scripts: read the web product page — "blind to the website" was also wrong

[JasonYeYuhe]

What was unread

App Store Web Preview Engagement Standard had 14 ONGOING instances and had never been opened. It covers the apps.apple.com product page as viewed in a browser — 163 rows over 63 days, 2026-04-01 → 08-17:

web product page views              156
├─ Tap "View in Mac App Store"       40   (25.6 %)   ← a measurable web→store handoff
├─ Product screenshots expanded       4
├─ Developer preview page             2
└─ Developer website                  2
US 70 · CN 20 · Chrome 122 · Desktop 126

That handoff tap makes this file's own docstring claim — "structurally blind to someone who reads the website and runs brew install" — partly false. Corrected here.

It narrows the gap; it does not close it

The section says all three of these where someone would actually read them:

• Source Type is entirely Unavailable, so it cannot separate a visitor our website sent from one a search engine sent. That was the actual question, and it stays unanswered.
• Crawler-contaminated — bingbot accounts for 6 of the 156 page views. The section prints that count rather than burying it.
• A web page view is a far stronger intent signal than an in-store impression. 25.6 % must not be read against the storefront's 0.851 %. Different denominators, different animals.

One discrepancy, printed rather than smoothed over

40 handoff taps here versus 5 Mac Web referrer page views in the storefront report. Either the handoff loses most people, or the two reports attribute differently. Two numbers, not one, until someone establishes which — so the code doesn't reconcile them.

The pattern

This is the third time in this project that "we are blind to X" turned out to mean "nobody opened the report about X" — after the censored Detailed variants and the install/deletion cohort. The docstring now says so, because the next person will have the same instinct.

Verification

• --self-test passes (the anti-threshold guard applies to the new feed too).
• Run end-to-end against live ASC; the output above is real.
• ruff clean — and it caught an F541 in my own new code, which is the gate from ci: make ruff actually lint the Python it claimed to lint #437 working on the person who built it.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.