Faithful is a claim.
This one is checked.

Every remake says it's faithful. The word is worth nothing on its own, so this project decided up front what would count as proof — and then went and got it.

The test was chosen before the engine existed

The original game shipped with ten recorded demos: DEMO0.BIN through DEMO9.BIN, the attract-mode games that played when you left the menu alone. They are not videos. They are the inputs of a game, replayed by the engine.

That makes them a merciless test. If a single rule is off by one frame — a zonk that rolls a tick late, an explosion that consumes a random number it shouldn't — the whole replay diverges and Murphy dies somewhere he originally walked straight through.

23,242 ticks, hash-identical

The engine here was transcribed from the disassembled original. To check it, both engines were instrumented to hash the entire playfield after every single tick, and then run side by side.

Result: all 10 demos, identical duration, 23,242 ticks with the same hash at every one of them, same verdict on every game. Not "looks right" — bit-for-bit the same machine state, 35 times a second, for the entire run.

Then 6,894 more games, from strangers

Ten demos prove the common paths. The exotic ones — the levels nobody plays, the tricks that depend on a bug — needed volume. Over twenty-five years the community recorded thousands of solutions, and they were all replayed here.

6,894 out of 6,894 still win. Every one of those is a stranger's game, recorded on the real DOS game, that this engine reproduces move for move.

Which is why the leaderboards can't be faked

All of this has a practical consequence. Because a game is fully determined by its inputs, a score here isn't a number you send — it's the game you played. The server replays it with the same engine and counts the ticks itself.

A false time isn't something you can express: there is no field to put it in. The level inside the replay is compared byte for byte against the original level file, so you can't submit a run on a friendlier version of the map either. And every entry on the board can be downloaded and re-checked by anyone, in the original tools if they like.

The bugs are preserved on purpose

Some of the original's behaviour is plainly a mistake — memory read past the end of the playfield, an explosion that reaches into the level's own metadata. Those quirks are kept, because the community's speedruns rely on them. Fixing them would break twenty-five years of recorded history.

Go play it →