productdevbook · Oct 26, 2026 · Oct 20, 2026 · Oct 20, 2026 · Oct 26, 2026 · Oct 26, 2026
diff --git a/README.md b/README.md
@@ -209,6 +209,17 @@ Anything a cell carries works there — `style`, `formula`, `richText`,
 `hyperlink`, `checkbox`. `cells` still takes a `"row,col"` map, and wins
 where both describe the same position.
 
+A whole row or column can carry a format of its own. `rowDefs` takes a
+`style` beside `height` and the rest — written as Excel writes a
+formatted row, `<row s="N" customFormat="1">` — and `columns[i].style`
+does the same for a column. Either one is also stamped onto the cells in
+it that have no style of their own; where a row and a column both cover
+such a cell, the row's wins.
+
+```ts
+rowDefs: new Map([[9, { height: 20, style: { font: { bold: true }, numFmt: "#,##0.00" } }]])
+```
+
 Features: cell styles, auto column widths, merged cells, freeze/split panes, auto-filter (with per-column value filters — `<filters><filter val="…"/></filters>`; custom/dynamic/colour criteria are not emitted), data validation, hyperlinks, images (PNG/JPEG/GIF/SVG/WebP), comments, tables, conditional formatting (all 15 rule types, with their dxf styles), named ranges, print settings, page breaks, sheet protection, workbook protection, rich text, shared/array/dynamic formulas, sparklines, textboxes, background images, number formats, hidden sheets, Excel 2024 native checkboxes, HTML/Markdown/JSON/TSV export, template engine.
 
 ### Auto Column Width
@@ -756,19 +767,19 @@ The ODS reader and writer model the same narrow set, so **ODS → ODS is
 lossless**. The loss shows up when converting _into_ ODS from a format
 that models more, which in practice means XLSX.
 
-|                                                            | ODS                                                                                         |
-| ---------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
-| Cell values — string, number, boolean, date                | yes                                                                                         |
-| Formulas, including the `of:=` translation                 | yes                                                                                         |
-| Merged cells                                               | yes                                                                                         |
-| Cell styles                                                | **six facets only**: bold, italic, font size, font colour, background colour, number format |
-| Document properties                                        | six fields: title, subject, creator, description, keywords, created                         |
-| Borders, alignment, font name, underline, strikethrough    | no                                                                                          |
-| Column widths, row heights, hidden rows and columns        | no                                                                                          |
-| Freeze and split panes, sheet views, tab colour            | no                                                                                          |
-| Data validation, conditional formatting, auto-filter       | no                                                                                          |
-| Named ranges, tables, images, page setup, sheet protection | no                                                                                          |
-| Hidden sheets                                              | no                                                                                          |
+|                                                                  | ODS                                                                                         |
+| ---------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
+| Cell values — string, number, boolean, date                      | yes                                                                                         |
+| Formulas, including the `of:=` translation                       | yes                                                                                         |
+| Merged cells                                                     | yes                                                                                         |
+| Cell styles                                                      | **six facets only**: bold, italic, font size, font colour, background colour, number format |
+| Document properties                                              | six fields: title, subject, creator, description, keywords, created                         |
+| Borders, alignment, font name, underline, strikethrough          | no                                                                                          |
+| Column widths, row heights, row formats, hidden rows and columns | no                                                                                          |
+| Freeze and split panes, sheet views, tab colour                  | no                                                                                          |
+| Data validation, conditional formatting, auto-filter             | no                                                                                          |
+| Named ranges, tables, images, page setup, sheet protection       | no                                                                                          |
+| Hidden sheets                                                    | no                                                                                          |
 
 So `readXlsx` → `writeOds` keeps values, formulas, merges and those six
 style facets, and drops the rest **silently** — there is no warning,

diff --git a/docs/PARITY.md b/docs/PARITY.md
@@ -85,10 +85,10 @@ printing, copies, DPI and printer defaults; the exception is `r:id`,
 which points at a binary printer-settings part and has no portable
 meaning — headers and footers, sheet views
 and tab colours, hidden and very-hidden sheets, tables, row and column
-definitions, manual page breaks, outline properties, sparklines, text
-boxes, background images, images, document properties, named ranges, the
-1904 date system, workbook protection, theme colours, Excel 2024
-checkboxes, and encryption.
+definitions (a row or column format included), manual page breaks,
+outline properties, sparklines, text boxes, background images, images,
+document properties, named ranges, the 1904 date system, workbook
+protection, theme colours, Excel 2024 checkboxes, and encryption.
 
 ### Formula results are write-only in practice
 
@@ -250,6 +250,17 @@ Only the **trailing** box shrinks. Interior positions are untouched: a
 value at D1 still sits at index 3 behind three nulls, and an empty row
 between two populated ones is still there.
 
+### A row format and the cells already in it
+
+A row's format (`<row s customFormat>`) is the format of the cells that
+are not in the row (ECMA-376 Part 1 §18.3.1.73). A cell that is present
+keeps its own `s`, and that includes `s="0"` or no `s` at all. With
+`readStyles`, such a cell inside a formatted row reads back with
+`style: {}`. That says the file chose the default format, and the writer
+keeps it at `s="0"` instead of giving it the row's format, as it does
+for a cell the caller left without a style. Without the `{}`, read →
+write moved the row's format onto it. Found in review of #612.
+
 ### `Sheet.rows` is a dense rectangle
 
 Every row is an array, every row is the same length, and no element is
@@ -603,11 +614,11 @@ document properties (six fields), and six style facets — bold, italic,
 font size, font colour, background colour, number format.
 
 Not modelled in **either** direction: borders, alignment, font name,
-underline, strikethrough, column widths, row heights, hidden rows and
-columns, freeze and split panes, data validation, conditional formatting,
-auto-filter, tables, images, page setup, sheet protection, tab colour,
-hidden sheets, and `time` cells (which read back as the raw ISO duration
-string).
+underline, strikethrough, column widths, row heights, row formats, hidden
+rows and columns, freeze and split panes, data validation, conditional
+formatting, auto-filter, tables, images, page setup, sheet protection,
+tab colour, hidden sheets, and `time` cells (which read back as the raw
+ISO duration string).
 
 Named ranges left that list: `<table:named-expressions>` carries them
 both ways. Only workbook-level ones — ODF scopes a name to a sheet by

diff --git a/docs/SPEC-COVERAGE.md b/docs/SPEC-COVERAGE.md
@@ -10,7 +10,7 @@ Corpus under `test/fixtures`: 35 OOXML workbooks (455 element names, 245 attribu
 
 ## SpreadsheetML (ECMA-376 Part 1)
 
-360 complex types. **31** names appear in the corpus and are _nowhere_ in `src/`; 6 appear in the corpus and occur in `src/` only inside a longer string, which usually means a writer emits them in a template; 644 the source switches on directly; 494 it knows but the corpus does not use; 917 are in neither.
+360 complex types. **30** names appear in the corpus and are _nowhere_ in `src/`; 6 appear in the corpus and occur in `src/` only inside a longer string, which usually means a writer emits them in a template; 645 the source switches on directly; 494 it knows but the corpus does not use; 917 are in neither.
 
 ### Not yet looked at
 
@@ -38,7 +38,6 @@ Corpus under `test/fixtures`: 35 OOXML workbooks (455 element names, 245 attribu
 | `lowestEdited`           | attribute | writer provenance — not modelled                                                                                                                                                                                                                                                                 |
 | `rupBuild`               | attribute | writer provenance — not modelled                                                                                                                                                                                                                                                                 |
 | `quotePrefix`            | attribute | measured, inert — marks a cell forced to text by a leading apostrophe. The value is already a string in the file, so `rows` is unaffected. Every occurrence in the corpus is `"0"`: openpyxl writes it explicitly false, Excel omits it. Nothing observed to carry                               |
-| `customFormat`           | attribute | restates that the row has a style, which the style says                                                                                                                                                                                                                                          |
 | `spans`                  | attribute | a hint at which columns a row uses. hucre derives that from the cells themselves, which is the authority — Excel treats a wrong `spans` as advisory too                                                                                                                                          |
 | `activeCell`             | attribute | view state — not modelled                                                                                                                                                                                                                                                                        |
 | `baseColWidth`           | attribute | measured, inert — the base every column width is relative to. Deriving `defaultColWidth` from it needs the normal font's maximum digit width, which hucre cannot measure. Every occurrence in the corpus is `8`, the schema default, and Excel omits it entirely — so nothing real is lost today |

diff --git a/scripts/spec-coverage.mjs b/scripts/spec-coverage.mjs
@@ -81,7 +81,7 @@
        const text = readFileSync(p, "utf8")
        corpusText += text + "\n"
        // A quoted literal — `case "sheetPr"` — is how a reader switches.
        for (const m of text.matchAll(/["'`]([A-Za-z_][\w:.\-]*)["'`]/g)) exact.add(m[1])
        // An object key — `xmlSelfClose("calcPr", { calcId: 0 })` — is how
        // a writer emits an attribute. Missing this form reported a dozen
        // attributes hucre writes on every workbook as unknown to it.
@@ -275,7 +275,6 @@
     // Out of scope by design.
     filterPrivacy: "privacy flag, no data — not modelled",
     pivotButton: "pivot UI affordance — pivots are round-trip only",
-    customFormat: "restates that the row has a style, which the style says",
     outlineLevelCol: "summary of the per-column outline levels hucre reads",
     outlineLevelRow: "summary of the per-row outline levels hucre reads",
 

diff --git a/src/_types.ts b/src/_types.ts
@@ -271,7 +271,9 @@ export interface ColumnDef {
   /**
    * Default style for every cell in the column. Applies whether the rows
    * come from {@link WriteSheet.data} or {@link WriteSheet.rows} — on the
-   * `data[]` path the generated header row gets it too.
+   * `data[]` path the generated header row gets it too. A row's
+   * {@link RowDef.style} takes its place, `numFmt` included, on the cells
+   * of that row (#601).
    */
   style?: CellStyle
   /** Number format. Folded into {@link style}; an explicit `style.numFmt` wins. */
@@ -908,6 +910,15 @@ export interface RowDef {
   outlineLevel?: number
   /** Whether this outline group is collapsed */
   collapsed?: boolean
+  /**
+   * Default style for the whole row — `<row s="N" customFormat="1">`. On
+   * write it also applies to every cell in the row that has no style of
+   * its own, and wins over {@link ColumnDef.style} there. Read back only
+   * with `readStyles`. A cell the file left at the default format inside
+   * such a row reads back with `style: {}`, so writing it again keeps it
+   * unformatted rather than filling it from the row. See #601.
+   */
+  style?: CellStyle
 }
 
 // ── Sheet ──────────────────────────────────────────────────────────

diff --git a/src/builder.ts b/src/builder.ts
@@ -239,7 +239,7 @@ export class SheetBuilder {
     return this
   }
 
-  /** Set row-level properties — height, hidden, outline level, collapsed. */
+  /** Set row-level properties — height, hidden, outline level, collapsed, format. */
   rowDef(
     row: number,
     def: NonNullable<WriteSheet["rowDefs"]> extends Map<number, infer T> ? T : never,

diff --git a/src/sheet-ops.ts b/src/sheet-ops.ts
@@ -956,7 +956,9 @@ export function cloneSheet(sheet: Sheet, newName: string): Sheet {
   if (sheet.rowDefs && sheet.rowDefs.size > 0) {
     const rowDefs = new Map<number, RowDef>()
     for (const [key, def] of sheet.rowDefs) {
-      rowDefs.set(key, { ...def })
+      // The style is an object of its own, so a spread alone would share
+      // it between the two sheets. See #601.
+      rowDefs.set(key, def.style ? { ...def, style: cloneCellStyle(def.style) } : { ...def })
     }
     cloned.rowDefs = rowDefs
   }

diff --git a/src/xlsx/stream-writer.ts b/src/xlsx/stream-writer.ts
@@ -81,11 +81,11 @@ export interface StreamWriterOptions {
    */
   repeatHeaders?: boolean
   /**
-   * Row-level properties — `height`, `hidden`, `outlineLevel`, `collapsed` —
-   * keyed by 0-based row index, counted the way the caller sees it: the key
-   * keeps rising past a rollover, where the per-sheet index restarts at 0.
-   * Consulted as each row is emitted, so it costs nothing per row that has
-   * no entry.
+   * Row-level properties — `height`, `hidden`, `outlineLevel`, `collapsed`,
+   * `style` — keyed by 0-based row index, counted the way the caller sees
+   * it: the key keeps rising past a rollover, where the per-sheet index
+   * restarts at 0. Consulted as each row is emitted, so it costs nothing
+   * per row that has no entry.
    */
   rowDefs?: Map<number, RowDef>
   /**
@@ -301,16 +301,22 @@ class RowSerializer {
     rowDef?: RowDef,
   ): string | null {
     const cellElements: string[] = []
+    // A row format reaches the cells that carry no style of their own and
+    // wins over the column's there, as it does in writeXlsx. See #601.
+    const rowStyle = rowDef?.style
 
     for (let c = 0; c < values.length; c++) {
       const colDef = this.columns?.[c]
       const raw = values[c]
       const styled = isStreamStyledCell(raw)
       const value = styled ? (raw.value ?? null) : (raw as CellValue)
-      let style: CellStyle | undefined = (styled && raw.style) || colDef?.style
+      const own = (styled && raw.style) || undefined
+      let style: CellStyle | undefined = own || rowStyle || colDef?.style
 
-      // If numFmt on column but not in style, merge
-      if (colDef?.numFmt && (!style || !style.numFmt)) {
+      // If numFmt on column but not in style, merge — except into a row
+      // format, which replaces the column's outright, number format
+      // included, the way writeXlsx resolves it. See #601.
+      if (colDef?.numFmt && (!style || !style.numFmt) && (own || !rowStyle)) {
         style = { ...style, numFmt: colDef.numFmt }
       }
 
@@ -324,11 +330,12 @@ class RowSerializer {
       if (cellXml) cellElements.push(cellXml)
     }
 
-    const attrs = rowAttributes(rowIndex, rowDef)
+    const attrs = rowAttributes(rowIndex, rowDef, this.styles)
 
     if (cellElements.length === 0) {
       // A row with no cells still has to be emitted when its definition asks
-      // for something — a height, or a hidden/grouped row the caller wants.
+      // for something — a height, a format, or a hidden/grouped row the
+      // caller wants.
       return hasRowAttributes(rowDef) ? xmlSelfClose("row", attrs) : null
     }
     return xmlElement("row", attrs, cellElements)

diff --git a/src/xlsx/styles-writer.ts b/src/xlsx/styles-writer.ts
@@ -311,8 +311,11 @@ export function createStylesCollector(
   const numFmtMap = new Map<string, number>()
   let nextNumFmtId = 164
 
-  // Cell XFs — default xf at index 0
-  const defaultXfKey = "nf:0|f:0|fl:0|b:0|a:|p:"
+  // Cell XFs — default xf at index 0. The key has to be spelled the way
+  // `registerXf` builds one, checkbox flag included, or a style that sets
+  // nothing — `{}` — never finds it and gets a second, identical xf. A
+  // cell read back as explicitly unformatted (#601) relies on landing here.
+  const defaultXfKey = "nf:0|f:0|fl:0|b:0|a:|p:|cb:0"
   const xfs: XfEntry[] = [
     {
       key: defaultXfKey,