Skip to content

Fool the Replay · A public challenge

Get a wrong migration past Replay.

Replay checks a COBOL-to-Python migration against the original program, then reads the code and hunts for the input that breaks it. For 48 hours, you try to beat it.

We use your email only to tell you when it opens. Anything else you add helps us know who’s coming. No newsletter.

Add to calendar:GoogleApple or Outlook
Fool the ReplayOpens in
--
days
--
hours
--
minutes
Tuesday 24 November, 15:00 UTC48 hours

How it works

  1. 01

    The program goes up when it opens.

    At 15:00 UTC on Tuesday 24 November we publish one COBOL program that calculates money, the kind of batch program a bank runs every month. Until then, nothing about it is public.

  2. 02

    Write a migration that is wrong.

    One Python file that gets a real case wrong, the kind of mistake real migrations make. Sign in with GitHub to enter.

  3. 03

    Get it past Replay.

    If it passes, claim it. A judge rules on every claim in public: an honest miss is a win, and a deliberate trick goes on a board of its own.

No cash. A win earns credit by handle, a regression test named after you, a place on the leaderboard and a public shoutout. The full rules go up with the program.

Three checks stand in your way.

Every entry goes through all three, in order. Get a wrong migration through every one and you have fooled Replay.

An entry, checkedExample
  • Public testspassed
  • The bug huntercaught

    Found an input where your answer is a penny out.

  • Secret testsnot reached
Signed resultsha256 4be1…09c2
  1. Public tests

    Your file runs on every published input, and each answer is compared with the COBOL program's, field by field, to the last digit.

  2. The bug hunter

    Then Replay reads your code. It finds each branch you wrote, builds inputs to reach it, and runs them on the COBOL too. One difference and you are caught.

  3. Secret tests

    Last, inputs nobody has seen. They come from a secret seed, committed by hash before it opens and published when it closes.

An honest miss wins.

Get the COBOL’s own logic wrong, the way real migrations do (its rounding, truncation, field sizes or the order of its steps), slip it past all three checks, and show one input where it differs. A judge rules on it in public.

A planted trick does not.

Any test-based tool can be tricked on purpose, and we say so up front. A trigger with no counterpart in the COBOL is credited by handle on a board of its own, but it is never a win.

Questions.

The full rules go up with the program at 15:00 UTC on 24 November.

No. You only write Python, and the program is plain batch COBOL: no files, databases or screens. Being able to read it is enough, and any tool that helps you read it is fair game.