Porygon

I The episode that wrote the rules

Roughly four and a half million households had it on. The sequence was an explosion drawn as red and blue frames alternating at about twelve hertz, held for some four seconds, and the reports came in through the evening: nausea, then convulsions in a few hundred children, most of whom were fine by morning. It is remembered in Japan as the Pokémon Shock. The programme went off the air for about four months.

What came out of it is the part worth borrowing. Japanese broadcasters wrote flash limits the following year and the British regulator wrote its own. Both eventually became an international recommendation, which is a number where there had been a judgement: how many flashes a second, over how much of the screen, with how much change in brightness. There is a machine that tests against it. Content is submitted to it before it airs.

None of that existed on 15 December 1997. Television had been going for fifty years. What it did not have was an instrument pointed at this particular way of hurting somebody, and the instrument was built the week after it was needed, which is the ordinary way these things arrive.

And Porygon, which the episode was named for, did not cause it. The flash came from Pikachu's attack on a set of missiles. Porygon was in the room. It has been kept off the screen ever since, for nearly thirty years, and that misattribution is the second thing this sheet borrows.

II Nine a second, and where it came from

Tuning into a programme already running has to catch up, so the valley steps hard until it reaches the moment on air. The numbers were all in the tree, each sensible on its own.

Read down that column and every entry is a good decision. The last two rows are not a decision at all. Nobody wrote nine hertz; it is what 600 and 0.05 and 200 come to when multiplied by a frame rate that appears in none of them.

The fix is four words long: the sky is held at the light of the moment being caught up to, until the catching up is done. Every step still runs and the settlers still live all of it. Only the painting waits.

III A gate that has to be able to fail

The gate has a 23rd check now, and it does two things rather than one. It runs the frames a viewer would get on tuning in three quarters of an hour late — 90 of them, 13.5 valley days — and measures the light the way an eye does, by how often it crosses the middle. Then it runs the same sequence without the fix and demands that one still read as dangerous.

held: swing 0.000, 0 crossings, 0.0 Hz · control: swing 1.000, 27 crossings, 9.0 Hz tools/check-flash.mjs, the line it prints when it passes

The second half is the whole value. A check on a property that is meant to be absent passes by doing nothing, and there are a dozen ways to write one that would pass for reasons having nothing to do with the fix. If the control ever stops finding the strobe, the check fails and says the instrument has stopped measuring, rather than printing a pass it has not earned.

It caught its own author first. The first version reported the fix as leaving one flash behind, and the fix was fine — the check painted the sky before stepping the clock where the game paints it after. The control said 9.0 Hz, correctly, while the measurement beside it was wrong, and the two disagreeing is the only reason anybody looked. Three more faults in the same evening's work read as working until a real click went near them. A button inside a label ran perfectly when a script pressed it and ticked the wrong thing when a hand did. A confirmation was written into a row the panel rebuilds every few seconds, so Copied. was true and invisible. And a clipboard call neither succeeded nor failed but sat there, which the button read as still working.

IV What would not clear the standards

Nobody was hurt here. 685 children were taken to hospital in 1997 and this is a settlement game with an audience you could count, so the comparison is one of shape and not of scale, and a sheet that let those two sit level would be helping itself to somebody else's seriousness.

The nine hertz is calculated, not measured off a screen. It follows from four numbers in the tree and a frame rate of 60. A viewer at 30 frames a second gets something else, and so does a machine dropping frames or a tab in the background. Possibly worse: 12 is nearer the middle of the band than 9. What was measured is the light the code computes, not the light a room saw.

And the gate is one check, on one route in. Any other place this valley speeds up is unexamined. The dial goes to 32×, which works out at a sixth of a hertz and is nowhere near the band. But that argument was made this evening by the same person who shipped the nine, and it has no test under it.

Hedge on the borrowed shape: the 1997 figures are the ones usually reported and the counts vary by source; the standards that followed were years of work by broadcasters and regulators rather than one document; and the neat parallel — that Porygon was blamed for somebody else's flash the way the simulation would have been blamed for the renderer's — is neat because it was chosen, out of a story that had a great many other parts.