install does not happen.
• The click is recorded before the redirect runs. Click counts therefore do not fall when this happens, and only the install count does.
• A visitor who leaves the webview and installs later can still be matched, because the recorded click stays available for the match window.

---

Frozen OS version on iOS 26

WebKit reports CPU iPhone OS 18_6 in the user agent regardless of the real iOS version. The device signal on iOS is exactly that comparison: the OS version from the click’s user agent against the systemVersion the SDK reports. A frozen token makes it fail.

Safari still sends the true version in its own Version/ token and Detour reads it from there. Chrome for iOS and in-app browsers (Facebook, Instagram, TikTok) send no such token.

Impact

Clicks from those browsers score 0 for the device signal, and iOS has no second device check to fall back on (see Scoring weights). The candidate has to clear the threshold on its remaining signals alone.

Workaround

Enable the copy-link feature. On these browsers the clipboard signal is usually what raises the score above the threshold.

---

False positives and false negatives

Probabilistic matching is a heuristic. It combines signals and accepts anything above a threshold, so there are two kinds of error:

• False negative — a correct match is missed, because the score fell below the threshold.
• False positive — the wrong click is matched, because its score cleared the threshold.

Impact

• A false negative loses the deferred attribution, and the user does not reach the content the link pointed to.
• A false positive corrupts your analytics, and the user reaches the wrong content.

---

Shared devices and shared IPs

Shared households, corporate devices and NATs all produce similar fingerprints and identical IPs for several distinct users.

Impact

• Two users clicking different links on the same device or network within a short window raise the risk of a false positive.
• Attribution on shared devices is hard to interpret, even when it is technically correct.