rhythms & feedback · metaphor 88 of 100
You never receive the event; you receive the event smeared by who you are. Every perceiver has a kernel — a characteristic blur, echo, and delay applied to whatever arrives — and convolution is the mathematics of that smearing: one pattern dragged through another.
Tell the same news to three friends. In the first it lands sharp and is gone by dinner. In the second it reverberates for a week, surfacing at odd hours, quieter each time but unmistakably still there. The third barely hears the news at all — what they hear is the older thing it reminds them of, arriving again wearing this year's clothes. The input was identical, word for word. What differed was the receiver.
Engineers have a merciless test for this. Hit a system with one clean spike — a single clap in the concert hall, a single volt for a single instant — and record what it does: ring, echo, smooth, delay. That recording is the impulse response, and the strange gift of linear systems is that whatever the system does to one spike, it does to everything, spike by spike, overlapped and summed. That running sum is convolution. It is how history reaches you — never raw, always through the kernel.
the instrument · i
Five receivers, five kernels. Choose one, fire a single clean event into it, and watch what comes out the other side. The kernel is the personality.
the five kernels are hand-designed illustrations of temperament, fully disclosed in the drawings above — not clinical measurements. everything downstream of them is computed exactly.
the instrument · ii
Now a full year of events. The top track is what actually happened — click or drag on it to add, raise, and erase events (an event's height is its size, not its goodness). The bottom track is that year as the chosen receiver lives it: every event dragged into the future by the kernel, overlaps added together. Then switch receivers and watch one objective year become five different lives.
the kernel is the character
The impulse response is temperament written as a curve: what one clean event does in you. Does it land at true size and fade by Thursday? Does it echo for two months at half volume? Does it arrive twice — once now, once again at day forty-two, larger, wearing the face of an older wound? You cannot see your own kernel directly, but everyone around you can read it from your last three reactions.
And linearity gives the strange gift: if the system is linear, the spike test is a complete character reference. No need to catalogue its response to promotions, diagnoses, betrayals, and lottery wins separately — every year of events is just spikes in a row, and the response to all of them is the response to one, copied, scaled, shifted, and summed. One clean observation, total predictive power. That economy is why the instrument above works at all — and, as the caveats admit, exactly where the metaphor is most likely to lie to you.
what to try
smeared history
Siblings share a house, a set of parents, a decade of dinners — nearly the same input signal — and write memoirs that read like accounts of different families. The convolution view says this needs no dishonesty to explain: the same spikes through different kernels are different lived years, arithmetically. The gallery above is a shelf of memoirs of one year.
Institutions have kernels too. The organization that perceives the world through quarterly averages has chosen the wide smooth box: nothing sharper than a quarter survives into its awareness, so a two-week crisis is not minimized — it is never perceived at all, merely diluted below the threshold of attention (a cousin of aliasing, where coarse sampling doesn't just blur the signal but replaces it with a false one). And cultures are point-spread functions: what a society does with one martyr, one scandal, one miracle — how long it echoes, how much it amplifies, what old grievance it re-arrives as — it does to all of them, summed. Read a nation's response to a single event and you have read its kernel; its next century is that kernel, convolved with whatever arrives.
the instrument · iii
Shape a receiver with the sliders, or draw one freehand on the strip. The year you built above is run through it live — and the mirror question is computed underneath: what would a year have to look like for this receiver to have a calm one?
editing the kernel
The spikes mostly arrive on their own schedule: the diagnosis, the windfall, the phone call. What was always yours to work on is the other operand. Therapy, in this dialect, is largely kernel work: locating the delayed second peak — the old event that every new event re-triggers — and shrinking it, so that things can finally arrive only once. Practice narrows the anxious lobe: the thousandth cold call still spikes, but the overshoot and the crash after it flatten with repetition. And ritual mourning is a designed echo: cultures that prescribe seven days, thirty days, a year of kaddish are engineering a tail that is long enough to honor the loss and — this is the design insight — has an ending built in.
None of this changes what happened. It changes what happening does. Same input, edited kernel, different life — that is the hopeful reading of the arithmetic above.
deconvolution · a teaser
Convolution runs forward; sometimes you need it backward. Given only the perceived year — a memory, a memoir, a friend's account — and a good estimate of the kernel that produced it, you can partially un-smear the record and estimate what actually happened. This is deconvolution, and it is half of what a good therapist or historian does: first estimate the witness's kernel, then divide it back out of the testimony. But it is a fragile art. Perception always carries noise, and naive un-smearing amplifies noise explosively — trust the reconstruction too literally and you manufacture vivid events that never occurred. Drag the trust slider below and watch recovery curdle into confabulation — and whose record can be un-smeared is itself revealing: the sharp one barely needs it, while the institutional's monthly average is nearly impossible to invert, because smoothing destroys information nothing fully recovers. The cure for over-trust is regularization, the mathematics of appropriate humility (see inverse problems).
the mapping
| Mathematics | Life |
|---|---|
| the input x[n] | What actually happens — the events themselves, at their true size and time. |
| the kernel h[k] | Who it happens to — the characteristic blur, echo, and delay of the receiver. |
| impulse response | What one clean event does in you: the complete character reference, read from a single spike. |
| convolution x∗h | The year as lived: every event smeared into the future and summed with all the others. |
| the second peak | The old wound that every new event reopens — arriving again, on a delay, at full size. |
| deconvolution | The hard, noise-limited art of estimating what actually happened from how it felt. |
where the metaphor tears
Convolution's whole power rests on two assumptions: that responses add, and that the kernel today is the kernel tomorrow. Both fail at the worst moments. A big enough event does not pass through the kernel — it rewrites it: the loss that changes how every later loss lands. That is precisely what convolution cannot represent, by construction, and it is this metaphor's deepest limit. The instrument shows you a life through a fixed kernel; real lives are the story of the kernel changing.
"The anxious," "the traumatized," and the rest are hand-set curves chosen to make the mathematics legible — disclosed in full above, but invented. No clinical population has been measured to have these kernels, and no real person is one curve. Treat them as caricature in the honorable sense: exaggeration in service of a true mechanism.
The frame slides easily into determinism — my kernel made me do it — and that reading should be resisted. An impulse response is a description of what has tended to happen, not a law about what must. The editing section is the point. Of the two operands in the product, the kernel is the one that was always partly yours.