@@ -1184,11 +1184,6 @@
 			"count": 1
 		}
 	},
-	"integrations/misc/__tests__/extract-text.spec.ts": {
-		"@typescript-eslint/no-explicit-any": {
-			"count": 4
-		}
-	},
 	"integrations/misc/__tests__/line-counter.spec.ts": {
 		"@typescript-eslint/no-explicit-any": {
 			"count": 2
 
@@ -0,0 +1,34 @@
+import fs from "fs/promises"
+import path from "path"
+import { extractTextFromBuffer } from "../extract-text"
+
+// Keep real parser fixtures separate from the mocked parser unit suite.
+describe("real document buffer decoding", () => {
+	afterEach(() => vi.restoreAllMocks())
+
+	it("extracts numbered PDF text from fixture bytes without reopening the metadata pathname", async () => {
+		const source = await fs.readFile(path.join(__dirname, "fixtures", "approved.pdf"))
+		const access = vi.spyOn(fs, "access").mockRejectedValue(new Error("Unexpected pathname access"))
+		const readFile = vi.spyOn(fs, "readFile").mockRejectedValue(new Error("Unexpected pathname read"))
+
+		// pdf-parse prefixes each page with two newlines, which retain their line numbers.
+		expect(await extractTextFromBuffer(source, "metadata-only.PDF")).toBe(
+			"1 | \n2 | \n3 | Approved PDF first line\n4 | Approved PDF second line\n",
+		)
+		expect(access).not.toHaveBeenCalled()
+		expect(readFile).not.toHaveBeenCalled()
+	})
+
+	it("extracts numbered DOCX paragraphs from fixture bytes without reopening the metadata pathname", async () => {
+		const source = await fs.readFile(path.join(__dirname, "fixtures", "approved.docx"))
+		const access = vi.spyOn(fs, "access").mockRejectedValue(new Error("Unexpected pathname access"))
+		const readFile = vi.spyOn(fs, "readFile").mockRejectedValue(new Error("Unexpected pathname read"))
+
+		// Mammoth separates paragraphs with two newlines and decodes XML entities across runs.
+		expect(await extractTextFromBuffer(source, "metadata-only.DOCX")).toBe(
+			"1 | Approved DOCX first paragraph\n2 | \n3 | Approved DOCX second & final paragraph\n4 | \n",
+		)
+		expect(access).not.toHaveBeenCalled()
+		expect(readFile).not.toHaveBeenCalled()
+	})
+})
@@ -1,7 +1,47 @@
 import ExcelJS from "exceljs"
-import { formatWorkbook } from "../extract-text-from-xlsx"
+import { extractTextFromXLSX, formatWorkbook } from "../extract-text-from-xlsx"
 import { extractTextFromFile } from "../extract-text"
 
+describe("XLSX buffer decoding", () => {
+	it("preserves empty columns in a populated row from real XLSX bytes", async () => {
+		const workbook = new ExcelJS.Workbook()
+		const worksheet = workbook.addWorksheet("Sparse")
+		worksheet.getCell("A1").value = null
+		worksheet.getCell("B1").value = "middle"
+		worksheet.getCell("C1").value = undefined
+		worksheet.getCell("D1").value = "end"
+
+		const expected = "--- Sheet: Sparse ---\n\tmiddle\t\tend"
+		expect(formatWorkbook(workbook)).toBe(expected)
+		const source = Buffer.from(await workbook.xlsx.writeBuffer())
+		expect(await extractTextFromXLSX(source)).toBe(expected)
+	})
+
+	it("decodes only the selected buffer view when another workbook follows in the backing buffer", async () => {
+		const approved = new ExcelJS.Workbook()
+		const worksheet = approved.addWorksheet("Selected")
+		worksheet.getCell("A1").value = "approved cell"
+		worksheet.getCell("B1").value = 42
+		const approvedBytes = Buffer.from(await approved.xlsx.writeBuffer())
+
+		const unapproved = new ExcelJS.Workbook()
+		unapproved.addWorksheet("Unapproved").getCell("A1").value = "outside selected bytes"
+		const unapprovedBytes = Buffer.from(await unapproved.xlsx.writeBuffer())
+		const prefix = Buffer.from("unapproved prefix")
+		const backing = Buffer.concat([prefix, approvedBytes, unapprovedBytes])
+		const source = backing.subarray(prefix.length, prefix.length + approvedBytes.length)
+
+		// Passing source.buffer would expose the later, unapproved ZIP archive to the decoder.
+		expect(source.byteOffset).toBeGreaterThan(0)
+		expect(source.byteLength).toBeLessThan(source.buffer.byteLength)
+		expect(await extractTextFromXLSX(source)).toBe("--- Sheet: Selected ---\napproved cell\t42")
+	})
+
+	it("rejects invalid workbook bytes instead of returning empty workbook text", async () => {
+		await expect(extractTextFromXLSX(Buffer.from("not an XLSX archive"))).rejects.toThrow()
+	})
+})
+
 describe("formatWorkbook", () => {
 	describe("basic functionality", () => {
 		it("should extract text with proper formatting", () => {
@@ -177,6 +217,19 @@ describe("formatWorkbook", () => {
 	})
 
 	describe("edge cases", () => {
+		it("includes row 50000, marks row 50001 as truncated, and still formats the next sheet", () => {
+			const workbook = new ExcelJS.Workbook()
+			const worksheet = workbook.addWorksheet("Large")
+			worksheet.getCell("A50000").value = "last included row"
+			worksheet.getCell("A50001").value = "excluded row content"
+			workbook.addWorksheet("Next").getCell("A1").value = "next sheet content"
+
+			expect(formatWorkbook(workbook)).toBe(
+				"--- Sheet: Large ---\nlast included row\n[... truncated at row 50001 ...]\n\n" +
+					"--- Sheet: Next ---\nnext sheet content",
+			)
+		})
+
 		it("should handle empty workbook", () => {
 			const workbook = new ExcelJS.Workbook()
 			workbook.addWorksheet("Empty Sheet")