@kortix/ui 0.0.0-stage → 0.14.4

package/dist/internal/lib/browser-noise/rules/graphics.js

@@ -0,0 +1,606 @@
+import { isFirstPartyResolvedSource, normalizeString, sourcesOf, stripErrorWrappers, } from '../evidence.js';
+// Paper Shaders (`@paper-design/shaders-react`) null-WebGL-context crash class.
+// On GPUs/browsers without working WebGL2 (context loss, blacklisted driver,
+// stripped WebView, headless renderer), Paper Shaders' shader-mount
+// `useEffect`/rAF callback reaches a WebGL2 context that has become `null` and
+// calls a WebGL API method on it → `TypeError`. The throw happens INSIDE an
+// async callback, so it ESCAPES the `<ShaderSafe>` React error boundary (which
+// only catches render-phase throws via `getDerivedStateFromError`) → global
+// error → Sentry → Better Stack. The two observed null-context method names are:
+//   - `getSupportedExtensions`  (Better Stack pattern `34127fa4…` / recurrence
+//                                `dfcb336b…`, call site `new b2` in chunk
+//                                `c76173f0.…`, prod)
+//   - `getAttribLocation`       (the known sibling already documented in
+//                                `shader-safe.tsx`'s probe rationale).
+// These are WebGL2 context method names — they are NEVER called from
+// first-party app code (only from Paper Shaders' library internals), so the
+// message wording alone is specific enough to safely classify as noise without
+// a chunk-frame anchor (unlike the generic old-browser SyntaxError class). The
+// matching covers all five JS-engine / DOM-binding wordings for the same
+// null-context bug:
+//   - V8 (Chrome/Edge):          `Cannot read properties of null (reading '<m>')`
+//   - old JSC (old Safari/iOS):  `Cannot read property '<m>' of null`
+//   - SpiderMonkey (Firefox):    `can't access property "<m>"<…>` (the variable
+//                                name after the method is library-specific, so
+//                                the pattern anchors on the stable method-name
+//                                prefix only — see the recurrence
+//                                `dfcb336b…` which shipped through PR #4544's
+//                                V8/JSC-only filter as
+//                                `can't access property "getSupportedExtensions",
+//                                this.gl is null`).
+//   - modern JSC (Safari / Chrome-on-iOS CriOS, which uses WebKit/JSC rather
+//                                than V8): `null is not an object (evaluating
+//                                'this.gl.<m>')` — the `this.gl.` token and the
+//                                `(evaluating '...')` wrapper are JSC-specific;
+//                                the pattern is the exact full message per
+//                                method so a generic JSC `null is not an object
+//                                (evaluating '<other expr>')` throw does NOT
+//                                match (pattern `a8754de5…`).
+//   - Gecko (Firefox) DOM-binding: `WebGL2RenderingContext.<m>: Argument 1 is
+//                                not an object.` — Firefox's DOM bindings throw
+//                                on the method call itself (a DIFFERENT code
+//                                path from SpiderMonkey's engine TypeError
+//                                above) when the `this` binding is not a valid
+//                                object (here the null WebGL2 context). Pattern
+//                                `fd773de2…` (Firefox 152 on Android 17,
+//                                `getAttribLocation`, marketing homepage).
+//   - Paper Shaders library's OWN internal guard: the bare `this.gl is null`
+//                                string. This is NOT a JS-engine TypeError and
+//                                NOT a Gecko DOM-binding message — it is the
+//                                library's OWN explicit `throw new Error(
+//                                'this.gl is null')` (or equivalent assertion
+//                                message) when its internal state check detects
+//                                that `this.gl` (the WebGL2 context it cached at
+//                                mount) is `null`. Whereas every other entry is
+//                                the JS engine / DOM binding wording the library
+//                                triggered by dereferencing the null context,
+//                                this is the library's OWN wording — it fires on
+//                                engines that DON'T surface a JS-engine
+//                                TypeError for the same deref (e.g. some Firefox
+//                                / SpiderMonkey builds where the method call is
+//                                short-circuited by the library's guard before
+//                                the engine ever throws). Pattern `f0c8c422…`
+//                                (Firefox 137.0 on Windows 10, Gecko engine,
+//                                `/projects/:id` project page, post-v0.12.7).
+// `TypeError: ` / `Error: ` / `Unhandled promise rejection: ` wrappers are
+// stripped before matching so all capture paths (window.onerror,
+// onunhandledrejection, Sentry exception) classify consistently.
+// `shouldIgnore*` here is the leak-path backstop for the throws that still
+// escape `<ShaderSafe>` after a context-loss event; the `supportsWebGL2()`
+// probe in `shader-safe.tsx` is the primary guard that degrades to the fallback
+// BEFORE the throw. The probe is engine-agnostic (it just calls
+// `ctx.getSupportedExtensions()`, which throws or returns null on any engine),
+// so it already prevents the throw at mount for Firefox — the filter backstop
+// catches the residual async-context-loss throws that bypass the one-shot probe.
+const PAPER_SHADER_NULL_CONTEXT_NOISE_PATTERNS = [
+    // V8 (Chrome/Edge).
+    "Cannot read properties of null (reading 'getSupportedExtensions')",
+    "Cannot read properties of null (reading 'getAttribLocation')",
+    // Old JSC (old Safari/iOS).
+    "Cannot read property 'getSupportedExtensions' of null",
+    "Cannot read property 'getAttribLocation' of null",
+    // SpiderMonkey (Firefox) — anchors on the stable method-name prefix; the
+    // `, this.gl is null` variable suffix is library-specific and dropped so the
+    // pattern matches regardless of which Paper Shaders internal variable holds
+    // the null context.
+    'can\'t access property "getSupportedExtensions"',
+    'can\'t access property "getAttribLocation"',
+    // Modern JSC (JavaScriptCore — Safari / Chrome-on-iOS CriOS, which uses
+    // WebKit/JSC rather than V8). JSC wraps the offending expression as
+    // `null is not an object (evaluating '<expr>')`; the Paper Shaders library
+    // accesses the WebGL2 context as `this.gl.<method>`, so the prod wording is
+    // `null is not an object (evaluating 'this.gl.getSupportedExtensions')` and
+    // the `getAttribLocation` sibling. The `this.gl.` token and the
+    // `(evaluating '...')` wrapper are JSC-specific; the stable anchor is the
+    // exact full JSC message (per-method), so a generic JSC `null is not an
+    // object (evaluating '<other expr>')` throw does NOT match. Seen as pattern
+    // `a8754de5…` (1 occurrence, 0 users) from Chrome 150 on iOS 26.5.2 on
+    // `/projects/:id/sessions/:sessionId`, the fourth engine variant of this
+    // class after V8 (#4544), old JSC, and SpiderMonkey (#5172).
+    "null is not an object (evaluating 'this.gl.getSupportedExtensions')",
+    "null is not an object (evaluating 'this.gl.getAttribLocation')",
+    // Gecko / Firefox DOM-binding wording. When the WebGL2 context is `null` /
+    // invalid (context loss, blacklisted GPU, stripped WebView), Firefox's DOM
+    // bindings throw on the method call itself with the canonical Gecko DOM-API
+    // shape `<Interface>.<method>: Argument 1 is not an object.` — the
+    // `Argument 1 is not an object.` is Gecko's standard message for a `this`
+    // binding that is not a valid object (here the null WebGL2 context). This is
+    // the SAME null-WebGL-context crash class as the V8/JSC/SpiderMonkey entries
+    // above, just with Firefox's DOM-API error wording instead of an engine
+    // TypeError. Better Stack pattern
+    // fd773de23b8dbee3551f1132df1dc048a80307133e1e513ca2422ca2bc4fd29a
+    // (Kortix Frontend prod, application_id 2346967): `TypeError`, message
+    // `WebGL2RenderingContext.getAttribLocation: Argument 1 is not an object.`,
+    // 1 occurrence / 0 identified users, first 2026-08-07 19:34:33 UTC
+    // (post-v0.12.5, release `e2540c341c6f43536a7cf0e0b51599e9928f055c`),
+    // call site `setupPositionAttribute` in chunk
+    // `app:///_next/static/immutable/chunks/24zv25pg_k-nz.js`, request URL
+    // `https://kortix.com/` (marketing homepage), browser Firefox 152.0 on
+    // Android 17 (Gecko engine), mechanism
+    // `auto.browser.global_handlers.onunhandledrejection` (UNCAUGHT,
+    // `handled:false`). The `getSupportedExtensions` sibling is added
+    // preemptively — same class, Firefox may emit it too. The
+    // `WebGL2RenderingContext.<method>:` prefix is the Gecko DOM-binding's own
+    // canonical marker (the interface + method name), never emitted by
+    // first-party app code, so the message wording alone is specific enough —
+    // same message-only contract as the other engine variants (no chunk-frame
+    // anchor, no first-party negative guard). Note: `stripErrorWrappers`'s
+    // `[A-Za-z]+Error:` regex does NOT strip the
+    // `WebGL2RenderingContext.<method>:` prefix (it contains a `.`), so the
+    // pattern is matched verbatim by `.includes()` after the `TypeError: ` /
+    // `Unhandled promise rejection: ` wrappers are stripped.
+    'WebGL2RenderingContext.getSupportedExtensions: Argument 1 is not an object.',
+    'WebGL2RenderingContext.getAttribLocation: Argument 1 is not an object.',
+    // Paper Shaders library's OWN internal guard wording — the SIXTH variant of
+    // this null-WebGL-context crash class, and the ONLY one that is the library's
+    // OWN throw rather than a JS-engine TypeError or a Gecko DOM-binding message.
+    // When the library's internal state check detects that `this.gl` (the WebGL2
+    // context it cached at mount) is `null` (after a context-loss / GPU-blacklist
+    // event, or a stripped WebView that returned `null` from `getContext('webgl2')`
+    // and bypassed the `supportsWebGL2()` probe), it throws its OWN message
+    // `this.gl is null` directly — NOT a JS-engine `TypeError` from dereferencing
+    // the null context, and NOT a Gecko DOM-binding message. This fires on engines
+    // that DON'T surface a JS-engine TypeError for the same deref (e.g. some
+    // Firefox / SpiderMonkey builds where the library's own guard short-circuits
+    // the method call before the engine ever throws). Better Stack pattern
+    // f0c8c42213b12122948f4c8307b1eedb6a51afe9072460604e3be14e0277d3f2
+    // (Kortix Frontend prod, application_id 2346967): `TypeError`, message
+    // `this.gl is null`, 1 occurrence / 0 identified users, first 2026-08-10
+    // 14:35:19 UTC (post-v0.12.7), request URL
+    // `https://kortix.com/projects/<project_id>` (project page), browser Firefox
+    // 137.0 on Windows 10 (Gecko engine), mechanism
+    // `auto.browser.global_handlers.onunhandledrejection` (UNCAUGHT,
+    // `handled:false`), 3 frames in chunk
+    // `app:///_next/static/immutable/chunks/2_t47hwky1w2m.js` (Paper Shaders
+    // library). Message-only contract (no chunk-frame anchor, no first-party
+    // negative guard) — same as the other engine variants — because (a) `this.gl
+    // is null` is the library's OWN canonical wording, never a coincidental
+    // app-logic phrase (no first-party `apps/web/src/…` code holds a `this.gl`
+    // field — confirmed by `rg "this\.gl" apps/web/src`), and (b) the unhandled-
+    // rejection stack carries only minified `@paper-design/shaders` chunk frames,
+    // so a first-party negative guard would never fire for this class anyway. The
+    // substring match is specific enough that near-worded first-party null-derefs
+    // (`this.foo is null`, `this.context is null`, `this.canvas is null`, …) do NOT
+    // match — only the exact `this.gl is null` token does. `stripErrorWrappers`
+    // strips `TypeError: ` / `Unhandled promise rejection: ` prefixes, leaving the
+    // bare `this.gl is null` to match verbatim.
+    'this.gl is null',
+];
+// Paper Shaders (`@paper-design/shaders-react`) WebGL-unsupported deliberate
+// throw — a SIBLING of the null-context crash class above, but a DIFFERENT
+// throw. When the library's shader mount detects that WebGL is unavailable
+// (a stripped-down/mobile WebView, a headless renderer, a browser with WebGL
+// disabled, or a GPU blacklisted at context creation), the library throws its
+// OWN deliberate `Error('Paper Shaders: WebGL is not supported in this browser')`
+// from its constructor — NOT a null-context `TypeError` from calling a WebGL2
+// method on a `null` context (the `getSupportedExtensions` / `getAttribLocation`
+// wording covered by `PAPER_SHADER_NULL_CONTEXT_NOISE_PATTERNS` above). The
+// `Paper Shaders:` prefix is the library's own canonical marker, so this exact
+// message is the library's deliberate signal that the browser cannot render the
+// decorative shader; it is an EXPECTED degradation state on WebGL-less browsers,
+// never a product bug.
+//
+// Better Stack pattern
+// f1abf79ece48a86faf8eb32cec8bbb6bf270627f9fd5d423fb1ee43b9abcfb23
+// (Kortix Frontend prod, application_id 2346967): `Error`, message
+// `Paper Shaders: WebGL is not supported in this browser`, 1 occurrence /
+// 0 identified users (anonymous), last 2026-07-23 17:26:32 UTC, release
+// `470fe6f3c88460212c3b187f6f86fb4ad456c4d6` (v0.10.13), route `/`
+// (marketing homepage), mechanism
+// `auto.browser.global_handlers.onunhandledrejection` (UNCAUGHT global
+// unhandledrejection — never reached a React error boundary, `handled:false`).
+// Browser: Chrome 150.0.0.0 on Android 10 (mobile), UA
+// `Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko)
+// Chrome/150.0.0.0 Mobile Safari/537.36` — a stripped-down/mobile Android
+// browser without WebGL. Stack frames: 2, both minified
+// `@paper-design/shaders` chunk frames — NO first-party `apps/web/src/…`
+// frame:
+//   - `app:///_next/static/chunks/81107-7c84018ef9475be5.js?dpl=dpl_FWCk2e9rGNxkUxaBwBGi2iMZDfno`
+//     function `?` lineno 251 colno 3276
+//   - same chunk function `new a` lineno 401 colno 1131  (call_site_function)
+//
+// The `supportsWebGL2()` probe in `shader-safe.tsx` is the primary guard that
+// degrades to the fallback BEFORE this throw fires (it calls
+// `getContext('webgl2')` + `getSupportedExtensions()` on a probe canvas and
+// treats a `null`/throw as unsupported). But the probe is a one-shot memo that
+// runs at first `render` of `<ShaderSafe>`, while the library's own `new a`
+// constructor throws synchronously on a browser where WebGL is `null` — and on
+// some code paths the probe's result is computed after the library has already
+// been dynamically imported and its constructor reached. The residual async
+// throw then escapes as an unhandled rejection (the library constructor runs
+// inside a dynamic import / `useEffect` that bypasses the React error
+// boundary). This matcher is the leak-path backstop for that residual throw,
+// the way `isPaperShaderNullContextNoise` is the backstop for the null-context
+// `TypeError` class.
+//
+// The message is the library's OWN canonical string (the `Paper Shaders:`
+// prefix is the library's deliberate marker, never emitted by first-party app
+// code), so an EXACT-message match is safe — a real first-party
+// `throw new Error('Paper Shaders: WebGL is not supported in this browser')`
+// regression is vanishingly unlikely AND would de-minify to `apps/web/src/…`
+// frames, which the mandatory negative guard below preserves. Unlike
+// `isPaperShaderNullContextNoise` (message-only, no negative guard — safe
+// because WebGL2 API method names are never called from first-party code),
+// this message COULD theoretically be thrown from first-party code, so the
+// first-party-frame negative guard MUST run when frames are present. The prod
+// event has only minified `81107` chunk frames, so the negative guard does
+// not fire for it. A frameless capture with this exact message still
+// classifies as noise (the message alone is specific — the `Paper Shaders:`
+// library prefix is part of the anchor). `Error: ` / `Unhandled promise
+// rejection: ` / `Unhandled promise rejection: Error: ` wrappers are stripped
+// before matching so all capture paths (window.onerror,
+// onunhandledrejection, Sentry exception) classify consistently. Deliberately
+// NOT added to `sentry.client.config.ts`'s `ignoreErrors` list — that gate has
+// no frame context, so a bare-string match there could swallow a real
+// first-party throw the negative guard exists to preserve; the frame-aware
+// `beforeSend` hook (which calls `shouldIgnoreSentryBrowserNoise`) is the
+// only safe gate.
+const PAPER_SHADER_WEBGL_UNSUPPORTED_NOISE_MESSAGE = 'Paper Shaders: WebGL is not supported in this browser';
+// Canvas `getImageData` out-of-memory noise — a third-party canvas library
+// (e.g. a decorative background / hyper-logo animation effect on the marketing
+// homepage) called `CanvasRenderingContext2D.getImageData()` and the browser ran
+// out of memory allocating the `ImageData` buffer, surfacing as
+//   `Failed to execute 'getImageData' on 'CanvasRenderingContext2D': Out of
+//    memory at ImageData creation`
+// (V8/Chrome wording — a `RangeError`, NOT a `TypeError`). This is TRANSIENT
+// browser resource exhaustion: the canvas was too large / the tab was under
+// memory pressure / the device is low-RAM, so the engine failed the buffer
+// allocation. It is NOT a deterministic code bug — the same canvas renders fine
+// on the next visit once memory frees up. The throw fires from a
+// third-party library's `addEventListener` callback (Sentry's `BrowserApiErrors`
+// integration auto-wraps `addEventListener` on `EventTarget` and captures the
+// throw as `handled:false`, UNCAUGHT — it never reached a React error
+// boundary), and the stack frames are all minified `_next/static/chunks/…`
+// library frames with NO resolved first-party `apps/web/src/…` source.
+//
+// Better Stack pattern
+// b4b4384734b09b411e476591e3f9ac3ad88f110e0be91aae390913038f6844f0
+// (Kortix Frontend prod, application_id 2346967): `RangeError`, message
+// `Failed to execute 'getImageData' on 'CanvasRenderingContext2D': Out of
+// memory at ImageData creation`, 1 occurrence / 0 identified users, last
+// 2026-08-07 10:09:13 UTC, release
+// `160f0b286f0ad5c53debc343d5e055241694e24d` (v0.12.4 prod), call site
+// function `Image.<anonymous>`, call site file
+// `app:///_next/static/chunks/0fl4m2af7bsiq.js` (minified), request URL
+// `https://kortix.com/` (marketing homepage), browser Chrome 130 on Linux,
+// mechanism `auto.browser.browserapierrors.addEventListener` (UNCAUGHT,
+// `handled:false`). Stack: 2 frames, both minified third-party canvas library
+// chunk frames — NO first-party `apps/web/src/…` frame.
+//
+// The message is the browser's OWN canonical out-of-memory wording for a
+// `CanvasRenderingContext2D.getImageData()` allocation failure (the
+// `Failed to execute 'getImageData' on 'CanvasRenderingContext2D':` prefix is
+// V8's DOM-bindings exception format; the `Out of memory at ImageData
+// creation` suffix is the specific allocation-failure reason). This exact
+// string is the browser's, never an app-logic phrase — a real first-party
+// `throw new RangeError('…Out of memory at ImageData creation…')` regression
+// is vanishingly unlikely AND would de-minify to `apps/web/src/…` frames.
+// BUT `getImageData` IS a Canvas 2D API method that first-party code CAN call
+// (e.g. an image-processing helper, a screenshot/export path, a pixel-reader),
+// so — mirroring `isSafariGenericSecurityErrorNoise` /
+// `isOldBrowserDomNullDerefNoise` — the matcher carries a NEGATIVE guard: if
+// ANY frame (or the window.onerror `filename`) resolves to a de-minified
+// first-party `apps/web/src/…` source path, the event KEEPS reporting (our
+// own code is the `getImageData` caller → a real first-party OOM regression
+// we want to fix). Only events with NO resolved first-party frame (the prod
+// noise shape: all minified third-party canvas library chunk frames, or
+// frameless) are dropped. A frameless capture with this exact message still
+// classifies as noise — the message alone is the browser's canonical OOM
+// wording and is specific enough (the `CanvasRenderingContext2D` +
+// `getImageData` + `ImageData creation` tokens together pin this single DOM
+// API call site). Deliberately NOT added to
+// `sentry.client.config.ts`'s `ignoreErrors` list — that gate has no frame
+// context, so a bare-string match there could swallow a real first-party
+// `getImageData` OOM regression the negative guard exists to preserve; the
+// frame-aware `beforeSend` hook (which calls `shouldIgnoreSentryBrowserNoise`)
+// is the only safe gate. The runtime `window.onerror` gate
+// (`shouldIgnoreBrowserRuntimeNoise`) is also wired so a runtime capture with
+// the exact message + no first-party `filename` drops.
+const CANVAS_GETIMAGE_DATA_OOM_NOISE_PATTERNS = [
+    // The exact V8/Chrome message. Anchored as a full-match (the trailing
+    // `Out of memory at ImageData creation` is the specific OOM reason).
+    /^Failed to execute 'getImageData' on 'CanvasRenderingContext2D': Out of memory at ImageData creation$/,
+];
+/**
+ * Whether a message is the Paper Shaders (`@paper-design/shaders-react`)
+ * null-WebGL-context crash class: a `TypeError` from calling a WebGL2 context
+ * method (`getSupportedExtensions` / `getAttribLocation`) on a context that
+ * became `null` (context loss, blacklisted GPU, stripped WebView). These fire
+ * from Paper Shaders' async shader-mount callback, ESCAPE the `<ShaderSafe>`
+ * React error boundary, and reach Sentry/Better Stack as global errors. The
+ * method names are WebGL2 API — never called from first-party app code — so the
+ * message wording alone is specific enough; no chunk-frame anchor is needed.
+ * Matches all six wordings of this class: five JS-engine / DOM-binding variants
+ * — V8 (`Cannot read properties of null (reading '<m>')`), old JSC
+ * (`Cannot read property '<m>' of null`), SpiderMonkey/Firefox
+ * (`can't access property "<m>"<…>`), modern JSC (Safari / Chrome-on-iOS
+ * CriOS, which uses WebKit/JSC rather than V8:
+ * `null is not an object (evaluating 'this.gl.<m>')`), and Gecko/Firefox
+ * DOM-binding (`WebGL2RenderingContext.<m>: Argument 1 is not an object.` —
+ * Firefox's DOM bindings throw on the method call itself when the `this`
+ * binding is the null WebGL2 context) — PLUS the library's OWN internal guard
+ * wording (`this.gl is null` — the library's own explicit throw when its state
+ * check detects the null context, distinct from any JS-engine TypeError).
+ * Never page Better Stack for this class. See
+ * `PAPER_SHADER_NULL_CONTEXT_NOISE_PATTERNS` for the full rationale and the
+ * `supportsWebGL2()` probe in `shader-safe.tsx` for the primary guard.
+ */
+export function isPaperShaderNullContextNoise(message) {
+    const stripped = stripErrorWrappers(normalizeString(message));
+    return PAPER_SHADER_NULL_CONTEXT_NOISE_PATTERNS.some((pattern) => stripped.includes(pattern));
+}
+/**
+ * Whether a Sentry event is the Paper Shaders
+ * (`@paper-design/shaders-react`) WebGL-unsupported deliberate-throw noise
+ * class: the library's OWN canonical
+ * `Paper Shaders: WebGL is not supported in this browser` `Error`, thrown from
+ * the library's shader-mount constructor when WebGL is unavailable (a
+ * stripped-down/mobile WebView, a headless renderer, a browser with WebGL
+ * disabled, or a GPU blacklisted at context creation). This is a SIBLING of
+ * the null-context crash class (`isPaperShaderNullContextNoise`), but a
+ * DIFFERENT throw — a deliberate library `Error`, NOT a null-context
+ * `TypeError` from calling a WebGL2 method on a `null` context. The throw
+ * escapes `<ShaderSafe>` (it fires from the library constructor inside a
+ * dynamic import / `useEffect` that bypasses the React error boundary) and
+ * reaches Sentry as an uncaught global `onunhandledrejection` — an EXPECTED
+ * degradation state on WebGL-less browsers, never a product bug. The
+ * `supportsWebGL2()` probe in `shader-safe.tsx` is the primary guard that
+ * degrades to the fallback BEFORE the throw; this matcher is the leak-path
+ * backstop for the residual async throw that bypasses the one-shot probe.
+ *
+ * Requires the EXACT library message (case-sensitive; the `Paper Shaders:`
+ * prefix is the library's canonical marker, never emitted by first-party
+ * app code) AND a NEGATIVE guard: if ANY frame resolves to a de-minified
+ * first-party `apps/web/src/…` source path, the event keeps reporting (a
+ * real first-party `throw new Error('Paper Shaders: WebGL is not supported
+ * in this browser')` regression de-minifies to `apps/web/src/…` and must not
+ * be hidden). The production noise pattern carries only minified
+ * `@paper-design/shaders` chunk frames, so the negative guard does not fire
+ * for it. A frameless capture with this exact message still classifies as
+ * noise (the message alone is specific — the `Paper Shaders:` library prefix
+ * is part of the anchor, so a near-worded `WebGL is not supported in this
+ * browser` without the prefix does NOT match). See
+ * `PAPER_SHADER_WEBGL_UNSUPPORTED_NOISE_MESSAGE` for the full rationale.
+ */
+export function isPaperShaderWebGLUnsupportedNoise(input) {
+    const stripped = stripErrorWrappers(normalizeString(input.message));
+    if (stripped !== PAPER_SHADER_WEBGL_UNSUPPORTED_NOISE_MESSAGE) {
+        return false;
+    }
+    const frames = input.frames ?? [];
+    // Negative guard: a resolved first-party `apps/web/src/…` frame means our
+    // own code threw this exact message → a real first-party regression (even
+    // though the `Paper Shaders:` prefix is the library's canonical marker, a
+    // hostile/copy-pasted first-party throw could share it). Keep reporting so
+    // the call site can be found + fixed. Only the library throw (minified
+    // `@paper-design/shaders` chunk frames, or frameless) is dropped. No second
+    // "any resolvable frame" guard is needed — the message is specific enough
+    // (the library's canonical string) that a frameless capture is safe to
+    // drop, unlike the generic `undefined` / `OperationError` matchers.
+    if (frames.some((frame) => isFirstPartyResolvedSource(frame?.filename))) {
+        return false;
+    }
+    return true;
+}
+// Paper Shaders (`@paper-design/shaders-react`) image-uniform load-race throw —
+// a THIRD sibling of the two Paper Shaders classes above, and like the
+// WebGL-unsupported one it is the library's OWN deliberate `throw`, not a
+// JS-engine TypeError.
+//
+// Every Paper Shaders effect that dithers or grains its output uploads a small
+// bundled noise bitmap to the `u_noiseTexture` sampler uniform. The library
+// guards that upload with an `img.complete` assertion and throws
+// `Paper Shaders: image for uniform u_noiseTexture must be fully loaded` when
+// the shader mount reaches the upload before the browser finished decoding the
+// bitmap. That is a race inside the library between its own image load and its
+// own rAF-driven mount — first-party code neither supplies this image nor names
+// this uniform (`rg "u_noiseTexture|noiseTexture" apps/web/src` matches
+// nothing). It is transient: the same page renders the shader normally on the
+// next paint once the bitmap is decoded, which is why it appears as scattered
+// single occurrences across visitors rather than a deterministic route failure.
+// Slow connections, throttled background tabs and cold caches all widen the
+// window. Better Stack, Kortix Frontend prod: 10+ occurrences, `Error`, call
+// site `?` in chunk `app:///_next/static/immutable/chunks/2-qxa2k33wllw.js`
+// (Paper Shaders library) — no first-party frame.
+//
+// The throw escapes `<ShaderSafe>` for the same reason the null-context class
+// does: it fires inside an async mount callback, and `getDerivedStateFromError`
+// only catches render-phase throws. `supportsWebGL2()` cannot prevent it
+// either — WebGL2 IS supported here; only the bitmap is late.
+//
+// Same message-only contract as its siblings: the `Paper Shaders:` prefix is
+// the library's canonical marker, never emitted by first-party app code, and
+// the `u_noiseTexture` uniform name pins the single library call site. The
+// first-party negative guard is kept so a hypothetical first-party throw
+// carrying this exact string still reports. NOT in `ignoreErrors`: that gate
+// has no frame context, so the first-party negative guard could not run there.
+const PAPER_SHADER_IMAGE_UNIFORM_NOISE_MESSAGE = 'Paper Shaders: image for uniform u_noiseTexture must be fully loaded';
+/**
+ * Whether a Sentry / window.onerror event is the Paper Shaders
+ * (`@paper-design/shaders-react`) `u_noiseTexture` image-uniform load-race
+ * throw. See {@link PAPER_SHADER_IMAGE_UNIFORM_NOISE_MESSAGE} for the full
+ * rationale.
+ *
+ * Requires the EXACT library message AND a NEGATIVE guard: any frame that
+ * resolves to a de-minified first-party `apps/web/src/…` source keeps
+ * reporting. The production shape carries only minified library chunk frames,
+ * so the guard does not fire for it; a frameless capture with this exact
+ * message still classifies as noise.
+ */
+export function isPaperShaderImageUniformNoise(input) {
+    const stripped = stripErrorWrappers(normalizeString(input.message));
+    if (stripped !== PAPER_SHADER_IMAGE_UNIFORM_NOISE_MESSAGE) {
+        return false;
+    }
+    const frames = input.frames ?? [];
+    if (frames.some((frame) => isFirstPartyResolvedSource(frame?.filename))) {
+        return false;
+    }
+    return true;
+}
+/**
+ * Whether a Sentry / window.onerror event is the Canvas `getImageData`
+ * out-of-memory noise class: a `RangeError` from
+ * `CanvasRenderingContext2D.getImageData()` running out of memory allocating
+ * the `ImageData` buffer — the browser's canonical
+ * `Failed to execute 'getImageData' on 'CanvasRenderingContext2D': Out of
+ * memory at ImageData creation` message. This is TRANSIENT browser resource
+ * exhaustion (the canvas was too large / the tab was under memory pressure /
+ * the device is low-RAM), fired from a third-party canvas library's
+ * `addEventListener` callback (Sentry's `BrowserApiErrors` auto-wrapper captures
+ * it as UNCAUGHT, `handled:false` — never reached a React error boundary).
+ * NOT a deterministic code bug — the same canvas renders fine on the next
+ * visit once memory frees up. See `CANVAS_GETIMAGE_DATA_OOM_NOISE_PATTERNS`
+ * for the full rationale and Better Stack pattern `b4b43847…`.
+ *
+ * Requires the EXACT V8/Chrome message AND a NEGATIVE guard: if any frame (or
+ * the window.onerror `filename`) resolves to a de-minified first-party
+ * `apps/web/src/…` source path, the event keeps reporting — our own code is
+ * the `getImageData` caller and a real first-party OOM regression is
+ * actionable. Only events with NO resolved first-party frame (the prod noise
+ * shape: all minified third-party canvas library chunk frames, or frameless)
+ * are dropped. A frameless capture with this exact message still classifies
+ * as noise (the message alone is the browser's canonical OOM wording and is
+ * specific enough — the `CanvasRenderingContext2D` + `getImageData` +
+ * `ImageData creation` tokens together pin this single DOM API call site).
+ * `RangeError: ` / `Unhandled promise rejection: ` wrappers are stripped
+ * before matching so all capture paths (window.onerror, onunhandledrejection,
+ * Sentry exception) classify consistently.
+ */
+export function isCanvasImageDataOOMNoise(input) {
+    const stripped = stripErrorWrappers(normalizeString(input.message));
+    if (!CANVAS_GETIMAGE_DATA_OOM_NOISE_PATTERNS.some((re) => re.test(stripped))) {
+        return false;
+    }
+    const sources = sourcesOf(input);
+    // Negative guard: a resolved first-party `apps/web/src/…` frame (or
+    // window.onerror `filename`) means our own code is the `getImageData` caller
+    // → a real first-party OOM regression; keep reporting so the call site can
+    // be found + fixed. A real first-party `getImageData` OOM de-minifies to
+    // `apps/web/src/…` and is never hidden.
+    if (sources.some(isFirstPartyResolvedSource)) {
+        return false;
+    }
+    return true;
+}
+// Spoofed WebGL `getParameter` noise — a broken anti-fingerprinting browser
+// extension (the `spoof` function is the extension's own wrapper name; nothing
+// first-party names a function `spoof` — confirmed by a repo-wide source
+// search) replaces `WebGL2RenderingContext.prototype.getParameter` with a
+// wrapper that loses the native receiver, so EVERY real call throws
+// `TypeError: Illegal invocation` from the extension's own frame
+// (`WebGL2RenderingContext.spoof [as getParameter]`, script URL hidden as
+// `<anonymous>`). The wrapper is installed before any page script, so the
+// failure is invisible to the page: every WebGL consumer on the page fails the
+// same way.
+//
+// On the marketing homepage the consumer is the shaders library
+// (`shaders` / Paper Shaders family, the homepage wallpaper and shader
+// components): its `initialize()` builds a `WebGPURenderer` whose
+// `Renderer.init()` executor falls back from `WebGPUBackend` to
+// `WebGLBackend`, whose `init()` builds `new WebGLState`, whose constructor
+// calls `this.gl.getParameter(this.gl.MAX_TEXTURE_IMAGE_UNITS)` — the throw
+// site. The library catches BOTH init stages (it warns `[Shaders] WebGPU
+// initialization failed, falling back to WebGL:` and then errors `[Shaders]
+// Both WebGPU and WebGL initialization failed:` and returns gracefully — the
+// breadcrumbs prove both catches ran, twice, one per shader component). The
+// `shaders-react` `Shader` wrapper's own failure log (`[Shader] Initialization
+// failed:`) is ABSENT from the breadcrumbs and no app action follows the
+// failure — the visitor gets the page's non-shader content; the canvas stays
+// blank. Somewhere along the library's promise plumbing one of these
+// init-failure rejections still reached the page's global
+// `onunhandledrejection` handler with the init-time throw stack (the capture
+// records no async frames beyond the throw site, so the exact leaking promise
+// cannot be pinned from the telemetry). That does not change the
+// classification: the throw originates in the visitor's extension breaking a
+// page-owned native, not in a first-party defect. Our code never wraps a
+// WebGL2 prototype method (no assignment to `WebGL2RenderingContext.prototype`
+// in the repo — confirmed by source search) and never names a function
+// `spoof`.
+//
+// Better Stack pattern
+// 1c3e5d8c40c89ea1c21fc33d524f8043e5a0d21a1ea001539cd1441f1ea37179
+// (Kortix Frontend prod, application_id 2346967): `TypeError`, message
+// `Illegal invocation`, 2 occurrences / 0 identified users, first 2026-10-15
+// 19:17:00 UTC, release `ed03160b76fccf04fb68bce4083737939faed829`, call site
+// `WebGL2RenderingContext.spoof [as getParameter]` at `<anonymous>` (line 97),
+// frames `40hy2smvy-pfn.js` (`?` async executor line 10, `bB.init` line 183,
+// `new bb` line 183), request URL `https://kortix.com/` (marketing homepage),
+// browser Chrome 146 on Windows 10, mechanism
+// `auto.browser.global_handlers.onunhandledrejection` (UNCAUGHT,
+// `handled:false`). Reproduced locally (KRTX-2465): the same spoof wrapper in
+// a real headless Chromium + the real `shaders` core module produces the
+// identical `Illegal invocation` from `WebGL2RenderingContext.spoof [as
+// getParameter]` → `new WebGLState` → `WebGLBackend.init` → the
+// `Renderer.init()` executor.
+//
+// The anchors are BOTH required: (a) the stripped message is exactly
+// `Illegal invocation` — V8's canonical wording for a native method called
+// with an invalid receiver (`TypeError: ` and `Unhandled promise rejection: `
+// wrappers stripped first so all capture paths match), and (b) a stack frame
+// whose function name is the extension's spoof wrapper for this exact native
+// (`WebGL2RenderingContext.spoof [as getParameter]` — V8's display string for
+// a function named `spoof` assigned to `WebGL2RenderingContext.prototype.
+// getParameter`). A bare `Illegal invocation` without the spoof frame is a
+// GENERIC V8 wording (any native called with a wrong receiver — a real
+// first-party DOM-API misuse) and keeps reporting; a spoof frame with a
+// different message is a different failure and keeps reporting. The
+// first-party negative guard preserves any event whose stack resolves to a
+// de-minified `apps/web/src/…` source (our own code is in the stack → real
+// regression, find + fix it).
+//
+// Sentry-only (`appliesTo: 'sentry'`): the runtime gate's evidence carries no
+// stack frames (`runtimeNoiseEvidence` sets `frames: []`), so it cannot prove
+// the third-party spoof origin and must not drop a generic `Illegal
+// invocation`. The Sentry `beforeSend` gate — which carries the frame stack —
+// is the gate that drops the class before it reaches Better Stack.
+// Deliberately NOT added to `sentry.client.config.ts`'s `ignoreErrors` list —
+// that gate has no frame context, so a bare `Illegal invocation` match there
+// would swallow a real first-party DOM-API misuse; the frame-aware
+// `beforeSend` hook (which calls `shouldIgnoreSentryBrowserNoise`) is the
+// only safe gate.
+const SPOOFED_GET_PARAMETER_NOISE_MESSAGE = 'Illegal invocation';
+const SPOOFED_GET_PARAMETER_FRAME_MARKER = 'WebGL2RenderingContext.spoof [as getParameter]';
+/**
+ * Whether a Sentry event is the spoofed WebGL `getParameter` noise class: a
+ * `TypeError: Illegal invocation` thrown from an anti-fingerprinting
+ * extension's own `WebGL2RenderingContext.spoof [as getParameter]` wrapper
+ * frame (the wrapper loses the native receiver, so every real WebGL call
+ * fails). See `SPOOFED_GET_PARAMETER_NOISE_MESSAGE` for the full rationale
+ * and Better Stack pattern `1c3e5d8c…`.
+ *
+ * Requires the EXACT stripped message AND the spoof wrapper frame AND the
+ * NEGATIVE guard: any frame that resolves to a de-minified first-party
+ * `apps/web/src/…` source keeps reporting. A frameless capture or a stack
+ * without the spoof frame keeps reporting (the message alone is a generic V8
+ * wording, not proof of third-party origin).
+ */
+export function isSpoofedWebGLGetParameterNoise(input) {
+    const stripped = stripErrorWrappers(normalizeString(input.message));
+    if (stripped !== SPOOFED_GET_PARAMETER_NOISE_MESSAGE) {
+        return false;
+    }
+    const frames = input.frames ?? [];
+    if (!frames.some((frame) => normalizeString(frame?.function).includes(SPOOFED_GET_PARAMETER_FRAME_MARKER))) {
+        return false;
+    }
+    if (frames.some((frame) => isFirstPartyResolvedSource(frame?.filename))) {
+        return false;
+    }
+    return true;
+}
+export const GRAPHICS_RULES = [
+    {
+        id: 'paper-shaders-null-context',
+        appliesTo: 'both',
+        match: ({ message }) => isPaperShaderNullContextNoise(message),
+    },
+    {
+        id: 'paper-shaders-webgl-unsupported',
+        appliesTo: 'sentry',
+        match: isPaperShaderWebGLUnsupportedNoise,
+    },
+    { id: 'paper-shaders-image-uniform', appliesTo: 'sentry', match: isPaperShaderImageUniformNoise },
+    { id: 'canvas-image-data-oom', appliesTo: 'both', match: isCanvasImageDataOOMNoise },
+    { id: 'spoofed-webgl-get-parameter', appliesTo: 'sentry', match: isSpoofedWebGLGetParameterNoise },
+];