Tokoyo holds twenty-nine observations.
What is here are the twenty-four that did not become one.
Each creature in the list looks as though it had always been that shape. It had not.
The ray (026) came out of seven remakings, and those seven are still here, still running.
The form of a creature is fixed only by its relation to the ones that were thrown away.
Take the discarded ones out, and the reason for the shape goes with them.
It is the same thing that happens when you thin the points in the list.
On the left is the ray. On the right are the three that came before it.
The shape is the same; only the speed differs. The one chosen was the middle of the seventh.
The numbers show that what was wrong was not the beat of the wings but the drift of the whole body.
The sixth drifts about four times faster than the ray does now. The wings were right from the start.
Observation 026Ray
move 0.210px ・ jerk 0.0022 ・ afterimage 0.88 ・ 20s per turn
Sixth generationD1
move 0.916px ・ jerk 0.0373 ・ afterimage 0.88 ・ 5s per turn
A wave travels along the wing from front to back; the wingtips lag most and swing widest
Seventh generationE1
move 0.272px ・ jerk 0.0043 ・ afterimage 0.88 ・ 20s per turn
One beat in five seconds
Seventh generationE3
move 0.228px ・ jerk 0.0024 ・ afterimage 0.88 ・ 20s per turn
One beat in ten seconds. Slower than the animal itself
Six with a straight spine and a transverse wave laid over it. At the time we ruled that the
skeleton was placed wrongly, and threw all six away.
That ruling was later withdrawn. The jellyfish is exactly a wave laid over a near-straight
spine, and it is one of the three he named as closest to being alive.
No comment was ever given on these six. They were discarded on our judgement alone.
First generationM1
move 1.258px ・ jerk 0.0489 ・ afterimage 0.86 ・ 5s per turn
A plain sine at constant speed. Movement from lag alone
First generationM2
move 1.309px ・ jerk 0.0597 ・ afterimage 0.86 ・ 5s per turn
Time itself bent, so it gathers slowly and releases fast
First generationM3
move 1.365px ・ jerk 0.0681 ・ afterimage 0.86 ・ 5s per turn
It moves forward only when the tail beats quickly
First generationM4
move 1.201px ・ jerk 0.0465 ・ afterimage 0.88 ・ 10s per turn
One strong beat per turn, then ringing on as it decays
First generationM5
move 0.855px ・ jerk 0.0250 ・ afterimage 0.86 ・ 5s per turn
The signal to turn travels from head to tail, late
First generationM6
move 1.238px ・ jerk 0.0518 ・ afterimage 0.86 ・ 10s per turn
The swimming beat and a slower breath running at once
The skeleton was replaced with one that turns as a whole. All three passed every one of the
eight tests we had then. They passed and were still not taken, because they looked like
turning ornaments rather than creatures.
They pass all eight, but as a creature it is thirty out of a hundred
Here we corrected the measurement twice. The reading that the tufts were not moving in any of
the three was an error in how we measured. We nearly read the unmeasurable as the absent.
Second generationA
move 0.896px ・ jerk 0.0183 ・ afterimage 0.74 ・ 10s per turn
The angle advances with the square of position along the body
Second generationB
move 1.605px ・ jerk 0.0591 ・ afterimage 0.74 ・ 10s per turn
The angle advances in an S — slow at the ends, fast in the middle
Second generationC
move 1.060px ・ jerk 0.0320 ・ afterimage 0.74 ・ 10s per turn
The amount of coiling itself breathes with time
One creature alone in the void reads as a specimen; several read as alive. Having noticed that,
we increased the number. A wave in the angle itself, regions, and nested trigonometry went in too.
Too fast. The rightmost is a little better, but too cluttered.
The brain cannot find the pattern, so it does not look alive
Third generationA2
move 0.878px ・ jerk 0.0290 ・ afterimage 0.74 ・ 10s per turn
Three bodies. Head and trunk follow different rules; the angle itself undulates
Third generationA3
move 1.041px ・ jerk 0.0447 ・ afterimage 0.74 ・ 10s per turn
Two bodies. The difference between regions strengthened, the projection flattened harder
Third generationA4
move 0.995px ・ jerk 0.0204 ・ afterimage 0.76 ・ 10s per turn
Five bodies. Small, tightly coiled, each swimming its own path on a scattered beat
Overlap, depth of the tufts, and sheer number of points — the causes of "cluttered" were
separated into three and fixed.
There is no comment on this generation. We went on to the next without showing it.
No video was ever shot. Of the seven, this is the only one no one has seen.
Run now, one turn takes forty seconds (a work is fixed at twenty at most).
One of them still drops sixty-five points, where a fractional power falls onto a negative base —
a hole we fell into three times over this work. They are left unrepaired, because repairing
them would make them no longer the things that were thrown away.
Fourth generationB1
move 0.129px ・ jerk 0.0027 ・ afterimage 0.7 ・ 40s per turn
Three bodies, placed where they cannot touch
Fourth generationB2
move 0.289px ・ jerk 0.0060 ・ afterimage 0.7 ・ 40s per turn
One large body alone — a control, to separate whether the clutter came from number or from making
Fourth generationB3
move 0.544px ・ jerk 0.0035 ・ afterimage 0.72 ・ 40s per turn
Five bodies, small, each circling its own territory slowly
The body was cut into head, trunk and tail by switching rather than blending: at each border the
ribs are pinched to zero. Joints appear, and the parts read as separate things.
The turning of the whole was halved. What was too fast was only the turning of the whole; the wave
along the body was right — measured and corrected, and then missed again in the next generation.
The turning feeling isn't it. The delicate points and smooth motion of the cocoon and the ripple
are closest to being alive. Basing it on a real animal and reproducing that might work better
Fifth generationC1
move 0.231px ・ jerk 0.0031 ・ afterimage 0.72 ・ 20s per turn
Three bodies, each with head, trunk and tail clearly divided
Fifth generationC2
move 0.472px ・ jerk 0.0064 ・ afterimage 0.72 ・ 20s per turn
One large body — the plainest form in which to see whether the parts read
Fifth generationC3
move 0.342px ・ jerk 0.0050 ・ afterimage 0.72 ・ 20s per turn
Two bodies, with the strongest contrast between parts
We withdrew the policy and rebuilt from how real animals swim. No more invented scrolls.
Here the drift coefficients were left at the standard one, two, three.
In the fifth we had measured that only the motion of the whole was too fast and corrected it —
and put it back in a new equation. The same oversight, twice.
The one on the left (the ray) is better. A little too fast. Slow, like the original
Sixth generationD1
move 0.916px ・ jerk 0.0373 ・ afterimage 0.88 ・ 5s per turn
A wave travels along the wing from front to back; the wingtips lag most and swing widest
Sixth generationD2
move 0.943px ・ jerk 0.0384 ・ afterimage 0.88 ・ 5s per turn
The mantle is stiff. A wave runs the fins, and ten arms follow late
Sixth generationD3
move 0.949px ・ jerk 0.1303 ・ afterimage 0.86 ・ 5s per turn
The body is nearly still and only the dorsal fin trembles fast. The stillness is what makes the tremble read as alive
Nothing about the shape was changed — only the speed, in three steps, so that what was too
fast could be measured apart from everything else.
One beat of the wings had been right from the beginning. What was four times too fast was the
drift alone. For seven generations, that was where the cause had been.
The middle one was chosen, and became the ray (026).
Seventh generationE1
move 0.272px ・ jerk 0.0043 ・ afterimage 0.88 ・ 20s per turn
One beat in five seconds
Seventh generationE2became observation 026
move 0.210px ・ jerk 0.0022 ・ afterimage 0.88 ・ 20s per turn
One beat in six point seven seconds. This became the ray
Seventh generationE3
move 0.228px ・ jerk 0.0024 ・ afterimage 0.88 ・ 20s per turn
One beat in ten seconds. Slower than the animal itself
Two rules were written, measured, and found to be wrong.
We wrote: *a straight spine cannot yield a creature; the whole must turn.*
The jellyfish is exactly a near-straight spine.
We wrote: *keep the afterimage thin; thick and the ribs dissolve.*
All three that read as alive had thick afterimages. The dissolving was what made the delicacy.
Both are wrong in the same way. We took a model apart to derive a law, and never measured our
own successes. Measure only one side, and the reason a thing works gets written into the law
as a thing absent.
Compared on smoothness, the studies were better. So smoothness was not the difference.
The difference was whether it turned, and how thick the afterimage was.
Both numbers here are measured directly from the equations.
Move is the average distance a point travels in one frame (the picture is four hundred units
across; a frame is a sixtieth of a second).
Jerk is the average of how sharply that motion changes.
The same rule is applied to the works, so a discarded study and a finished work can stand side by side.