chore: back-merge main → dev (pre Sprint 157 release) - #2803
Conversation
…try (TASK-21147) Five send flows fell back to txResult.userOpHash when the receipt wait gave up. A userOp hash is bundler-internal — eth_getTransactionReceipt can never resolve it, so backend validation burned its full retry window and marked SUCCESSFUL payments FAILED (11 in the 28-day prod baseline). - useZeroDev: when waitForUserOperationReceipt fails, re-poll getUserOperationReceipt (6x3s) before giving up — the tx usually landed seconds later; rescued receipts recorded to PostHog. - resolveSettledTxHash: shared resolution for all 5 sites — receipt hash first, coordinator txHash next, userOp hash only as measured last resort (send_txhash_fallback event per use). - direct-send flow: send_latency_breakdown event splits the client leg (charge create / sign+bundler+receipt / record payment) — the p50 7.9s block prod DB timing cannot attribute.
- Receipt rescue is a capped retry of viem's own wait (15s wall clock, 500ms polling) instead of a hand-rolled 3s-step poller; skipped entirely after a genuinely exhausted 120s timeout; checks the userOp success flag so a mined-but-REVERTED op is never rescued into a recordable hash (fail-open risk); carries elapsed_ms + chain_id. - mixedEphemeralSpend gets the same rescue (its stamp path was the remaining userOpHash-as-txHash producer); a rescued reverted op takes the ok:false exit so the passkey fallback still runs. - resolveSettledTxHash applied to the 4 sites the first pass missed (withdraw/crypto x2, useSpendBundle stamps x2, bank withdraw) — the exact pre-fix ordering bug lived on in withdraw/crypto. - Demo mode no longer pollutes send_txhash_fallback / latency metrics; an all-empty txResult throws instead of recording undefined; latency capture is throw-guarded; stale fallback comments removed; timing consts scoped inside the try.
The normal return exposed a reverted userOp's successful EntryPoint bundle receipt — downstream flows recorded a transfer that never moved funds as a successful payment. Same guard the rescue path already has.
…ions only /careers was a 30-word interstitial: a mascot, one sentence, and a link out to Notion. No <h1>, no footer, no Crisp, and metadata still selling 'Instant Global P2P Payments' from the protocol era. The footer's Jobs link bypassed it entirely, pointing at a second URL for the same Notion page. The page now carries its own content — what Peanut is, how the work works, and the open roles — sourced from content/press/en.md so nothing is invented. Notion is reached only by 'Apply on Notion'. The footer Jobs link now points at /careers instead of routing around it. Uses the public *.notion.site URL; the old www.notion.so workspace URL bounced logged-out visitors to a login.
shhhhh.faq.q6 read 'We're rolling out to select regions first and adding more as our partners' coverage expands. Eligibility is shown during signup.' That is issuance and availability framing; product/card.md requires public card copy to describe acceptance only and never name countries or promise eligibility. Now answers where the card can be used — accepted wherever Visa is accepted — and sends people to the door for their own account status. es-AR gains a q6 answer override for the voseo imperatives, since es-419 uses tuteo here. The key exists in es-419, so the delta stays a strict subset. The $10 reward claims this FAQ used to carry were already removed on dev.
Three marketing routes that carried no unique content, all now permanent redirects. (/lp/card was retired separately on dev by #2760.) - /exchange -> /. Rendered the same NoFees component already on the homepage plus a footer, and shipped an empty client shell: 'ZERO TRANSFER FEES', 'no hidden fees' and 'See the markup on' all return zero matches in the live server HTML, while it sat in the sitemap at priority 0.7. Its metadata title promised 'Live Exchange Rates' while its H1 said 'ZERO TRANSFER FEES'. NoFees itself is untouched and still used by the homepage and /quests. - /:locale/blog -> /:locale/content. Byte-identical to /content with fixedType='blog', which /content's own Blog chip already produces. Only the index is deleted; all 21 posts under blog/[slug] survive. - /:locale/team -> /:locale/press. /press already renders the same two members from the same content file, with real photos. Fallout cleaned up in the same pass: robots.ts no longer allows a route that is now a 308, and the blog post breadcrumb — which feeds a BreadcrumbList JSON-LD — points at /content?type=blog instead of naming a redirecting URL. That form matches how /compare and /use-cases already link back, so the crumb still reads 'Blog'. Also drops the now-dead 'team' entry from ROUTE_SLUGS and refreshes three comments that cited the retired routes as live.
…root identity - /crisp-proxy, /kyc/success and /maintenance defined no metadata, so all three inherited the root layout's canonical: '/' and declared the homepage as their canonical while serving index, follow. Each is a client component, so metadata cannot live in the page — added sibling layouts carrying a self-canonical plus robots noindex, the pattern /exchange already used. - Root title was 'Peanut - Instant Global P2P Payments in Digital Dollars', protocol-era framing inherited by every page without an override. The root description said the same thing. Both now describe the money app. - The X handle @peanutprotocol appeared in six places while the footer links x.com/joinpeanut. All six now say @joinpeanut, including the Organization JSON-LD sameAs and llms-full.txt. - Adds the five card legal pages to the sitemap. They return 200 with index, follow and were listed nowhere. English only: the localized copies serve English and already canonicalize to /en, so listing all twenty would create fifteen duplicates competing with their own canonical.
…ate, localize /press
- /en/content and /en/blog emitted zero article links in server HTML while the
sitemap listed 21 posts. ContentLanding is a client component read through
nuqs, so Next prerenders the Suspense fallback — and the fallback was a
pulsing skeleton. The list rendering is now an exported ContentLinkList that
touches no URL state, and the fallback renders it, so the links exist in the
prerendered HTML. Interactive filtering is unchanged.
- Content cards rendered the raw frontmatter title, leaking the SEO suffix into
the UI ('Lynn Pham: The Brand Girl | Peanut'). Stripped at render time only;
sitemap and metadata still read the raw value.
- /en/stories linked to /en/stories/index, a live second copy of the hub with a
self-referencing canonical and different hero copy. The page now applies the
same slug === 'index' guard the sitemap already had.
- /press rendered all six section headings as hardcoded English on a [locale]
route. They now come from the marketing catalogs, translated in all four.
- /press absorbs the team data: its Organization JSON-LD gains the member[]
array that only /team used to emit, so deleting /team does not drop the
site's only Person schema.
Drops the now-unused fixedType prop and the teamTitle/teamSubtitle keys, both
orphaned by the routes retired in this branch.
…nce and the waitlist
The ragdoll had no prefers-reduced-motion handling at all. It now follows the
same shape as confetti.ts — matchMedia read live per mount — and degrades
rather than disappearing: the world is stepped off-screen until the peanut
settles, then painted once, with no RAF loop and no pointer listeners.
Two new homes, both non-scrolling panes with a definite height:
- /maintenance replaces a static crying-peanut image. The copy was the worst
voice on the site ('We are currently going through maintenance... Sorry for
the inconvenience') and a button labelled 'Contact Support?'. Rewritten to
match the 404's register.
- The waitlist jail pane, gated on step === 'jail' so plain /invite — a signup
conversion surface — does not get a draggable toy next to its CTA. The
illustration pane moves from min-h to a definite h-[55dvh]; min-h gives
h-full nothing to resolve against, so the canvas would have silently
collapsed to zero height while the loop kept running.
Fixes a real bug found while placing it: the (mobile-ui) layout registers
pull-to-refresh on document before the waitlist early-return, and its
#scrollable-content guard is absent on that screen, so the handler fell
through to true. Dragging the peanut down ~160px would have reloaded the page
and wiped the user's typed invite code. The host now swallows touchstart — a
native listener, because React hydrates on document and a synthetic
stopPropagation would not stop a same-target sibling.
startRagdoll now runs in a try/catch: it throws synchronously on a failed 2D
context, which would white-screen /maintenance, the one page whose job is to
work when nothing else does.
Also retires the stale 'triple-click Rewards title' comments — that easter egg
was deliberately cut in 347757d and never shipped.
The footer nav now points at /careers rather than the Notion board, so the label should name the page it opens. Renames footerJobs to footerCareers and sets the English value to 'Careers' (it read 'Career'). pt-BR moves from 'Vagas' — openings — to 'Carreiras'; es keeps 'Empleo', which already reads correctly. Note the SEO footer's own jobs entry, which still carries a Notion href in the content manifest, is already suppressed by RESOURCES_MOVED_ELSEWHERE, so it renders nothing and needs no change here.
The hub delink hid the duplicate but generateStaticParams still emitted
/{locale}/stories/index. Filter it out so dynamicParams=false 404s it.
chore(marketing): consolidate the stray pages, and give /careers a real one
- Rescued-but-REVERTED ops now THROW from the rescue path too (they returned success-shaped results and flowed into the userOpHash fallback + success screen). - returnRevertedReceipt opt-in restores the kernel-migration contract: the noop wiring sites receive the bundle receipt on revert so ensureRootValidatorMigrated verifies on-chain truth instead of hard-failing a healthy wallet. - Rescue extracted to shared rescueUserOpReceipt (used by useZeroDev + mixedEphemeralSpend) with telemetry on BOTH paths + 4 tests.
roboto-flex-bold.ttf was the full 13-axis Roboto Flex variable font — 1.6 MB on disk, 964 KB over the wire, on every page. The CSS only ever drives the wght axis, so the woff2 is instanced down to wght 100–1000 (all other axes pinned to their defaults — rendering is identical) and subsetted to latin + latin-ext plus the arrows/math symbols the UI uses (fonttools). 964 KB -> 53 KB transfer. knerd-outline/knerd-filled get the same subset treatment (113 KB -> 73 KB combined). Their .ttf sources stay: the OG image routes read them at runtime (satori needs TTF). roboto-flex-bold.ttf had no other reader, so it's gone. Lighthouse (local prod build): font transfer 1,244 KB -> 311 KB, total page weight -926 KiB. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013wrRmcBRas8FdBr4n8btrL
…usel The whistling mascot is the mobile LCP element, but without `priority` next/image emits loading="lazy" — Lighthouse measured a 7s image load delay inside a 19.5s LCP. With priority the image is preloaded and the load delay drops to ~0.8s (LCP 19.5s -> 6.9–14.7s on a throttled run; the remainder is main-thread render delay from the app-wide JS bundle, not the image). TweetCarousel statically imported the testimonials manifest (~64 KB of JSON) into the landing chunk while rendering far below the fold — moved behind next/dynamic (SSR stays on, so crawlers still see the tweets). Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013wrRmcBRas8FdBr4n8btrL
Lighthouse a11y/SEO fixes, no visual change: - RegulatedRails wrapped its "Learn more" link in an h6 under an h1 — it's a link line, not a heading, so it's a p now; the link also gets an sr-only suffix (the section heading) so it isn't a bare generic "Learn more" for screen readers or the SEO link-text audit. - securityBuiltIn's section heading sat as an h1 directly above h3 feature titles (heading-order skip) — h2 now. - PartnerLockup's aria-label replaced the visible "Learn more" text instead of containing it (WCAG 2.5.3 label-in-name) — the visible label now prefixes the description. Landing a11y 93 -> 95, SEO 92 -> 100. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013wrRmcBRas8FdBr4n8btrL
…images Next 16 deprecates next/image's `priority` in favor of `preload` (same semantics: preload || priority disables lazy loading and emits the preload link). Flagged by review on #2788. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013wrRmcBRas8FdBr4n8btrL
…phash fix: never record a userOp hash as a tx hash + client-leg send telemetry (TASK-21147)
…0821-191753 content: publish latest to production (src/content → peanut-content@4eb2c3f)
…-speed-mngr5x perf(landing): fix lazy-loaded LCP, ship subsetted woff2 fonts, a11y/SEO audit fixes
…cation-error-uzgmxr fix(claim): stop blaming the sender, and offer a retry when one would work
…0824-100955 content: publish latest to production (src/content → peanut-content@26a4263)
…0824-115455 content: publish latest to production (src/content → peanut-content@41ba07b)
…into-dev-20260824
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
Code-analysis diffPainscore total: 7317.77 → 7334.68 (+16.91) 🆕 New findings (166)
…and 146 more. ✅ Resolved (175)
…and 155 more. 📈 Painscore deltas (top movers)
|
🧪 UI test report — ✅ all greenSuites
📊 Coverage (unit)
⏱ 10 slowest test cases
|
Back-merge of
mainintodevahead of the Sprint 157 prod release, so the dev→main release PR is a clean superset.Carries the content-submodule bumps, SEO/perf landing hotfixes, #2779 (userOpHash), #2776, #2783 (stray pages). Clean merge — no conflicts.
Part of the Sprint 157 release prep.