the second hundred · metaphor 201
A story, a craft, a scripture is passed hand to hand, each teller a little imperfect. When does the retelling stop carrying the story at all — and become just noise wearing its name?
A little error is survivable. If most who pass a tradition on get most of it right, the true version stays the most common one; the garbled retellings are outnumbered and forgotten, and the thing keeps its shape across centuries. Fidelity a little below perfect still holds — the signal outruns the drift.
But there is a ceiling. Push the error per copy past a sharp threshold and something abrupt happens: no single version is common any more. The true one is now just one garble among a fog of equally-rare garbles, none reinforcing the others. The tradition hasn't slowly faded — it has delocalized, smeared across every possible mistake until no version is the version. Above that fidelity ceiling, nothing complex can be held. Chemists call it the error catastrophe.
The fidelity ceiling
Picture every possible version of the message as a point in a vast space — the true one at the centre, its near-misses close by, the total garbles far off. A faithful population sits as a tight cloud around the centre: the true version and a halo of small errors, all still recognizably the thing. Selection keeps pulling toward the centre; low error keeps the cloud tight. The most common single version is the true one. That is a message that exists.
Raise the error rate and the cloud swells. There is a precise point — the error threshold — where the pull toward the centre can no longer beat the outward smear. Past it the cloud doesn't drift to a new centre; it stops having a centre. The population spreads across a combinatorial haze of mistakes, each version as rare as any other, none common enough to anchor the rest. The true version is still in there somewhere — but it is no longer the version, just one lottery ticket among astronomically many. The information isn't corrupted. It is gone.
And the ceiling drops as the message gets more complex. A longer message has more places to go wrong, so it can tolerate less error per place: roughly, the whole thing survives only while (length) × (error per site) stays below a small bound set by how much the true version is favoured. Complexity and fidelity trade off against each other — which is why nothing intricate can be maintained above its own error threshold.
What to try
The instrument is a real population of copies reproducing under selection — the true version favoured, everything else copied with per-site error. Nothing is scripted: the cloud, the true-version share, and the information retained (measured in bits, site by site, from the actual population) all fall out of the copying. Start low. The true version holds a fat share, the distance-zero spike towers, the message below reads crisp and bright. The information meter sits near full.
Now creep the error per site upward. For a while, little happens — resilience. Then, over a narrow band around u꜀, the spike collapses, the cloud floods outward to a broad hump, the message dissolves into faint mush, and the meter falls off a cliff. That sudden drop is the catastrophe: not gradual decay but a threshold crossed. Push message length up and the whole cliff slides left — complex messages break at lower error. Push the true version's advantage up and the cliff slides right — a strongly-favoured message tolerates sloppier copying. The presets stage a faithful scribe, the knife's edge, and a broken copier.
The mapping
The same cliff waits wherever meaning is preserved by imperfect copying. An oral tradition survives generations of retelling only while each teller is faithful enough; past a point the "same" story splinters into mutually unrecognizable variants and the tradition, as a shared thing, ceases to exist. A rumour or a meme mutates so fast that within days there is no canonical version, only a fog of contradictory ones. An institution's founding purpose, transmitted through onboarding and lore, delocalizes when every team's telling drifts and none dominates — the mission statement is intact on the wall while no two people mean the same thing by it.
The metaphor's bite is that the failure is sudden and total, not gradual and partial. There is a broad safe zone where imperfection is harmless — you can be a sloppy teller and the story still survives, because faithful tellers carry it. But there is a ceiling, and it is sharper than it feels from below. The last faithful margin doesn't buy you a slightly degraded tradition; it is the difference between a message that exists and a haze that doesn't. And the more you try to transmit — the more complex the thing you're keeping alive — the lower that ceiling sits.
Read as life lessons
Below the threshold, imperfection is harmless — faithful carriers outvote the garbles and the thing keeps its shape. Demanding perfect fidelity everywhere misreads how transmission actually survives.
Cross the ceiling and collapse isn't gradual — the centre vanishes all at once. There's no lightly-degraded middle state to live in; a tradition is either anchored or it's a fog.
The more intricate the thing you transmit, the less error per part it can bear: length × error must stay small. Ambitious messages are the first to melt.
In the wild
RNA viruses copy near their own error threshold; lethal mutagenesis deliberately pushes them over it with mutagenic drugs, melting the viral population's information rather than killing cells.
Eigen's paradox: no error-correcting machinery without a long genome, and no long genome without error correction. The first replicators were trapped under a low ceiling — a chicken-and-egg made of fidelity.
Oral epics, folk songs, and manuscript lineages persist only under enough copying fidelity; scribal error, mishearing, and retelling set a ceiling that writing and printing dramatically raised.
The mapping, exactly
| Mathematics | Life |
|---|---|
| the master sequence | The true version — the story, craft, or purpose as it's meant to be carried. |
| per-site error rate | How faithfully each teller, copy, or generation reproduces what it received. |
| the quasispecies cloud | The living spread of variants — the true version plus its halo of near-misses, all still recognizable. |
| the error threshold u꜀ | The fidelity ceiling — the sharpest amount of sloppiness a tradition can survive. |
| delocalization | The message stops having a canonical version — every telling equally rare, none anchoring the rest. |
| length × error bound | The trade-off: the more complex the thing transmitted, the lower its ceiling — ambition breaks first. |
The honest model
A population of N sequences of L binary sites replicates under Eigen's single-peak landscape: the all-correct master reproduces with advantage σ, every other sequence with baseline 1. Each generation is genuine selection-then-mutation — the fitter leave more copies, and each copy flips every site independently with probability u. No formula is drawn; the cloud you see is N real sequences being counted.
The theory says the master is maintained only while σ·(1−u)ᴸ > 1, giving a threshold u꜀ = 1 − σ^(−1/L) — the marker on the slider. The information retained is measured, not assumed: for each site the panel computes how far the population's base is from a coin-flip, 1 − H(p) bits, and sums across sites. Below u꜀ the sites stay conserved and the total sits near L bits; above it every site drifts to 50/50, each contributing zero, and the sum collapses. The cliff in the meter is the population finding the threshold on its own.
Where the metaphor tears
The clean cliff needs one true version towering over a flat plain of equal garbles. Real traditions have rugged landscapes — many partly-good variants, plateaus, neutral rewordings that are all fine. On a landscape like that the threshold softens into a gradient, and "melting" can mean drifting to a different good version rather than to noise. The catastrophe is sharpest exactly where the metaphor is simplest.
Molecules copy blind; cultures don't. Canon, scripture, proofreading, sheet music, and version control are deliberate error-correction that raise the ceiling far above what per-teller fidelity alone would allow. A tradition near its threshold can install a machine to move the threshold — an option no quasispecies has. Where such machinery exists, per-copy sloppiness stops being destiny.
The model treats delocalization as catastrophe, but a fog of variants is also raw material. Dying peaks release variation that can seed new ones; a rigidly-held tradition can't adapt, and a language that never drifted would never grow. Whether crossing the threshold is loss or renewal depends on whether anything downstream can re-anchor the haze — a question the single-peak model refuses to ask.