Why does my error tracker show one bug as hundreds of different errors?

What splits one bug into many

Build output does it too. Bundlers put content hashes in file names, such as chunk-3f9a1c2b.js or .next/server/chunks/8123.js, so the same frame gets a new file name on every deploy.

Normalize before hashing

Replace the values that change, then hash. Order matters: replace UUIDs, timestamps and emails before plain numbers, or the number rule eats them first.

import crypto from 'node:crypto';

const RULES: Array<[RegExp, string]> = [
  [/\b[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}\b/gi, '<uuid>'],
  [/\b\d{4}-\d{2}-\d{2}[T ]\d{2}:\d{2}(:\d{2}(\.\d+)?)?(Z|[+-]\d{2}:?\d{2})?/g, '<time>'],
  [/[\w.+-]+@[\w-]+\.[\w.-]+/g, '<email>'],
  [/\b0x[0-9a-f]+\b/gi, '<hex>'],
  // Numbers on their own or stuck to a unit: 8124, user_4821, 10032ms. E11000 and utf8 stay.
  [/(^|[^a-z0-9])\d+(?:\.\d+)*(ms|s|m|h|kb|mb|gb|b)?(?![a-z0-9])/gi, '$1<n>'],
];

export function normalize(message: string): string {
  let out = message;
  for (const [re, to] of RULES) out = out.replace(re, to);
  return out;
}

export function fingerprint(type: string, message: string, file?: string, fn?: string): string {
  const parts = [type, normalize(message), file ?? '', fn ?? ''];
  return crypto.createHash('sha256').update(parts.join('|')).digest('hex').slice(0, 32);
}

Keep the words that tell bugs apart. ECONNREFUSED and ETIMEDOUT are different problems, and so are E11000 and E11001, and reading 'total' and reading 'items'. A rule that strips every digit or every quoted string merges them.

Leave the line number out

File and function are enough to place an error. With the line number in the fingerprint, adding an import at the top of a file changes the fingerprint of every error below it. They all look new, and a bug you marked fixed doesn't reopen when it comes back, because it now has a different fingerprint.

Use the first frame from your own code, not from a library, so the same bug reached through two call paths into a shared helper still groups by where your code went wrong.

Test both directions

• Pairs that must group: the same message with different IDs, durations, emails and timestamps, and the same frame in two different bundle chunks.
• Pairs that must stay apart: different error codes, different property names, different functions.
• Run the old and new fingerprint over a day of real events and count the groups before you switch. Existing groups will split from new ones once, so tell people.