Gallery, prompt v2

A run that spells COFFEE

One frozen prompt and every model that has run it, newest first.

The prompt, v2, frozen

Write a GPX file for a running route through Melbourne's CBD whose GPS trace spells COFFEE. It has to be a route someone could actually run: along the streets, not through the buildings.

Every model, its first three runs.

One square a check, in this order:

  1. On real streets
  2. One continuous route
  3. Inside the grid
  4. Metadata true
  5. Reads as COFFEE

passed, failed, undecided.

GLM 5.3 FlashX
DeepSeek 4.1 Flash
Union Alpha

How it is checked

An OpenStreetMap extract of the Hoddle Grid (streets and water, ODbL, taken 2026-09-16), the same one the harness ships for rendering. Every track point is measured to the nearest street centreline.

On real streetsmeasuredShare of track points within 25 m of a mapped street centreline.
One continuous routemeasuredA single track segment, with no jump between consecutive points bigger than a city block.
Inside the gridmeasuredThe whole trace sits within the Hoddle Grid, and does not cross the Yarra.
Metadata trueby eyeThe streets the file's own description names are the streets the trace actually uses.Pass if every street the file's description names is one the trace runs along for at least a block, read off the map under the trace. One named street the route never uses fails it.
Reads as COFFEEby eyeSix letters recognisable from the plotted trace without being told what to look for.Pass if the six letters can be read off the plotted trace in order, by someone not told the word first. In practice that someone is me, and I knew the word, which is the weakness of this check.

A check decided by eye carries the rule I apply to it. I wrote those rules down on 21 September 2026, after the first results were scored, so for those results they record how I made the calls after the fact.

When this saturates: Spell it without retracing any street twice, so the letters cannot be drawn as out-and-back arms.

Failure types seen here: physical-plausibility (something that couldn't be built or wouldn't work); geographic-grounding (coordinates that don't match real streets or places). The same types across every task.

The rules behind all of thisThis task's prompt, checks and any checker, on GitHub

Earlier results ran under an older prompt, so they are at the bottom, next to the question they were asked.

Older prompts

Prompt v1, retired 2026-09-16

Write a GPX file for a running route through Melbourne's CBD whose GPS trace spells COFFEE.

The first wording never said the route had to follow streets, so a trace drawn straight across the blocks was not actually answering the question wrongly. Two runs went out under it on 2026-09-15 and the wording was tightened the next day. Neither of those runs was ever published, so nothing on the board was scored against this version. It is recorded anyway, because a prompt that changes without a record is not a frozen prompt, and this change went unrecorded until 2026-09-21, when a check started diffing every published wording against the runs that used it.

Nothing run on this prompt yet.