not the code.
the scatter.
the scatter.
same sensor, same led, same room. the difference is how far your readings move when nothing moves.
same light.
how far does it
get through?
how far does it
get through?
image: ai-generated (gpt-image-2)
your results
| team | alphabet | hit rate | what they did about noise |
|---|---|---|---|
| team 7 | 16 | 98 % | cardboard tube, fixed distance, averaged over 8 |
| team 2 | 8 | 96 % | longer integration time, shielded from the window |
| team 11 | 8 | 91 % | picked colours further apart |
| team 5 | 4 | 100 % | nothing, played it safe |
| team 14 | 4 | 82 % | nothing, and it still failed |
same hardware. same task. a factor of four between first and last.
ten readings of one colour
nothing changed. the numbers did.
the same signal, measured differently
longer windows, calmer values. and fewer of them per second.
two symbols, two clouds
same distance. wider clouds. some readings land on the wrong side.
the ratio is what counts
your alphabet size measures nothing but this ratio.
how many symbols fit?
the range divided by the width of one band. that is your alphabet.
try it: the distinguishability lab
the toolbox, with price tags
three of them cost nothing but craft. that is why the optics are allowed.
averaging is not free
def read_stable(n=10):
values = [sensor.get_clear() for _ in range(n)]
return sum(values) / n
ten times calmer readings. ten times slower.
both roads lead to the same wall
both spend the same margin. that wall is where challenge 2 begins.
and one honest sentence
we believed:more gain gives us more colours
we measured:the scatter grew with it, and we lost two symbols
the channel is the only authority in this room.
the fog stays.
measure how far
you see.
measure how far
you see.
image: ai-generated (gpt-image-2)