both causes are one question
when does a new symbol begin?
answer one: time tells you.
answer two: change tells you.
how do eighty people
stay together?
stay together?
image: ai-generated (gpt-image-2)
your homework: where did it tip?
| team | symbols per second | what broke |
|---|---|---|
| team 7 | 14 | letters shifted by one, slowly, after about a minute |
| team 2 | 9 | suddenly nothing matched a profile any more |
| team 11 | 6 | every second word was garbled from the start |
| team 5 | 4 | held, we did not dare go faster |
same limit for everyone. very different numbers.
what one dot means on the next slides
one dot in the pictures = many readings, averaged.
cause one: no two clocks run alike
the tiny error per symbol is not the problem. the sum is.
cause two: the sensor needs time to look
no error, no warning, just a meaningless number.
when does it tip?
a shared start is not a question of if it tips, only when.
eighty people, one clap
agreeing once is not enough. agreeing again and again is.
recurring markers reset the drift
the error can never grow beyond one block.
try it: the drift simulator
the price of the clocked way
how bad are your clocks? that is a measurement, not a belief.
first idea: a change means "next symbol"
no change, no symbol. "red red" arrives as one.
the fix: the transition carries the bit
measure as often as you like. decode only on change.
freedom is paid in capacity
want a clean 2 bits per transition? you need five colours, not four.
the most common wrong conclusion
clock-free is not time-free
how long must a colour hold to count?
how many transitions per second can the sensor follow?
what your file costs in minutes
two levers, nothing else: bits per symbol, symbols per second.
today
1 the clapping experiment
2 workshop: find your maximum rate, with evidence
3 write down which way you lean, clocked or clock-free, and why
no two clocks agree.
the only choice is
what you do about it.
the only choice is
what you do about it.
image: ai-generated (gpt-image-2)