spread & inequality · metaphor 92 of 100

The finding mind
putters, then leaps.

The productive mind putters — small steps around the current spot — and then, rarely, leaps somewhere unconnected entirely. Foraging albatrosses, browsing readers, wandering attention: when the good things are scattered and sparse, the optimal search is mostly small steps punctuated by huge jumps. The leaps are not distractions. They are the strategy.

Watch yourself research anything: minutes of local digging — this page, its neighbors, its citations — then an unexplainable jump to a different shelf, a different field, a walk outside. Guilt says the jump is weakness of focus. The mathematics of search says otherwise: pure local digging over-searches exhausted neighborhoods — Brownian motion revisits itself endlessly; pure jumping never mines a find.

The mixture — heavy-tailed steps, mostly small, occasionally vast — is what optimal foragers actually do when targets are sparse and patchy. There is even a number for the mixture: an exponent μ that tunes your putter-to-leap ratio. Below, a searcher hunts hidden targets under your control of μ. Drag it, and let the field tell you which strategy the world rewards.

The foraging grounds · one searcher, one exponent

Targets (green) are hidden, sparse, and clustered. The searcher takes steps whose lengths are drawn from a power law — small ones mostly, huge ones rarely. Low μ = nearly ballistic, all leaps. μ≈2 = classic Lévy. High μ = effectively Brownian, all putter. The efficiency curve at right is an honest Monte Carlo, filling in as trials accumulate across every μ.

the field is:
0.0
finds per 1,000 traveled · current μ, live
peak μ so far · the world's answer
0
Monte-Carlo trials pooled into the curve

efficiency = finds ÷ distance, normalized · current μ   μ=2   running peak

Your μ is how often you let yourself jump shelves. Keep targets sparse and the efficiency peak sharpens at the Lévy middle; flood the field and the peak flattens — when good things are everywhere, strategy barely matters.
Honest simulation. Step lengths are drawn from a truncated power law P(ℓ)∝ℓ−μ (ℓmin=detection radius, ℓmax=field width) by inverse-transform sampling; directions are uniform. A target is found when the searcher's straight segment passes within the detection radius; foraging is non-destructive (targets persist), the regime whose optimum sits near μ≈2. Each Monte-Carlo trial runs a fresh field to a fixed distance budget; the curve is a running mean per μ-bin, recomputed when you move the density or patchiness sliders. Nothing is asserted — the peak is where the trials put it.

The shape of finding

Two ways to search badly, and the middle that works.

Turn μ up past 3 and the searcher becomes Brownian — a jitterer, taking only tiny steps. Watch what it does to a sparse field: it burns its whole budget grinding over ground it has already covered. A random walk in two dimensions is recurrent — it comes home again and again — so most of its effort is spent re-checking neighborhoods it already knows are empty. Diligence, here, is just expensive re-reading.

Now pull μ down toward 1 and the searcher becomes ballistic — nothing but enormous leaps in straight lines. It reaches fresh ground instantly, but it never mines: it blows straight through a patch, catching one target where a patch holds a dozen, and rockets off before it has worked the cluster. All tail, no knots. Novelty without finding.

The heavy-tailed middle — the celebrated μ≈2 — does both. Mostly it putters, working the patch it has landed in; rarely, on a draw from the fat tail, it leaps clear across the field to somewhere it has never been. For sparse, patchy targets this is provably close to optimal, and the instrument reproduces it rather than asserting it: leave the Monte Carlo running and the peak climbs to μ≈2 on its own. Then flood the field with targets and watch the peak flatten: when good things are everywhere, every strategy finds them fast and the Lévy advantage fades. The wandering mind's edge is a phenomenon of scarcity — it earns its keep exactly when targets are sparse and patchy, and hardly at all when they are not. The right strategy is a fact about your world, not your character.

The revisit meter · oversampling, made visible

Two searchers, equal travel budget, same field. The Brownian putterer and the Lévy mixer both walk the same total distance — but colour each patch of ground by how many times it was searched. Blue is fresh; burning red is ground worked over and over for nothing.

BROWNIAN · μ=3.2
new ground per step:
LÉVY · μ=2.0
new ground per step:

Same steps, same distance. The putterer keeps re-searching the block it already knows; the mixer spends its budget on ground it has never seen.

Honest computation. Both walks take the same number of steps on the same grid; "new ground per step" is the count of grid cells entered for the first time, divided by steps. The heat is a literal visit-count per cell. Brownian steps are fixed-length; Lévy steps are power-law with μ=2. No field of targets here — this isolates the geometry of coverage.

What to try

Three experiments on the field above.

01

Sweep μ and find the peak.

Leave the density on sparse and drag μ slowly from 1.1 to 3.5, watching the live finds-per-1,000 and the accumulating curve. The efficiency peak forms near μ≈2 — the wandering mind out-finds both the jitterer and the rocket.

02

Flood the field. Watch the edge vanish.

Push target density to the right. As the good things stop being scarce, efficiency climbs everywhere and the peak flattens — when everything is nearby, how you move barely matters. The Lévy advantage is a fact about scarcity: match your strategy to how sparse your world actually is.

03

Check the revisit meter.

Run a fresh pair and read "new ground per step." The Brownian searcher spends most of its steps re-covering worked ground; the Lévy searcher keeps buying new territory. Diligence that revisits is just √t creeping back home.

Attention as forager

Your mind is a searcher with an adjustable μ.

The same dial names two familiar pathologies. Rumination is μ→∞: attention circling the same searched ground, re-checking the same grievance, the same worry, the same paragraph, finding nothing because there was nothing new to find and the search never leaves. Mania is μ→1: nothing but leaps, an idea abandoned mid-sentence for the next, huge distance covered and nothing mined — the tail without the knots. Health is not a fixed point between them; it is a heavy-tailed mixture — mostly present and local, occasionally, deliberately, gone.

Which means the long jump can be institutionalized. The scheduled wander, the walk with no destination, the unrelated book pulled off a shelf you had no business at, the sabbatical, the conversation with someone outside your field — these are the fat tail, kept alive on purpose, because a searcher that has quietly set its own μ to infinity will grind a single exhausted neighborhood forever and call it focus. If you cannot remember your last real jump, your μ has drifted, and the field has stopped yielding.

Worlds that set your μ for you

Someone else is holding the dial.

You rarely choose your step-length distribution alone; the environment proposes it. Feeds and "more like this" clamp you local — every recommendation is a small step to an adjacent item, and an engine optimized for engagement has every reason to keep your μ high, because a putterer is predictable and monetizable in a way a leaper is not. Libraries, cities, and sabbaticals lengthen your tail — the physical stacks that put an unrelated field one shelf away, the street that opens onto a neighborhood you had no reason to enter, the season with no agenda. The toy below makes the cost concrete: two ways to comb the same catalog of hidden gems.

The playlist test · more-like-this vs. the mix

A catalog of items; the good ones — the gems (green) — are clustered and rare. More like this only ever steps to a neighbour. The Lévy mix mostly steps local, but keeps a live tail of long jumps. Equal number of moves. Count the gems each one finds.

MORE LIKE THIS · local only0 gems
THE LÉVY MIX · local + jumps0 gems

Engagement engines set your μ high because putter is monetizable — and this is what it costs you in gems the local walk can never reach.

Honest simulation. The catalog is a grid; gems sit in a few clusters. Both walkers get the same move budget and the same start. The local walker steps to a random neighbour each move; the Lévy walker does too, except a power-law fraction of moves are long jumps. Gems are counted once, on first visit. Re-run to see the distribution — the mix wins on average, not every single time.

So the practical question is not only "what is my μ?" but "who benefits from it?" Audit your step-length distribution the way you would audit a diet. If everything you read, hear, and consider arrived by adjacency — one more like the last — your tail has been trimmed for you, and the gems that live three shelves over, in the clusters your neighbours never point to, are going unfound on your behalf. The remedy is not more discipline. It is a longer jump.

The mapping

Mathematics ↔ life.

MathematicsLife
step length ℓHow far the next move of attention goes — a neighbouring idea, or a different field entirely.
the power-law tailRare, huge jumps kept possible: the walk outside, the unrelated book, the shelf you had no business at.
exponent μYour putter-to-leap ratio — how often you let yourself jump shelves.
sparse, patchy targetsGood ideas, good jobs, good books — clustered and rare, not spread evenly for the taking.
oversampling (√t recurrence)Rumination, the re-read inbox, the block circled for the tenth time — effort spent on searched ground.
the efficiency peakWhy the wandering mind out-finds the disciplined one — in the right worlds, and only there.

Where the metaphor tears

Three honest failures.

The empirical claim was famously overstated.

The idea that real animals forage by Lévy flights is contested. The celebrated albatross data was later corrected — Edwards et al. (2007) showed the original power-law signature was largely an artifact of the birds sitting on their nests, and much of it vanished on reanalysis. So the honest claim is not a law of nature but a conditional optimality: in sparse, patchy, non-destructive search, a heavy tail near μ≈2 beats both extremes. Whether any given creature — or reader — actually walks that way is a separate, empirical question.

Human jumps carry a re-immersion tax.

The model's steps are free; a mind's are not. Every long jump pays a switching cost — the minutes of re-orientation before a new field yields anything, the lost thread you must pick back up. Those costs are real and the pure Lévy model omits them entirely, which pushes the truly optimal μ higher than the model says: with a re-immersion tax, you should putter somewhat more and leap somewhat less than a frictionless forager. The lesson survives, but the number moves.

Leaping only counts if you land and dig.

The strategy is putter-then-leap, and the "then" is load-bearing. A jump is only strategy if you mine after landing — work the patch you arrive in. The jump-addicted browser has the tail without the knots: novelty as a compulsion, forty tabs and no finding, μ→1 dressed up as curiosity. Long jumps redeem local digging; they do not replace it. Without the knots, the tail is just distraction after all.