# Packet loss has a shape, and the shape is the diagnosis > Thirty squares missing from a grid of four hundred tell you nothing by their count. Where they cluster is what actually points at the cause. August 19, 2026 · https://justog.club/stories/packet-loss-pattern Flat white on near-black, twenty rows of twenty small solid squares — four hundred in total. Around thirty are missing. They're not scattered evenly through the grid. They cluster into two or three ragged patches, and one patch runs right up to the edge. Count the gaps and they tell you nothing. Look at where they sit and they tell you almost everything. ## What is this grid actually recording? A count of something sent against a count of something received, laid out spatially instead of as a single percentage. Send four hundred and get back three hundred and seventy, and the number alone says "some loss." Laid out as a grid instead, the same result says where in the sequence the loss happened, which is the piece a single percentage throws away the moment it's calculated. Two connections can report an identical ninety-two and a half per cent and mean two entirely different things by it, and the grid is the only one of the two formats that shows which. ## Why does the shape of the missing squares matter more than the count? Because random loss and caused loss look completely different once they're drawn, even at the same total. Squares missing at random, one here and one three rows down, is the signature of background noise — a connection with a slightly worse baseline than it should have, losing a little everywhere without any single point being to blame. Squares missing in a tight, ragged patch is the signature of an event: something happened in a narrow stretch of time or in one specific stretch of the route, and everything sent during that window paid for it. Same thirty squares, opposite explanation. ## Why does the patch touching the edge matter specifically? Because it tells you the event hadn't finished when the recording stopped. A patch fully enclosed inside the grid describes something that started and ended within the window being measured — a fault that recovered. A patch that runs off the edge describes something still happening, or something that started before the recording began. It's the one detail in the whole grid that points outside the frame instead of describing what's inside it, the way [an internal diagram only ever shows the machine as it was the moment somebody opened it up](/stories/hacked-hardware-diagram). ## What makes a good version of this design? A grid large enough that a cluster reads as a genuine shape rather than a coincidence, and a loss count light enough that most of the grid still stands. Twenty by twenty is close to the right size — small enough to read as a shape at chest scale, large enough that two or three separate patches don't just look like one big blank patch of nothing. Around thirty gaps out of four hundred keeps the pattern legible: enough missing squares to form a real shape, few enough that the grid still reads as mostly intact. ## Does the same reading apply to loss outside of networks? Anywhere a regular sequence gets interrupted, the same rule holds — gaps spread evenly point at a general weakness, and gaps clustered together point at a specific moment or a specific place. A cable with worn insulation drops signal the same way along its whole length, and [the colour coding along its run tells you where to start looking rather than what actually happened](/stories/wiring-loom-colour-code) — a fault at one join in the loom clusters exactly the way this grid does. The pattern is the diagnosis either way, before anyone opens anything up to look closer. A gift for somebody who spends their working hours reading exactly this kind of shape out of a screen lands because it's a joke only they're fully equipped to get. ## What's the strongest objection to reading a grid this small so precisely? That thirty squares out of four hundred is a small sample to draw conclusions from, and somebody could read a shape into noise that isn't actually there. That risk is real with any small sample, which is why the clustering has to be obvious rather than subtle — a handful of adjacent gaps forming a ragged block is a different claim from two gaps that happen to sit near each other. The design earns the reading by making the patches unambiguous rather than asking anyone to squint, closer to a fault map than to [the aestheticised static of glitch art, which multiplies noise on purpose rather than trying to locate it](/stories/glitch-art-print). ## Does it print well? Yes, provided the grid stays crisp rather than photographic. Flat white squares on near-black hold their edges at chest scale far better than a soft dot pattern would; softness is exactly what would blur a cluster back into noise. Keep the squares small and evenly spaced with the gaps as clean absences rather than faded ones — a faded square reads as "sent but weak," a different claim from "never arrived." One ink, one grid, no shading. Near-black as the garment ground keeps the missing squares reading as true gaps in the dark rather than as a pattern printed onto a lighter field. ## How do I get one made? Describe your grid at [JustOG](/) — the size, the loss count, where the patches sit. Pick a direction, drag the crop frame, see it composited on the real garment, and it's made to order and shipped. Designs other people have published are in the [shop](/shop). Three hundred and seventy squares standing, thirty gone, and every one of the gaps in the same two places.