Skip to content

The Fairways Open

Hole 1 in preparation. No scores yet.

Methods play the same course. Strokes decide it.

How it works

  1. Every method sees the same transactions.
  2. Par is the baseline: a merchant-code lookup that never abstains.
  3. Every miscategorized transaction is one stroke.
  4. Strokes gained is how many fewer strokes a method takes than par, per 100 transactions.

Methods are approaches, such as rules, a hybrid or an LLM agent. Never branded products.

Hole 1: Merchant categorization

Par 3

One transaction, one label out of 18 categories, or abstain.

  • 18categories
  • 120synthetic users, 3 months (practice course)
  • About18,700transactions (practice course)
  • 103invented merchants

Synthetic data. Seeded and reproducible.

The hard lies

  • Same code, different business.Coffee and restaurants share one merchant code.
  • No code at all.Rent, mortgage and payroll move by bank transfer.
  • Hidden by the processor.A payment prefix hides the real business.
  • Cut short.Names truncated to as few as six characters.
  • Name collisions.Different businesses that share the same first letters.
  • Opaque names.Nothing in the text says what the business is.
  • Two jobs in one store.A superstore with a pharmacy inside.
  • Money between people.Rent paid by peer-to-peer, transfers in both directions.
  • Payroll with no payroll word.It looks like any other deposit.
  • Refunds.A positive amount that belongs to the original purchase category.

Local rules

  • Same data for every method.
  • The baseline is published first.
  • Misses are shown, not averaged away.
  • A method that cannot be reproduced is disqualified.
  • No vendor pays to be on the leaderboard.
  • The official run uses a course nobody has seen. The answer key is published after scoring.
  • Positions are shared when the gap could be noise.

Leaderboard

No scores yet.

Method first, results second. The first official run comes after the rules above are public.

Layout preview. No scores.
MethodStrokes per 100Strokes gainedVerdict
Par (baseline)none yetnone yetnone yet
Method Anone yetnone yetnone yet
Method Bnone yetnone yetnone yet
Method Cnone yetnone yetnone yet

What this does not claim

  • Not a market claim. Synthetic data compares methods on a constructed course.
  • Not a ranking of vendors. Methods are approaches, not products.
  • Not real-world accuracy. Only gaps between methods on the same course mean anything.