diff --git a/src/cli/completions/directive-scanner.ts b/src/cli/completions/directive-scanner.ts
index 36841d2..1291511 100644
--- a/src/cli/completions/directive-scanner.ts
+++ b/src/cli/completions/directive-scanner.ts
@@ -51,6 +51,18 @@ function looksLikeDice(raw: string): boolean {
   return /^\d*d\d+/i.test(raw) || /^[+-]/.test(raw);
 }
 
+/** Parse key=value pairs from directive extra attrs string */
+function parseDirectiveAttrs(extraStr: string | undefined): Record<string, string> {
+  if (!extraStr) return {};
+  const attrs: Record<string, string> = {};
+  const re = /(\w+)\s*=\s*("[^"]*"|\S+)/g;
+  let m: RegExpExecArray | null;
+  while ((m = re.exec(extraStr)) !== null) {
+    attrs[m[1]] = m[2].replace(/^"|"$/g, "");
+  }
+  return attrs;
+}
+
 // ---------------------------------------------------------------------------
 // Markdown table → CSV conversion
 // ---------------------------------------------------------------------------
@@ -136,6 +148,7 @@ export function buildSparkTableCompletion(
   csv: string,
   filePath: string,
   csvPath: string,
+  remix: boolean,
   slugger: Slugger,
 ): SparkTableCompletion | null {
   const dataHeaders = inspectSparkTableCsv(csv);
@@ -159,6 +172,7 @@ export function buildSparkTableCompletion(
     filePath: basePath,
     csvPath,
     headers: dataHeaders,
+    remix,
   };
 }
 
@@ -248,7 +262,7 @@ export function scanDirectives(
     newIndexEntries[resolvedPath] = csv;
 
     // Collect spark table completion
-    const st = buildSparkTableCompletion(csv, filePath, resolvedPath, slugger);
+    const st = buildSparkTableCompletion(csv, filePath, resolvedPath, false, slugger);
     if (st) {
       sparkTables.push(st);
     }
@@ -290,7 +304,11 @@ export function scanDirectives(
     let csv = index[csvPath] ?? newIndexEntries[csvPath];
     if (!csv) continue;
 
-    const st = buildSparkTableCompletion(csv, filePath, csvPath, slugger);
+    // Parse extra attrs for remix flag
+    const attrs = parseDirectiveAttrs(extraStr);
+    const isRemix = attrs["remix"] === "true";
+
+    const st = buildSparkTableCompletion(csv, filePath, csvPath, isRemix, slugger);
     if (!st) continue;
 
     // Check if data-spark is already set in extra attrs
diff --git a/src/cli/completions/spark-scanner.ts b/src/cli/completions/spark-scanner.ts
new file mode 100644
index 0000000..44c2c50
--- /dev/null
+++ b/src/cli/completions/spark-scanner.ts
@@ -0,0 +1,114 @@
+/**
+ * Spark table scanner — CLI-side markdown table detection and conversion.
+ *
+ * Parses markdown tables, detects spark tables (first column header is a dice
+ * formula), and converts them to CSV for the directive scanner.
+ *
+ * Not imported at runtime — the frontend only needs CSV parsing + rolling.
+ */
+
+import Slugger from "github-slugger";
+
+// ---------------------------------------------------------------------------
+// Types
+// ---------------------------------------------------------------------------
+
+export interface MarkdownTable {
+  headers: string[];
+  rows: string[][];
+}
+
+// ---------------------------------------------------------------------------
+// Markdown table parser
+// ---------------------------------------------------------------------------
+
+/**
+ * Parse all markdown tables from a markdown string.
+ * Handles both leading/trailing `|` styles and bare styles.
+ */
+export function parseMarkdownTables(markdown: string): MarkdownTable[] {
+  const tables: MarkdownTable[] = [];
+  const lines = markdown.split(/\r?\n/);
+
+  for (let i = 0; i < lines.length; i++) {
+    const headerCells = splitTableRow(lines[i]);
+    if (!headerCells || headerCells.length < 2) continue;
+
+    // Peek at the next line — must be a separator row
+    if (i + 1 >= lines.length) continue;
+    const sepCells = splitTableRow(lines[i + 1]);
+    if (!sepCells || sepCells.length < headerCells.length) continue;
+    if (!sepCells.every((c) => /^:?-{3,}:?$/.test(c))) continue;
+
+    // Valid table header + separator — collect body rows
+    const rows: string[][] = [];
+    let j = i + 2;
+    while (j < lines.length) {
+      const rowCells = splitTableRow(lines[j]);
+      if (!rowCells) break;
+      // Allow rows with fewer cells (unfilled trailing columns)
+      rows.push(rowCells);
+      j++;
+    }
+
+    // Only include tables with at least one data row
+    if (rows.length > 0) {
+      tables.push({ headers: headerCells, rows });
+    }
+    i = j - 1;
+  }
+
+  return tables;
+}
+
+/** Split a pipe-delimited table row, stripping optional leading/trailing `|` */
+function splitTableRow(line: string): string[] | null {
+  const trimmed = line.trim();
+  if (!trimmed.includes("|")) return null;
+
+  // Strip optional leading and trailing `|`
+  let inner = trimmed;
+  if (inner.startsWith("|")) inner = inner.slice(1);
+  if (inner.endsWith("|")) inner = inner.slice(0, -1);
+
+  return inner.split("|").map((c) => c.trim());
+}
+
+// ---------------------------------------------------------------------------
+// Spark table detection & metadata
+// ---------------------------------------------------------------------------
+
+const DICE_HEADER_RE = /^d\d+$/i;
+
+/** Check whether a table is a spark table (first header is a dice formula) */
+export function isSparkTable(table: MarkdownTable): boolean {
+  if (table.headers.length < 2) return false;
+  return DICE_HEADER_RE.test(table.headers[0]);
+}
+
+/**
+ * Generate the spark table slug by concatenating slugs of all data column
+ * headers (excluding the dice column).
+ */
+export function sparkTableSlug(table: MarkdownTable): string {
+  const slugger = new Slugger();
+  return table.headers
+    .slice(1)
+    .map((h) => slugger.slug(h.toLowerCase()))
+    .join("-");
+}
+
+/**
+ * Scan all spark tables in a markdown file and return their metadata
+ * (without rows — suitable for listing available tables).
+ */
+export function scanSparkTables(
+  markdown: string,
+): { notation: string; slug: string; dataHeaders: string[] }[] {
+  const tables = parseMarkdownTables(markdown);
+  return tables.filter(isSparkTable).map((table) => ({
+    notation: table.headers[0],
+    slug: sparkTableSlug(table),
+    dataHeaders: table.headers.slice(1),
+  }));
+}
diff --git a/src/cli/completions/types.ts b/src/cli/completions/types.ts
index 15d273f..ba1c110 100644
--- a/src/cli/completions/types.ts
+++ b/src/cli/completions/types.ts
@@ -52,6 +52,8 @@ export interface SparkTableCompletion {
   csvPath: string;
   /** Data column headers for display */
   headers: string[];
+  /** Whether to roll each column independently (remix mode) */
+  remix: boolean;
 }
 
 export interface CompletionsPayload {
diff --git a/src/components/journal/JournalInput.tsx b/src/components/journal/JournalInput.tsx
index 9eb99de..9659c50 100644
--- a/src/components/journal/JournalInput.tsx
+++ b/src/components/journal/JournalInput.tsx
@@ -128,7 +128,8 @@ export const JournalInput: Component = () => {
         const key = (parsed.payload as { key: string }).key;
         const match = comp.data.sparkTables.find((s) => s.slug === key);
         const csvPath = match?.csvPath ?? "";
-        const p = await resolveSparkPayload({ key, csvPath });
+        const remix = match?.remix ?? false;
+        const p = await resolveSparkPayload({ key, csvPath, remix });
         const result = sendMessage("spark", p);
         const r = unwrap(result);
         finish(r.ok, r.err);
diff --git a/src/components/journal/completions.ts b/src/components/journal/completions.ts
index c8417e1..56277b8 100644
--- a/src/components/journal/completions.ts
+++ b/src/components/journal/completions.ts
@@ -54,6 +54,7 @@ export interface SparkTableCompletion {
   filePath: string;
   csvPath: string;
   headers: string[];
+  remix: boolean;
 }
 
 export interface JournalCompletions {
diff --git a/src/components/journal/types/spark.tsx b/src/components/journal/types/spark.tsx
index 80c4d81..f28c662 100644
--- a/src/components/journal/types/spark.tsx
+++ b/src/components/journal/types/spark.tsx
@@ -70,10 +70,12 @@ export type SparkPayload = z.infer<typeof schema>;
  *
  * `key` is the combined slug (pageName-columnSlug).
  * `csvPath` is the resolved .csv file path from completions.
+ * `remix` controls whether each column gets an independent roll.
  */
 export async function resolveSparkPayload(raw: {
   key: string;
   csvPath: string;
+  remix: boolean;
 }): Promise<SparkPayload> {
   let csv: string;
   try {
@@ -89,7 +91,7 @@ export async function resolveSparkPayload(raw: {
     );
   }
 
-  const sparkResult = rollSparkTable(meta);
+  const sparkResult = rollSparkTable(meta, { remix: raw.remix });
   const firstRoll = rollFormula(meta.notation);
 
   return {
diff --git a/src/components/utils/spark-table.ts b/src/components/utils/spark-table.ts
index 77c1e40..9059d88 100644
--- a/src/components/utils/spark-table.ts
+++ b/src/components/utils/spark-table.ts
@@ -1,13 +1,19 @@
 /**
- * Spark Table — markdown table where the first column header is a dice
- * formula (d6, d20, d100, etc.). Rolling a spark table means:
+ * Spark Table — runtime rolling and CSV parsing.
  *
- * 1. Roll the dice formula once for each data column (non-dice columns)
- * 2. Look up the row whose dice-column value matches each roll
- * 3. Return the rolled values keyed by column slug
+ * A spark table is a CSV whose first column header is a dice formula
+ * (d6, d20, d100, etc.). Rolling a spark table means:
+ *
+ * 1. Roll the dice formula once
+ * 2. Look up the row whose dice-column value matches the roll
+ * 3. Return all column values from that row
+ *
+ * When `remix` is true, each data column gets its own independent roll
+ * and may come from different rows.
  */
 
 import Slugger from "github-slugger";
+import { parse } from "csv-parse/browser/esm/sync";
 import { parseCSVString } from "../utils/csv-loader";
 import { rollFormula } from "../md-commander/hooks";
 
@@ -15,11 +21,6 @@ import { rollFormula } from "../md-commander/hooks";
 // Types
 // ---------------------------------------------------------------------------
 
-export interface MarkdownTable {
-  headers: string[];
-  rows: string[][];
-}
-
 export interface SparkTableColumn {
   header: string;
   slug: string;
@@ -40,6 +41,8 @@ export interface SparkTableMeta {
   notation: string;
   /** Concatenated slug of data columns */
   slug: string;
+  /** The header name of the dice column (first column) */
+  diceHeader: string;
   /** Data column headers (excluding dice column) */
   dataHeaders: string[];
   /** Full list of rows as objects keyed by header */
@@ -47,185 +50,52 @@ export interface SparkTableMeta {
 }
 
 // ---------------------------------------------------------------------------
-// Markdown table parser
-// ---------------------------------------------------------------------------
-
-/**
- * Parse all markdown tables from a markdown string.
- * Handles both leading/trailing `|` styles and bare styles.
- */
-export function parseMarkdownTables(markdown: string): MarkdownTable[] {
-  const tables: MarkdownTable[] = [];
-  const lines = markdown.split(/\r?\n/);
-
-  for (let i = 0; i < lines.length; i++) {
-    const headerCells = splitTableRow(lines[i]);
-    if (!headerCells || headerCells.length < 2) continue;
-
-    // Peek at the next line — must be a separator row
-    if (i + 1 >= lines.length) continue;
-    const sepCells = splitTableRow(lines[i + 1]);
-    if (!sepCells || sepCells.length < headerCells.length) continue;
-    if (!sepCells.every((c) => /^:?-{3,}:?$/.test(c))) continue;
-
-    // Valid table header + separator — collect body rows
-    const rows: string[][] = [];
-    let j = i + 2;
-    while (j < lines.length) {
-      const rowCells = splitTableRow(lines[j]);
-      if (!rowCells) break;
-      // Allow rows with fewer cells (unfilled trailing columns)
-      rows.push(rowCells);
-      j++;
-    }
-
-    // Only include tables with at least one data row
-    if (rows.length > 0) {
-      tables.push({ headers: headerCells, rows });
-    }
-    i = j - 1;
-  }
-
-  return tables;
-}
-
-/** Split a pipe-delimited table row, stripping optional leading/trailing `|` */
-function splitTableRow(line: string): string[] | null {
-  const trimmed = line.trim();
-  if (!trimmed.includes("|")) return null;
-
-  // Strip optional leading and trailing `|`
-  let inner = trimmed;
-  if (inner.startsWith("|")) inner = inner.slice(1);
-  if (inner.endsWith("|")) inner = inner.slice(0, -1);
-
-  return inner.split("|").map((c) => c.trim());
-}
-
-// ---------------------------------------------------------------------------
-// Spark table detection & metadata
+// CSV parsing
 // ---------------------------------------------------------------------------
 
 const DICE_HEADER_RE = /^d\d+$/i;
 
-/** Check whether a table is a spark table (first header is a dice formula) */
-export function isSparkTable(table: MarkdownTable): boolean {
-  if (table.headers.length < 2) return false;
-  return DICE_HEADER_RE.test(table.headers[0]);
-}
-
-/**
- * Generate the spark table slug by concatenating slugs of all data column
- * headers (excluding the dice column).
- */
-export function sparkTableSlug(table: MarkdownTable): string {
-  const slugger = new Slugger();
-  return table.headers
-    .slice(1)
-    .map((h) => slugger.slug(h.toLowerCase()))
-    .join("-");
-}
-
-/**
- * Find a spark table in the given markdown content matching `slug`.
- * Returns null if not found.
- */
-export function findSparkTable(
-  markdown: string,
-  slug: string,
-): SparkTableMeta | null {
-  const tables = parseMarkdownTables(markdown);
-  for (const table of tables) {
-    if (!isSparkTable(table)) continue;
-    if (sparkTableSlug(table) === slug) {
-      return {
-        notation: table.headers[0],
-        slug,
-        dataHeaders: table.headers.slice(1),
-        rows: rowsToObjects(table),
-      };
-    }
-  }
-  return null;
-}
-
-/**
- * Find a spark table by its combined slug (`pageName-columnSlug`).
- * Iterates every spark table in the content, constructs the combined slug,
- * and returns the first match.  Avoids the need to guess where the page
- * name ends and the column slug begins (page names may contain `-`).
- */
-export function findSparkTableByCombinedSlug(
-  markdown: string,
-  combinedSlug: string,
-  pageName: string,
-): SparkTableMeta | null {
-  const tables = parseMarkdownTables(markdown);
-  for (const table of tables) {
-    if (!isSparkTable(table)) continue;
-    const colSlug = sparkTableSlug(table);
-    const candidate = `${pageName}-${colSlug}`;
-    if (candidate === combinedSlug) {
-      return {
-        notation: table.headers[0],
-        slug: colSlug,
-        dataHeaders: table.headers.slice(1),
-        rows: rowsToObjects(table),
-      };
-    }
-  }
-  return null;
-}
-
-/** Convert a MarkdownTable's string[][] rows to Record<string, string>[] */
-function rowsToObjects(table: MarkdownTable): Record<string, string>[] {
-  return table.rows.map((row) => {
-    const obj: Record<string, string> = {};
-    for (let i = 0; i < table.headers.length; i++) {
-      obj[table.headers[i]] = row[i] ?? "";
-    }
-    return obj;
-  });
-}
-
 /**
  * Parse a CSV string into a SparkTableMeta.
  * The CSV must have a dice formula as its first column header.
  * Returns null if the CSV is not a valid spark table.
  */
 export function parseSparkTableCsv(csv: string): SparkTableMeta | null {
-  const parsed = parseCSVString(csv);
-  const headers = Object.keys(parsed[0] ?? {});
+  // Parse raw headers first — before parseCSVString injects frontmatter keys
+  // into rows. We use csv-parse directly to get the header order reliably.
+  const rawParsed = parse(csv, {
+    columns: false,
+    comment: "#",
+    trim: true,
+    skipEmptyLines: true,
+    bom: true,
+  }) as string[][];
 
+  if (rawParsed.length < 2) return null;
+
+  const headers = rawParsed[0];
   if (headers.length < 2) return null;
   if (!DICE_HEADER_RE.test(headers[0])) return null;
 
+  // Now parse with csv-loader for full frontmatter + quoting support
+  const parsed = parseCSVString(csv);
+
   const slugger = new Slugger();
+  const diceHeader = headers[0];
   const dataHeaders = headers.slice(1);
   const slug = dataHeaders
     .map((h) => slugger.slug(h.toLowerCase()))
     .join("-");
 
   return {
-    notation: headers[0],
+    notation: diceHeader,
     slug,
+    diceHeader,
     dataHeaders,
     rows: parsed as Record<string, string>[],
   };
 }
 
-/** Scan all spark tables in a markdown file and return their metadata */
-export function scanSparkTables(
-  markdown: string,
-): Omit<SparkTableMeta, "rows">[] {
-  const tables = parseMarkdownTables(markdown);
-  return tables.filter(isSparkTable).map((table) => ({
-    notation: table.headers[0],
-    slug: sparkTableSlug(table),
-    dataHeaders: table.headers.slice(1),
-  }));
-}
-
 // ---------------------------------------------------------------------------
 // Range parsing
 // ---------------------------------------------------------------------------
@@ -267,33 +137,62 @@ function matchesCell(diceCell: string, rolledValue: number): boolean {
 // Rolling
 // ---------------------------------------------------------------------------
 
+export interface RollSparkTableOptions {
+  /** When true, each data column gets its own independent roll. Default false. */
+  remix?: boolean;
+}
+
 /**
- * Roll a spark table: for each data column, roll the dice formula and look
- * up the corresponding row value.
+ * Roll a spark table.
+ *
+ * By default, rolls the dice once and reads all columns from the matched row.
+ * When `remix` is true, each data column gets its own independent roll and
+ * may come from different rows.
  */
-export function rollSparkTable(meta: SparkTableMeta): SparkTableResult {
+export function rollSparkTable(
+  meta: SparkTableMeta,
+  options: RollSparkTableOptions = {},
+): SparkTableResult {
   const slugger = new Slugger();
   const columns: SparkTableColumn[] = [];
 
-  const diceHeader = Object.keys(meta.rows[0] ?? {})[0] ?? "";
+  if (options.remix) {
+    // Independent roll per column
+    for (const header of meta.dataHeaders) {
+      const slug = slugger.slug(header.toLowerCase());
+      const roll = rollFormula(meta.notation);
+      const rolledValue = roll.result.total;
 
-  for (const header of meta.dataHeaders) {
-    const slug = slugger.slug(header.toLowerCase());
+      let value = `(no row for ${rolledValue})`;
+      for (const row of meta.rows) {
+        if (matchesCell(row[meta.diceHeader] ?? "", rolledValue)) {
+          value = row[header] ?? "";
+          break;
+        }
+      }
 
+      columns.push({ header, slug, value });
+    }
+  } else {
+    // Single roll — all columns from the same row
     const roll = rollFormula(meta.notation);
     const rolledValue = roll.result.total;
 
-    // Find the row matching the rolled value
-    let value = `(no row for ${rolledValue})`;
+    let matchedRow: Record<string, string> | null = null;
     for (const row of meta.rows) {
-      const diceCell = row[diceHeader] ?? "";
-      if (matchesCell(diceCell, rolledValue)) {
-        value = row[header] ?? "";
+      if (matchesCell(row[meta.diceHeader] ?? "", rolledValue)) {
+        matchedRow = row;
         break;
       }
     }
 
-    columns.push({ header, slug, value });
+    for (const header of meta.dataHeaders) {
+      const slug = slugger.slug(header.toLowerCase());
+      const value = matchedRow
+        ? (matchedRow[header] ?? "")
+        : `(no row for ${rolledValue})`;
+      columns.push({ header, slug, value });
+    }
   }
 
   return {