[HelloOjasMutreja]

Fixes #5392 — step 1, the small verified mobile half, per your own framing in the issue. Left step 2 (bounding the desktop-pinned layers, restoring native rubber-band on macOS) for a follow-up, since it needs its own visual pass on the tuned pin-and-cover effect.

Fix

overscroll-behavior-y: none on html was global and unscoped by width. Below 1024px, backdropGlow (the aurora blur behind the hero) is the only layer still position: fixed at all widths — everything else you listed (heroContent, stage) already goes in-flow or is hidden there. So:

• backdropGlow goes position: absolute below 1024px, inside heroScope (already position: relative, already bounding the pin-and-cover), mirroring the exact pattern heroContent uses ~100 lines away in the same file. top drops to 0 in the absolute case, since heroScope already starts below the header — keeping the header offset would've shifted the glow down 48px, which is the bug you flagged in the naive version of this fix.
• overscroll-behavior-y: none is scoped to @media (min-width: 1024px), where the remaining pinned layers still need it.

Desktop is untouched — same three layers, same suppression.

Verification

Computed styles at both widths (Playwright, real next dev build):

	390px (mobile)	1280px (desktop)
overscroll-behavior-y	auto	none (unchanged)
backdropGlow position	absolute	fixed (unchanged)
backdropGlow top	0px	48px (unchanged)

That confirms the mechanism directly: pull-to-refresh's gate (overscroll-behavior-y) is auto on mobile now, and the layer that was reaching past the footer (backdropGlow) no longer has position: fixed there.

On the screenshot: I tried to also produce a before/after of the actual overscroll-bleed artifact you documented (translating <body> past its scroll bounds to fake the rubber-band gap), the same shape as your own iOS-simulator evidence. It didn't work — a CSS transform on <body> doesn't reproduce genuine elastic-overscroll compositing, so both attempts rendered an identical clean strip regardless of whether the fix was applied, which means it wasn't testing anything real. I'm not including those as evidence since they'd be misleading. I don't have a way to drive genuine touch-overscroll or an iOS Safari session in this environment, so real pull-to-refresh and the visual-bleed check are unverified beyond the computed-style proof above — flagging that plainly rather than claiming coverage I don't have.

Lint clean, no typecheck errors from either changed file (ran tsc --noEmit and grepped for the two file paths specifically — a full production next build currently fails on an unrelated pre-existing type error in a generated example, BannerCollapsibleContent.tsx, present on a clean checkout before I touched anything).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
astryx	Error		Aug 25, 2026 1:14am

[xinzi-bot]

🔴 Blocking — line-ending churn in apps/docsite/src/app/globals.css (needs:code-review, community contribution).

Triage: bug fix · non-breaking · docsite-only · low blast radius → standard path (visual/behavioral change, verified in a real browser)

Thanks for this — the diagnosis is right, the mechanism is right, and I appreciate you being explicit about what you could and couldn't verify rather than claiming coverage you didn't have. I verified the fix independently and it does what it says. One mechanical blocker before it lands.

Verification

Local next dev at 650bad43, Playwright, same server with the two files flipped between main and this head, canary banner hidden so the layout matches production:

viewport	html overscroll-behavior-y	backdropGlow position	glow document Y at scroll 0
390 × 844	none → auto	fixed → absolute	48 → 48 (unchanged)
820 × 1180	none → auto	fixed → absolute	48 → 48 (unchanged)
1280 × 900	none → none	fixed → fixed	48 → 48 (unchanged)

Document height and scrollWidth are identical before/after at all three widths — no horizontal overflow from width: min(1200px, 100vw) resolving against heroScope instead of the viewport. Desktop is unchanged on every axis I measured.

I also ran it on a real iOS 26.5 Safari simulator (iPhone 17 Pro, genuine touch gestures), which closes the gap you flagged:

• Pull-to-refresh completes — the same 660pt pull at the top of / produces a reload navigation entry on this head, and nothing on live production.
• Bottom overscroll is clean — 89px elastic gap under the footer, elementFromPoint in the gap returns <html>, no aurora.
• Top overscroll is clean too — the strip the pull exposes is the cream nav band, not the glow.
• Fixed-layer census at 390px across /, /components/Button, /blog and /docs/getting-started: the only position: fixed element left below 1024px anywhere is navBackdrop, a 48px strip at the very top of the home page. It cannot reach the bottom gap.

So your computed-style proof holds up under real touch, and both halves of the change are load-bearing — auto alone bleeds.

🔴 B1 — globals.css was re-saved with CRLF, rewriting all 205 lines

git diff --numstat                     → 211 / 205
git diff --numstat --ignore-cr-at-eol  →  10 /   4

The actual change is 10 lines. Every other file in the repo is LF, and nothing normalizes CSS line endings today — your own #5229 proposing exactly that is still open, so it can't be relied on here.

Three concrete costs: the real change is invisible in review, git blame for the whole file lands on this commit, and any concurrent edit to globals.css conflicts wholesale.

git config core.autocrlf false
git add --renormalize apps/docsite/src/app/globals.css

Purely mechanical — nothing about the fix itself is in question.

🟡 N2 — the new comment inverts the causal chain

HeroThemeReel.tsx, on backdropGlow:

a fixed glow below 1024px reached past the footer into the bottom-overscroll gap, which blocked pull-to-refresh from ever registering a release

The glow never blocked pull-to-refresh — overscroll-behavior-y: none did. The glow is the reason that rule was written, which is exactly why bounding it is what lets the rule retreat to desktop. Worth correcting, because this comment is the durable record of the decision and the next person to touch these layers will reason from it.

Something like: "a fixed glow below 1024px reached past the footer into the bottom-overscroll gap — the exposure the global overscroll-behavior-y: none was suppressing, at the cost of pull-to-refresh site-wide (#5392)."

Note on CI

The Vercel check failure is not yours — #5414, #5398 and #5362 all fail the same deployment check on unrelated heads. Every GitHub Actions check on this PR is green.

---

Requesting changes on B1 alone. Fix the line endings and this is good to go on my end — the mobile fix is correct, complete, and now verified on-device. Deferring desktop to a follow-up is the right call and matches how #5392 scopes it.

[imdreamrunner]

Thanks for your contribution! (human here, I added a small fix to change the newline style, otherwise nothing is changed)