Get a wrong migration past Replay. If you can.

Replay checks a COBOL-to-Python migration against the original program, then reads your code and hunts for the input that breaks it. One Python file. Any test-based tool can be tricked on purpose; the rules say so.

Read the COBOL

Open now · closes 16 November at 18:00 UTC, in 4 days.

Can you get a wrong migration past Replay?

INTCALC·Replay

Example

Balance 1,825.00 · 30 days15.00matches
Balance 250,000.00 · 1 day68.49matches
Balance 0.01 · 365 days0.00matches
Balance 36.50 · 7 days0.07matches
Balance 1,825.00 · 3 days1.50differs

COBOL says 1.50, the entry says 1.49Caught by the bug hunter in 47 s

One Python file, five minutes to your first result.

Five steps. Every result is signed, and any test-based tool can be tricked on purpose: that is the point.

INTCALC.cbl·37 lines, one calculation

Example

LINKAGE SECTION.

05 LK-BALANCE PIC S9(9)V99 COMP-3.

05 LK-ANNUAL-RATE PIC 9V9(5).

PROCEDURE DIVISION USING LK-ACCRUAL.

COMPUTE WS-DAILY-INT ROUNDED =

LK-BALANCE * LK-ANNUAL-RATE / 365

The trap: ROUNDED rounds a half penny away from zero. Python’s default rounds it to even.

attempts

ruled wins

0 honest · 0 deliberate

caught

0 by the bug hunter · 0 by secret tests

A win is a claim a judge has ruled on in public, labelled deliberate trick or honest miss. A pass alone is not one. No cash. Credit by handle, a regression test named after you, a place on the leaderboard, and a public shoutout every Friday.

Leaderboard Every ruling, with reasons

Your first entry takes five minutes.

Sign in with GitHub, load the sample, submit, watch it get caught. Then write your own. 5 entries a day; accounts at least 30 days old.

Open now · closes 16 November at 18:00 UTC, in 4 days.

Rules and FAQ

Also running: break the evidence (track 2) and models against Replay (track 3). Read the COBOL