diff --git a/docs/src/components/docs-search/search-logic.test.ts b/docs/src/components/docs-search/search-logic.test.ts
index a33a96be..bbfd8907 100644
--- a/docs/src/components/docs-search/search-logic.test.ts
+++ b/docs/src/components/docs-search/search-logic.test.ts
@@ -314,6 +314,12 @@ describe("snippet", () => {
     expect(result.endsWith("\u2026")).toBe(true);
   });
 
+  it("centers on the match after text whose lowercase is longer, such as a dotted capital I", () => {
+    // "İ".toLowerCase() is two characters: an offset found in a lowercased copy lands past the match.
+    const text = "İ".repeat(100) + " match " + "b".repeat(40);
+    expect(snippet(text, "match", tokens("match"))).toContain("match");
+  });
+
   it("keeps output to roughly 120 characters of text", () => {
     const text = "word ".repeat(100);
     const result = snippet(text, "word", tokens("word"));
diff --git a/docs/src/components/docs-search/search-logic.ts b/docs/src/components/docs-search/search-logic.ts
index 2961bdad..4e3013c2 100644
--- a/docs/src/components/docs-search/search-logic.ts
+++ b/docs/src/components/docs-search/search-logic.ts
@@ -171,6 +171,20 @@ export function score(e: SearchEntry, q: string, toks: string[]): number {
   return 0;
 }
 
+/**
+ * Where offset `idx` of `lower` (`text.toLowerCase()`) falls in `text`. Lowercasing can lengthen a string ("İ"
+ * becomes two characters), so past such a character the two offsets differ.
+ */
+function indexInOriginal(text: string, lower: string, idx: number): number {
+  if (idx <= 0 || lower.length === text.length) return idx;
+  var lowered = 0;
+  for (var i = 0; i < text.length; i++) {
+    if (lowered >= idx) return i;
+    lowered += text[i].toLowerCase().length;
+  }
+  return text.length;
+}
+
 /** Returns ~120 chars centered on the first query/token match in text. */
 export function snippet(
   text: string | null | undefined,
@@ -187,6 +201,7 @@ export function snippet(
       if (idx >= 0) break;
     }
   }
+  idx = indexInOriginal(text, lo, idx);
   var start = Math.max(0, idx > 40 ? idx - 40 : 0);
   return (
     (start > 0 ? "\u2026" : "") +
diff --git a/lsp/src/preserved-targets.ts b/lsp/src/preserved-targets.ts
index c804d374..bb995e22 100644
--- a/lsp/src/preserved-targets.ts
+++ b/lsp/src/preserved-targets.ts
@@ -94,6 +94,28 @@ const directiveBits = (value: string | undefined): number => {
 };
 
 /** The offset just past the `>` closing a start tag that begins at `from`, honoring quotes; -1 when unclosed. */
+/**
+ * The index of `</name` at or after `from`, in any ASCII case, or -1. `name` is lowercase letters. Tag names are
+ * compared in place, as an HTML parser compares them: a lowercased copy of the markup can be longer than it ("İ"
+ * becomes two characters), so an index found there points past the real end tag.
+ */
+const findEndTag = (markup: string, name: string, from: number): number => {
+  for (
+    let at = markup.indexOf("</", from);
+    at !== -1;
+    at = markup.indexOf("</", at + 2)
+  ) {
+    let k = 0;
+    while (
+      k < name.length &&
+      (markup.charCodeAt(at + 2 + k) | 0x20) === name.charCodeAt(k)
+    )
+      k++;
+    if (k === name.length) return at;
+  }
+  return -1;
+};
+
 const startTagEnd = (markup: string, from: number): number => {
   let quote = "";
   for (let i = from; i < markup.length; i++) {
@@ -118,7 +140,6 @@ export const collectPreservedTargets = (
   preserve: readonly string[],
 ): PreservedTargets => {
   const matchers = preserveMatchers(preserve);
-  const lower = markup.toLowerCase();
   const stack: { name: string; bits: number }[] = [];
   const preserved = {
     classes: new Set<string>(),
@@ -196,7 +217,7 @@ export const collectPreservedTargets = (
     classTarget.tags.add(name);
 
     if (RAW_TEXT_ELEMENTS.has(name)) {
-      const rawEnd = lower.indexOf(`</${name}`, end);
+      const rawEnd = findEndTag(markup, name, end);
       i = rawEnd === -1 ? markup.length : rawEnd;
       continue;
     }
diff --git a/pkg/src/lib/generated-id-classes.test.ts b/pkg/src/lib/generated-id-classes.test.ts
index c3893813..b2a2337b 100644
--- a/pkg/src/lib/generated-id-classes.test.ts
+++ b/pkg/src/lib/generated-id-classes.test.ts
@@ -157,6 +157,24 @@ describe("generated #id class collisions", () => {
     expect(error.message.match(/^error: /gm)).toHaveLength(2);
   });
 
+  it("finds a <style> rule after text whose lowercase is longer, such as a dotted capital I", async () => {
+    // "İ".toLowerCase() is two characters: an offset found in a lowercased copy of the markup lands past the rule.
+    write(
+      "src/components/site-nav_x.html",
+      '<nav><button id="btn">Menu</button></nav>\n',
+    );
+    write("src/components/site-nav_x.css", "#btn { color: red; }\n");
+    write(
+      "src/components/site-nav.html",
+      `<nav>\n  <a id="x_btn">${"İ".repeat(40)}</a>\n</nav>\n<style>\n  #x_btn { color: blue; }\n</style>\n`,
+    );
+
+    const error = await collisionOf();
+    expect(error.message).toContain(
+      "src/components/site-nav.html:5:3  #x_btn in <site-nav>",
+    );
+  });
+
   it("reads CRLF line endings as one line each", async () => {
     write("src/components/one_two.html", '<p id="three">a</p>');
     write(
diff --git a/pkg/src/lib/generated-id-classes.ts b/pkg/src/lib/generated-id-classes.ts
index 8abb33c7..233082aa 100644
--- a/pkg/src/lib/generated-id-classes.ts
+++ b/pkg/src/lib/generated-id-classes.ts
@@ -133,16 +133,42 @@ const lineAndColumn = (
   return { line, column: offset - lineStart + 1 };
 };
 
+/**
+ * The index of `tag` (`<` and lowercase ASCII, such as `</style`) at or after `from` in `html`, in any ASCII case,
+ * or -1. Compared in place: a lowercased copy of `html` can be longer than it ("İ" becomes two characters), so an
+ * index found there points past the text it names.
+ */
+const indexOfTag = (html: string, tag: string, from: number): number => {
+  for (
+    let at = html.indexOf("<", from);
+    at !== -1;
+    at = html.indexOf("<", at + 1)
+  ) {
+    let k = 1;
+    for (; k < tag.length; k++) {
+      const code = html.charCodeAt(at + k);
+      const wanted = tag.charCodeAt(k);
+      // A lowercase letter also matches its uppercase; anything else only itself.
+      if (
+        code !== wanted &&
+        !(wanted >= 97 && wanted <= 122 && code === wanted - 32)
+      )
+        break;
+    }
+    if (k === tag.length) return at;
+  }
+  return -1;
+};
+
 /** The offset of the first `#id` rule naming `idName` inside a `<style>` element of `html`, or -1. */
 const styleElementOffset = (html: string, idName: string): number => {
-  const lower = html.toLowerCase();
   let from = 0;
   for (;;) {
-    const open = lower.indexOf("<style", from);
+    const open = indexOfTag(html, "<style", from);
     if (open === -1) return -1;
-    const contentStart = lower.indexOf(">", open);
+    const contentStart = html.indexOf(">", open);
     if (contentStart === -1) return -1;
-    const close = lower.indexOf("</style", contentStart);
+    const close = indexOfTag(html, "</style", contentStart);
     const contentEnd = close === -1 ? html.length : close;
     const offset = cssIdSelectorOffset(
       html.slice(contentStart + 1, contentEnd),
diff --git a/pkg/src/lib/preserved-targets.test.ts b/pkg/src/lib/preserved-targets.test.ts
index 5e396ea6..8e714afc 100644
--- a/pkg/src/lib/preserved-targets.test.ts
+++ b/pkg/src/lib/preserved-targets.test.ts
@@ -271,6 +271,27 @@ document.querySelector(".inline");`;
   });
 });
 
+describe("collectPreservedTargets: raw-text elements", () => {
+  it("finds the end of a raw-text element in any ASCII case", () => {
+    const targets = collectPreservedTargets(
+      '<TITLE>x</Title ><p class="kept">a</p><STYLE>p{}</STYLE><p class="also">b</p>',
+      ["p"],
+    );
+    expect([...targets.classes]).toEqual(["kept", "also"]);
+  });
+
+  it("reads past a raw-text element whose lowercase is longer, such as one holding a dotted capital I", () => {
+    // "İ".toLowerCase() is two characters: an end found in a lowercased copy of the markup lands past the next tag.
+    for (const element of ["title", "style", "textarea", "script"]) {
+      const targets = collectPreservedTargets(
+        `<${element}>${"İ".repeat(12)}</${element}><p class="kept">a</p>`,
+        ["p"],
+      );
+      expect([...targets.classes], element).toEqual(["kept"]);
+    }
+  });
+});
+
 describe("preserved-targets copies", () => {
   it("keeps the language server's copy byte-identical to the compiler's", () => {
     const pkgCopy = readFileSync(
diff --git a/pkg/src/lib/preserved-targets.ts b/pkg/src/lib/preserved-targets.ts
index c804d374..bb995e22 100644
--- a/pkg/src/lib/preserved-targets.ts
+++ b/pkg/src/lib/preserved-targets.ts
@@ -94,6 +94,28 @@ const directiveBits = (value: string | undefined): number => {
 };
 
 /** The offset just past the `>` closing a start tag that begins at `from`, honoring quotes; -1 when unclosed. */
+/**
+ * The index of `</name` at or after `from`, in any ASCII case, or -1. `name` is lowercase letters. Tag names are
+ * compared in place, as an HTML parser compares them: a lowercased copy of the markup can be longer than it ("İ"
+ * becomes two characters), so an index found there points past the real end tag.
+ */
+const findEndTag = (markup: string, name: string, from: number): number => {
+  for (
+    let at = markup.indexOf("</", from);
+    at !== -1;
+    at = markup.indexOf("</", at + 2)
+  ) {
+    let k = 0;
+    while (
+      k < name.length &&
+      (markup.charCodeAt(at + 2 + k) | 0x20) === name.charCodeAt(k)
+    )
+      k++;
+    if (k === name.length) return at;
+  }
+  return -1;
+};
+
 const startTagEnd = (markup: string, from: number): number => {
   let quote = "";
   for (let i = from; i < markup.length; i++) {
@@ -118,7 +140,6 @@ export const collectPreservedTargets = (
   preserve: readonly string[],
 ): PreservedTargets => {
   const matchers = preserveMatchers(preserve);
-  const lower = markup.toLowerCase();
   const stack: { name: string; bits: number }[] = [];
   const preserved = {
     classes: new Set<string>(),
@@ -196,7 +217,7 @@ export const collectPreservedTargets = (
     classTarget.tags.add(name);
 
     if (RAW_TEXT_ELEMENTS.has(name)) {
-      const rawEnd = lower.indexOf(`</${name}`, end);
+      const rawEnd = findEndTag(markup, name, end);
       i = rawEnd === -1 ? markup.length : rawEnd;
       continue;
     }
diff --git a/skills-tests/tests/skill-scripts.test.ts b/skills-tests/tests/skill-scripts.test.ts
index 8dd7a6a6..8668d6ed 100644
--- a/skills-tests/tests/skill-scripts.test.ts
+++ b/skills-tests/tests/skill-scripts.test.ts
@@ -548,6 +548,15 @@ describe("audit-seo", () => {
     expect(facts.imagesWithoutAlt).toBe(0);
   });
 
+  it("reads past a script or style whose lowercase is longer, such as one holding a dotted capital I", () => {
+    // "İ".toLowerCase() is two characters: an end found in a lowercased copy of the page lands past the next tag.
+    const facts = inspectPage(
+      `<body><script>${"İ".repeat(12)}</script><h1>real</h1><style>${"İ".repeat(12)}</style><img src=x></body>`,
+    );
+    expect(facts.h1Count).toBe(1);
+    expect(facts.imagesWithoutAlt).toBe(1);
+  });
+
   it("does not form new HTML from either side of a script or style", () => {
     const facts = inspectPage(
       '<body><h<script>inert</script>1>fake</h1><SCRIPT>"<h1>ignored</h1>"</script\t\n data-test><h1>real</h1></body>',
@@ -822,6 +831,17 @@ describe("check-links", () => {
     ).toEqual(["/a.png", "/a.png", "/a@2x.png"]);
   });
 
+  it("reads past a masked element whose lowercase is longer, such as one holding a dotted capital I", () => {
+    // "İ".toLowerCase() is two characters: an end found in a lowercased copy of the page lands past the next tag.
+    const links = extractLinks(
+      `<pre>${"İ".repeat(12)}</pre><a href="/after-pre">a</a><script>${"İ".repeat(12)}</script><a href="/after-script">b</a>`,
+    );
+    expect(links.map((link: { url: string }) => link.url)).toEqual([
+      "/after-pre",
+      "/after-script",
+    ]);
+  });
+
   it("decodes entities once and ignores links in masked elements", () => {
     const links = extractLinks(
       '<!-- <a href="/comment"> --><SCRIPT>"<a href=\'/script\'>"</SCRIPT><pre><a href="/pre">x</a></pre><a href="/a&amp;lt;b&amp;gt;">real</a>',
diff --git a/skills/bascik-seo/scripts/audit-seo.mjs b/skills/bascik-seo/scripts/audit-seo.mjs
index 5f93fa59..ea0fa2d2 100644
--- a/skills/bascik-seo/scripts/audit-seo.mjs
+++ b/skills/bascik-seo/scripts/audit-seo.mjs
@@ -88,7 +88,6 @@ export function inspectPage(html) {
     metas.find((m) => (m[key] ?? "").toLowerCase() === value)?.content;
   // Preserve boundaries between visible regions, rather than joining text on
   // either side of a removed element into a new tag on another pass.
-  const lower = html.toLowerCase();
   const visible = [];
   const opening = /<(script|style)\b[^>]*>/gi;
   let from = 0;
@@ -97,9 +96,11 @@ export function inspectPage(html) {
     if (match.index < from) continue;
     visible.push(html.slice(from, match.index));
     const name = match[1].toLowerCase();
-    let close = lower.indexOf(`</${name}`, opening.lastIndex);
-    while (close >= 0 && !/[\s/>]/.test(lower[close + name.length + 2] ?? ""))
-      close = lower.indexOf(`</${name}`, close + name.length + 2);
+    // Found in the page itself, in ASCII case: a lowercased copy can be longer than
+    // the page ("İ" becomes two characters), so its offsets point past the tag.
+    const endTag = new RegExp(`</${name}(?=[\\s/>])`, "gi");
+    endTag.lastIndex = opening.lastIndex;
+    const close = endTag.exec(html)?.index ?? -1;
     const end = close < 0 ? -1 : html.indexOf(">", close + name.length + 2);
     from = end < 0 ? html.length : end + 1;
     opening.lastIndex = from;
diff --git a/skills/bascik/scripts/check-links.mjs b/skills/bascik/scripts/check-links.mjs
index 4ca4df90..3951f8d9 100644
--- a/skills/bascik/scripts/check-links.mjs
+++ b/skills/bascik/scripts/check-links.mjs
@@ -62,7 +62,6 @@ export function extractLinks(html) {
   };
   // Scan disjoint visible regions. Removing markup and joining the leftovers can
   // turn <scr<script>ipt> into a new <script> on the next pass.
-  const lower = html.toLowerCase();
   let from = 0;
   const ignored = /<!--|<(script|template|textarea|code|pre)\b[^>]*>/gi;
   let match;
@@ -76,9 +75,11 @@ export function extractLinks(html) {
       const name = match[1].toLowerCase();
       // External scripts have a real src link; their contents are still inert.
       if (name === "script" && /\bsrc\s*=/i.test(match[0])) scan(match[0]);
-      let close = lower.indexOf(`</${name}`, ignored.lastIndex);
-      while (close >= 0 && !/[\s/>]/.test(lower[close + name.length + 2] ?? ""))
-        close = lower.indexOf(`</${name}`, close + name.length + 2);
+      // Found in the page itself, in ASCII case: a lowercased copy can be longer than
+      // the page ("İ" becomes two characters), so its offsets point past the tag.
+      const endTag = new RegExp(`</${name}(?=[\\s/>])`, "gi");
+      endTag.lastIndex = ignored.lastIndex;
+      const close = endTag.exec(html)?.index ?? -1;
       const end = close < 0 ? -1 : html.indexOf(">", close + name.length + 2);
       from = end < 0 ? html.length : end + 1;
     }