The layout is a Michelson interferometer, scaled up and made monstrously precise. A laser is split in two, sent down two perpendicular arms, bounced back and recombined; change one arm's length relative to the other and the beams interfere differently. GEO600's arms are 600 metres long and the beam runs each one twice, for an effective optical path of 1,200 metres. The optics sit in ultra-high vacuum at less than a hundred-millionth of a millibar. The main mirrors are fused silica cylinders eighteen centimetres across and ten thick; the beam splitter, twenty-six across and eight thick, is the only piece of glass the full-power beam must pass through rather than bounce off, so it was cut from a special grade of silica. Unlike LIGO and Virgo, GEO600 has no resonant cavities in its arms at all - a deliberate divergence, and part of why it became a laboratory rather than a rival.

At this sensitivity everything is noise. A mirror hanging on a wire hears the wire; a mirror with a magnet glued to it hears the glue. So GEO600's main mirrors hang from fibres of the same fused silica the mirrors themselves are made of - monolithic suspensions, chosen because silica loses less mechanical energy than steel, and mechanical loss is noise wearing a different hat. The secondary mirrors, less critical, keep their steel wires and glued-on magnets pushed around by coils. The main mirrors are moved instead by electrostatic drives: combs of electrodes behind the glass that raise an uneven electric field the mirror can feel but never touches. A ring of heaters at the far east mirror warms it so that thermal expansion tunes its radius of curvature. And the entire suspension cage rides on piezoelectric crystals that shove it in the opposite direction to whatever the ground is doing, cancelling the planet out.

Construction began in September 1995, a German-British project run by the Max Planck Institute for Gravitational Physics and Leibniz University Hannover with Glasgow, Birmingham and Cardiff. It has taken data alongside LIGO since 2002 and reached design sensitivity in 2006, with a peak of about two parts in ten thousand billion billion per root hertz at 600 Hz - limited by laser power at the high end and by seismic ground motion at the low. What matters more is what it demonstrated. Signal recycling, in which an extra mirror at the output forms a resonant cavity with the far mirrors and amplifies a potential signal. Homodyne readout. An output mode cleaner filtering light that cannot be carrying a signal. And, most striking, squeezed vacuum injected into the dark port of the beam splitter - manipulating quantum uncertainty itself to buy a factor of about 1.5 in sensitivity above 700 Hz. These are now standard equipment in every large detector on Earth.

By 2007 the collaboration was reporting, openly and in public talks, that it had unexplained noise in the middle of its frequency band. Craig Hogan, a physicist at Fermilab, proposed something more exciting than a loose cable: that the noise was real - the quantum graininess of spacetime itself, predicted if the universe encodes its information holographically. In 2009 New Scientist ran it as a cover story about living inside a hologram. The chronology is less cinematic than the coverage. Hogan's own Physical Review D paper of May 2008 cites a 2007 GEO600 talk that had already described the mid-band 'mystery' noise, and a GEO600 paper submitted in October 2007 said much the same. Karsten Danzmann, the project's principal investigator, offered the deflating and correct answer: the daily business of improving these experiments always throws up excess noise. Usually you find the cause and it goes away. This one did.

On 14 September 2015, when the two Advanced LIGO detectors registered the first gravitational wave ever observed, GEO600 was in engineering mode and would not have been sensitive enough regardless. It joined the observing run four days later. Its contribution has been stubbornness: from September 2011, with LIGO and Virgo both down for upgrades, it was the only large laser interferometer on the planet still watching, and it stayed on again through the shutdowns of 2020. Its output includes not just the main photodiode but microphones, seismometers, accelerometers and magnetometers, and volunteers have long sifted that data on home computers through Einstein@Home. After thirty-one years, the instrument is scheduled to shut down for good on 31 December 2026. Its techniques leave the field in far better shape than they found it.