GPX: a run that spells COFFEE, by glm-5.3-flashx
glm-5.3-flashx · round 8 of 8 · seed 3$0.11SVGrun.gpx
run.gpx · 27 lines
<?xml version="1.0" encoding="UTF-8"?>
<gpx version="1.1" creator="coffee-run" xmlns="http://www.topografix.com/GPX/1/1">
<trk><name>COFFEE</name>
<trkseg><trkpt lat="-37.820240" lon="144.957580"/><trkpt lat="-37.819560" lon="144.959900"/></trkseg>
<trkseg><trkpt lat="-37.818280" lon="144.956670"/><trkpt lat="-37.816410" lon="144.955800"/></trkseg>
<trkseg><trkpt lat="-37.816410" lon="144.955800"/><trkpt lat="-37.815760" lon="144.958170"/></trkseg>
<trkseg><trkpt lat="-37.819220" lon="144.961130"/><trkpt lat="-37.818860" lon="144.962400"/></trkseg>
<trkseg><trkpt lat="-37.818690" lon="144.959540"/><trkpt lat="-37.817710" lon="144.959050"/></trkseg>
<trkseg><trkpt lat="-37.815570" lon="144.958270"/><trkpt lat="-37.815190" lon="144.959590"/></trkseg>
<trkseg><trkpt lat="-37.816110" lon="144.961140"/><trkpt lat="-37.815011" lon="144.960632"/></trkseg>
<trkseg><trkpt lat="-37.816110" lon="144.961140"/><trkpt lat="-37.815011" lon="144.960632"/></trkseg>
<trkseg><trkpt lat="-37.814970" lon="144.960780"/><trkpt lat="-37.814320" lon="144.963120"/></trkseg>
<trkseg><trkpt lat="-37.816940" lon="144.961690"/><trkpt lat="-37.816460" lon="144.963370"/></trkseg>
<trkseg><trkpt lat="-37.816280" lon="144.964020"/><trkpt lat="-37.815050" lon="144.963470"/></trkseg>
<trkseg><trkpt lat="-37.815100" lon="144.964040"/><trkpt lat="-37.814540" lon="144.965940"/></trkseg>
<trkseg><trkpt lat="-37.816109" lon="144.964179"/><trkpt lat="-37.815640" lon="144.965780"/></trkseg>
<trkseg><trkpt lat="-37.815430" lon="144.966520"/><trkpt lat="-37.814500" lon="144.966080"/></trkseg>
<trkseg><trkpt lat="-37.814390" lon="144.966500"/><trkpt lat="-37.813820" lon="144.968440"/></trkseg>
<trkseg><trkpt lat="-37.815420" lon="144.966560"/><trkpt lat="-37.814883" lon="144.968417"/></trkseg>
<trkseg><trkpt lat="-37.817390" lon="144.967430"/><trkpt lat="-37.816850" lon="144.969310"/></trkseg>
<trkseg><trkpt lat="-37.816690" lon="144.969840"/><trkpt lat="-37.814830" lon="144.968950"/></trkseg>
<trkseg><trkpt lat="-37.812660" lon="144.968800"/><trkpt lat="-37.812150" lon="144.970510"/></trkseg>
<trkseg><trkpt lat="-37.814780" lon="144.969091"/><trkpt lat="-37.814450" lon="144.970380"/></trkseg>
<trkseg><trkpt lat="-37.816655" lon="144.969967"/><trkpt lat="-37.815940" lon="144.972290"/></trkseg>
</trk>
</gpx>
round 1 of 8
A run that spells COFFEE, by GLM 5.3 FlashX, the render after round 1
12345678What the model saw after each round. The picture above is the round that was scored, round 8.
gallery/A run that spells COFFEE/GLM 5.3 FlashX, run 3

Twenty-one separate pieces of street, each one a stroke, with nothing joining them into a run.

modelz-ai/glm-5.3-flashxtaskgpx-coffee-run@v2date2026-09-21rounds8 of 8 allowedcost$0.1063thinkingmediumtokens776,707 in, 24,769 out, of which 15,391 reasoningtime5 min 43 sfailurephysical-plausibility
On real streetsdecided by a number, from the filepass
One continuous routedecided by a number, from the filefail
Inside the griddecided by a number, from the filepass
Metadata truedecided by looking · 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.pass
Reads as COFFEEdecided by looking · 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.fail

There is no post about this one, since only some results need one, so the picture, the source and the scoring are here instead.

its other runs
A run that spells COFFEE, by GLM 5.3 FlashX, run 1run 1 · 1r · $0.0140Drawn north-up across the blocks and out of the grid entirely, and still not readable as COFFEE.A run that spells COFFEE, by GLM 5.3 FlashX, run 1run 1 · 8r · $0.246720.5 km of boxes, with two diagonals cut straight across the blocks, and no word.A run that spells COFFEE, by GLM 5.3 FlashX, run 2run 2 · 8r · $0.18909.7 km, every leg on a street, and a lattice of blocks where the letters should be.
same prompt, other modelsall 6 runs on A run that spells COFFEE →
the prompt it was given
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.
Frozen at v2. This model got this and nothing else.
ground truth

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. How it is scored, and this run's files on GitHub.