@@ -247,7 +247,7 @@ export default function PdfViewer(props: { src: string; title: string }) {
             throw new Error(
               "This PDF has too much text to search. Save it to search in another app.",
             );
-          if (text.toLowerCase().includes(query.toLowerCase())) found.push(index);
+          if (pdfFindMatchRuns([text], query)[0]) found.push(index);
         }
         if (cancelled) return;
         setRatios(sizes);
 
@@ -105,6 +105,20 @@ describe("pdfFindMatchRuns", () => {
     ]);
   });
 
+  it("matches a space in the query across a line end", () => {
+    expect(
+      pdfFindMatchRuns(["Q3 quarterly", "\n", "outlook", " ahead"], "quarterly outlook"),
+    ).toEqual([true, true, true, false]);
+  });
+
+  it("matches a word hyphenated across a line end", () => {
+    expect(pdfFindMatchRuns(["the quar-", "\n", "terly report"], "quarterly")).toEqual([
+      true,
+      false,
+      true,
+    ]);
+  });
+
   it("highlights nothing for a blank or missing query", () => {
     expect(pdfFindMatchRuns(["a b", "c"], " ")).toEqual([false, false]);
     expect(pdfFindMatchRuns(["a b", "c"], "")).toEqual([false, false]);
 
@@ -56,18 +56,50 @@ export function pdfPageText(
   return items.map((item) => ("str" in item ? item.str + (item.hasEOL ? "\n" : "") : "")).join("");
 }
 
+/**
+ * Lowercased search text with each whitespace run, line ends included, read as one space, and a
+ * hyphen that ends a line between letters dropped, as pdf.js's own find does. `owner` maps each
+ * character back to the run it came from.
+ */
+function pdfFindText(runs: readonly string[]) {
+  const chars = runs.flatMap((run, index) =>
+    Array.from(run.toLowerCase(), (char) => ({ char, run: index })),
+  );
+  let text = "";
+  const owner: number[] = [];
+  for (let index = 0; index < chars.length; index++) {
+    const { char, run } = chars[index]!;
+    if (/\s/.test(char)) {
+      if (!text.endsWith(" ")) {
+        text += " ";
+        owner.push(run);
+      }
+    } else if (
+      char === "-" &&
+      chars[index + 1]?.char === "\n" &&
+      /\p{L}/u.test(text.at(-1) ?? "") &&
+      /\p{L}/u.test(chars[index + 2]?.char ?? "")
+    ) {
+      index++;
+    } else {
+      text += char;
+      for (let unit = 0; unit < char.length; unit++) owner.push(run);
+    }
+  }
+  return { text, owner };
+}
+
 /** Which text runs hold part of a case-insensitive match, when pdf.js splits words across spans. */
 export function pdfFindMatchRuns(runs: readonly string[], query: string) {
-  const texts = runs.map((run) => run.toLowerCase());
-  const text = texts.join("");
-  const needle = query.toLowerCase();
-  let match = needle.trim() ? text.indexOf(needle) : -1;
-  let offset = 0;
-  return texts.map((run) => {
-    while (match !== -1 && match + needle.length <= offset)
-      match = text.indexOf(needle, match + needle.length);
-    const hit = match !== -1 && match < offset + run.length;
-    offset += run.length;
-    return hit;
-  });
+  const { text, owner } = pdfFindText(runs);
+  const needle = query.toLowerCase().replace(/\s+/g, " ");
+  const hits = runs.map(() => false);
+  if (!needle.trim()) return hits;
+  for (
+    let match = text.indexOf(needle);
+    match !== -1;
+    match = text.indexOf(needle, match + needle.length)
+  )
+    for (let index = match; index < match + needle.length; index++) hits[owner[index]!] = true;
+  return hits;
 }
Original file line number	Diff line number	Diff line change
`@@ -247,7 +247,7 @@ export default function PdfViewer(props: { src: string; title: string }) {`
`247`	`247`	`throw new Error(`
`248`	`248`	`"This PDF has too much text to search. Save it to search in another app.",`
`249`	`249`	`);`
`250`		`- if (text.toLowerCase().includes(query.toLowerCase())) found.push(index);`
	`250`	`+ if (pdfFindMatchRuns([text], query)[0]) found.push(index);`
`251`	`251`	`}`
`252`	`252`	`if (cancelled) return;`
`253`	`253`	`setRatios(sizes);`