What we scan
- Linked stylesheets (
<link rel="stylesheet">) — top 8 per site, capped at ~5 MB of CSS payload - Inline
<style>…</style>blocks in the document - Inline
style="…"attributes on elements
PalettePunk scans your iOS or web codebase, finds every gradient stuck in muddy sRGB and every color literal locked out of Display P3, and writes the wide-gamut fix as one git-reversible commit. $99 lifetime license. Nothing uploads. Old browsers and old iOS keep working exactly as they did.
| apple.com | 13 sRGB-only gradients · 1,291 sRGB tokens · zero color(display-p3) calls |
40/100 |
| figma.com | 5 sRGB-only gradients · 735 sRGB tokens · zero wide-gamut anywhere | 40/100 |
| youtube.com | 192 sRGB-only gradients · 2,745 sRGB tokens · 2.9 MB of CSS, zero modern color | 40/100 |
Three sites. Three 40s. The internet is shipping 2014 color in 2026 and somehow nobody else is mentioning it.
wikipedia-ios / WMFYearInReviewSlideHighlightsView.swift:42.
Best viewed on a Display P3 screen (any Mac, any iPhone Pro since 2016).
SmoothGradient Swift package — exists only because Apple's default is muddy. 187 devs hand-built the fix.| App | Files | Status | Score |
|---|---|---|---|
| wikipedia-ios | 1,230 | CRITICAL · SRGB MUD | 42 |
| wordpress-ios | 2,609 | CRITICAL · SRGB MUD | 58 |
| element-ios | 1,884 | HIGH · GAMUT LOCKED | 78 |
| openterface | 147 | HIGH · GAMUT LOCKED | 78 |
| firefox-ios | 1,988 | MED · sRGB BIAS | 84 |
| loop | 149 | MED · GRADIENT LERP | 88 |
| duckduckgo-ios | 1,197 | LOW · MINOR DRIFT | 92 |
| kickstarter-ios | 2,035 | LOW · MINOR DRIFT | 94 |
| mastodon-ios | 793 | LOW · MINOR DRIFT | 98 |
| signal-ios | 2,542 | CLEAN | 100 |
| wire-ios | 4,875 | CLEAN | 100 |
| netnewswire | 601 | CLEAN | 100 |
PLAIN ENGLISH Twelve of these ten apps have real visible color tech debt that PalettePunk catches in seconds. The worst — Wikipedia at 42/100 — has a Year-in-Review hero gradient with 5 sRGB-locked color stops that an iPhone Pro can render dramatically better with one engine import. If Wikipedia ships this, your app probably does too.
Three steps. No installs, no SDKs, no servers. The whole tool runs in your browser.
Don't trust a web tool with your whole repo? Fair. Start with one file. Drop just your Colors.swift, Theme.swift, or tailwind.config.js to see PalettePunk work locally before you give it the project. Nothing uploads. Browser reads the file in-place via the File System Access API. WebAssembly engine scans it. Findings render in your tab.
Once Micro-Drop convinces you it stays local, drop the project folder. PalettePunk walks every .swift file and flags the three rule families: gradients lerped in RGB, Color literals locked to sRGB, missing MeshGradient opportunities on hero surfaces. Severity-tagged. File:line referenced. Side-by-side rendered.
Each finding gets Smart Context (why it matters, the math, references, plain-English close). Pick the fixes you want. PalettePunk generates a palettepunk/<project>/<timestamp> git branch with one clean commit. Review in Xcode. git revert if you hate it.
PalettePunk Web runs as a WebAssembly bundle in your browser. There's no database. No upload endpoint. No "we anonymize the data" lie. We literally do not have a place to put your code even if we wanted to.
Every patch is additive. Your existing LinearGradient stays as the fallback; the modern syntax is added as a second declaration (CSS) or wrapped in if #available(iOS 18, *) (Swift). Old browsers and old iOS see exactly what they used to. New browsers and new iOS see the win. Nobody loses anything.
Zero dependencies. PalettePunk outputs raw CSS and native SwiftUI / UIKit color literals. No npm package. No Swift Package Manager bloat. No proprietary PalettePunk-SDK. We change your hex codes and get out of your way.
Every Apply Fixes is one PalettePunk branch with one commit via libgit2. Don't like it? Switch back to your branch and delete the PalettePunk branch in 5 seconds. Or git revert. The tool refuses to operate on a dirty working tree — your changes never tangle with ours.
PalettePunk is deterministic SwiftSyntax-based static analysis with a small color-science library (RadioArtColor) underneath. The only on-device language model use is Apple's FoundationModels framework for short fix-rationale explanations — we call that "Smart Context," not "AI." Your scans are reproducible. Nothing hallucinates.
Run on 2026-06-23 against the public marketing/product surfaces of each site. Open-source script. Open-source results. Re-run it yourself.
| Site | Top finding | Score |
|---|---|---|
| apple.com | 13 sRGB-only gradients · 1,291 sRGB tokens · 0 modern color fns | 40 |
| behance.net (Adobe-owned) | 30 sRGB-only gradients · 1,108 sRGB tokens · 0 modern fns | 40 |
| figma.com | 5 sRGB-only gradients · 735 sRGB tokens · 0 color(display-p3) | 40 |
| google.com | 17 sRGB-only gradients · 462 sRGB tokens · 0 modern fns | 40 |
| youtube.com | 192 sRGB-only gradients · 2.9 MB CSS · 0 modern fns anywhere | 40 |
| notion.so | 9 sRGB-only gradients · 197 sRGB tokens · 0 modern fns | 40 |
| linear.app | 17 of 18 gradients sRGB-only · 131 sRGB tokens · 3 modern fns | 43 |
ADOBE NOTE adobe.com kills our HTTP/2 stream mid-connection. We literally couldn't audit them. So we audited Behance, which Adobe owns. It scored 40/100 with zero modern color functions. Make of that what you want.
ALSO RAN Vercel scored 76, Shopify 72, Spotify 75, Framer 70 — they actually ship some color(display-p3) and oklch(). The good news is the floor is rising. The bad news is half a dozen of the biggest brands on the internet are still at the floor.
We are going to tell you the limits of our own script. That way the HN commenter saving the takedown for tomorrow morning can save it.
<link rel="stylesheet">) — top 8 per site, capped at ~5 MB of CSS payload<style>…</style> blocks in the documentstyle="…" attributes on elementslinear-gradient, radial-gradient, conic-gradient) without modern interpolation (in oklch, in oklab, in display-p3)#hex, rgb(), rgba()) without any complementary wide-gamut usage (color(display-p3 …), oklch(), oklab())color-mix, oklch, oklab, color(display-p3)) as a signal of modernity — present, absent, frequencyOur scanner parses static and dynamically loaded CSS payloads. It does not currently evaluate <canvas> output, inline <svg> injected post-load, or encoded color profiles in raster images. If Apple is hiding Display P3 inside a Shadow DOM root we couldn't scrape, we'll update the board. But for standard CSS tokens — which is the entire color-system layer for every site on this audit — these numbers are exact.
0–100 scale. Gradients without modern interpolation lose up to 40 points (proportional to ratio). A large palette of sRGB color literals with zero wide-gamut alternatives loses 30. Presence of modern color functions earns up to 10 back. Floor: 40. Ceiling: 100. The full Python source ships with the v1.0 release; pre-release copy lives at palettepunk.com/methodology.
If a single number on the leaderboard is wrong on a site you can prove it on, email [email protected] with the receipt and we update it within 24 hours. We'd rather be corrected than be cute.
One price. Pay once. Use forever. No subscription. No SaaS. No expiring license.
Web tool ships within 2 weeks · Mac app ships Q1 2027 · HDR upgrade ships Q1 2027 — all included.
PalettePunk patches are additive, never replacive. Your existing code stays as the fallback. The modern syntax is added as a second declaration (CSS) or wrapped in if #available(iOS 18, *) (Swift). Old devices see exactly what they used to. Modern devices see the win. No user sees a worse experience than they would have before.
Yes. The same rule families apply: linear-gradient(...) without in oklch, color(display-p3 ...) opportunities, oklch() palette extensions. PalettePunk Web scans your tailwind.config.js, your generated CSS, and any inline style="..." attributes. The patch follows Tailwind conventions — we don't replace your design tokens, we add wide-gamut variants.
No. Never. The web tool uses the browser's File System Access API to read your project folder in-place; the SwiftWasm engine scans it locally inside your browser tab. The Mac app uses SwiftSyntax to parse on your Mac. We never see your code. We don't run a server that could see your code. We don't have a database that could store your code. Our entire backend is a static HTML/CSS/WASM bundle on Cloudflare Pages.
Two options. (1) Micro-Drop: drop just your Colors.swift or tailwind.config.js — one file, almost-zero IP exposure, lets you verify the tool works without uploading. (2) Offline CLI: brew install palettepunk, runs entirely on your machine, never touches the network. The CLI ships free with the $99 lifetime license.
Yes — PalettePunk doesn't change your brand colors. We keep your existing source hex codes; we just render them with better math (perceptual OKLab interpolation through Display P3). The brand red stays the brand red — it just looks like the brand red the designer originally specified in Figma, not the slightly-flatter version sRGB gives you on a Pro display.
It can't. We require a clean git working tree before applying. We create a separate palettepunk/<project>/<timestamp> branch with one commit. If anything looks wrong: switch back to your main branch and delete the PalettePunk branch. Under 5 seconds to undo.
Because shipping color science correctly requires six teams to coordinate — designer, brand, developer, build system, browser/OS, asset pipeline. Most teams have at least one broken link in that chain. PalettePunk solves the developer's slice conservatively and correctly, so the dev can ship the win without needing anyone else's approval.
No. PalettePunk is deterministic SwiftSyntax-based static analysis with a small color-science library underneath. The only on-device language model use is Apple's FoundationModels framework for short fix-rationale explanations — and we call that "Smart Context," not "AI." Your scans are reproducible across runs; nothing hallucinates.
For a small project (under 100 Swift files): 10-30 seconds end-to-end. Medium (100-500 files): 30-90 seconds. Large (500-3,000 files): 1-4 minutes. The scan itself is under 10 seconds even on huge projects.
v1.0 is Swift + CSS only. The rules transfer 1-for-1 to web CSS (we already verified — see our parallel audit of apple.com, stripe.com, vercel.com). If you ship a React Native iOS app and want the underlying native Swift wrappers audited, PalettePunk v1.0 still helps with the wrapper layer.
PalettePunk reads your project's minimum deployment target from Package.swift or .xcodeproj and adapts. iOS 18+ targets get the modern syntax directly. iOS 17 and earlier get the modern syntax wrapped in if #available(iOS 18, *) { ... } else { ... } with your existing code as the else branch. Old-iOS users get exactly what they had before.
v1.0 is the first chapter. Part 2 (HDR) is already specced. Part 3 (paint match) is on the horizon. $99 lifetime includes all of it.
Display P3 · OKLab · MeshGradient
Catch every SwiftUI gradient interpolating in muddy sRGB. Catch every Color(red:green:blue:) locked out of the wider gamut your customers' devices have shipped since 2016. Surface every hero surface that could use a MeshGradient. Fix all of it with one git-reversible commit.
"The gradient looks muddy in the middle" — solved.
EDR headroom · selective glow · graceful SDR fallback
iPhone Pro displays have 1.6x-2x EDR headroom most apps never use. PalettePunk v1.5 surfaces every hero surface that could glow — sun glints in your weather app, ice sparkle on your hero card, lightning in your animation. Renders brighter-than-white selectively. SDR fallback on standard devices is automatic. Same additive-patch model: no user loses.
Be the first to know when v1.5 ships →
Display P3 → Benjamin Moore · Sherwin Williams · Pantone
Pull any Display P3 color out of your app. Get the closest Benjamin Moore, Sherwin Williams, Behr, Farrow & Ball, RAL, or Pantone code — with fidelity scores. For the designer who wants their app's hero color to match their living room wall. For the brand that wants pixel and print to actually align.
"Your app and your living room, same color."