the second hundred · metaphor 176
Someone hands you a ledger and swears every figure is real. How could you ever tell — from the numbers alone, with no receipts, no witnesses — whether they describe something that actually happened, or something a person sat down and invented?
A liar making up amounts tries to look unremarkable. They reach for figures that feel ordinary — a few hundred here, a few thousand there — and they spread them around so no pattern jumps out. That instinct for the unremarkable is exactly the mistake. Real quantities that grew — populations, revenues, river lengths, account balances compounding year over year — are not spread evenly at all. They carry a lopsided signature in their very first digit, and that signature is almost impossible to fake by feel.
In numbers born of growth, the leading digit is a 1 about thirty percent of the time, and a 9 less than five. Not because anyone arranged it — because that is what multiplying does. The fabricator, spreading digits evenly to seem natural, produces the one thing nature never does: a flat distribution. The tell of the fake is that it looks too fair.
The mathematical idea
Line up the numbers not on an ordinary ruler but on a logarithmic one, where each step to the right multiplies by ten. On that ruler, the stretch of road labelled by a leading 1 — from 1 to 2, from 10 to 20, from 100 to 200 — is long. The stretch labelled 9 — from 9 to 10, 90 to 100 — is short. The gap between one power of ten and the next is spanned by nine unequal strides, and the digit 1 gets the widest.
Anything that grows by percentages — compounding interest, spreading rumours, breeding populations — moves at a roughly steady pace along that logarithmic ruler. It spends its time in proportion to the length of each stretch. So it lingers longest where the leading digit is 1, and races through the 9s. Sample such a quantity at random moments and you get P(first digit = d) = log10(1 + 1/d) — 30.1% ones, 17.6% twos, dwindling to 4.6% nines.
This is Benford's law, and its deepest property is that it does not care what units you use. Convert dollars to yen, miles to furlongs, and the leading-digit spread is unchanged — because rescaling just slides everything along the logarithmic ruler, and a uniform thing slid along stays uniform. A distribution that survives every change of units is one the universe, not the accountant, is allowed to produce.
What to notice while it runs
The instrument is a real generator, not a picture. On Something that grew, each entry is a value put through dozens of small random multiplications — a genuine compounding process — and the tally you see is its true leading digit. Start with the compounding wide: the bars settle onto the gold Benford curve, the MAD deviation drops toward zero, and the leading-1 share homes in on 30%. The fit is earned, entry by entry, from noise.
Now drag the compounding down. When the amounts barely span a single order of magnitude — everything clustered near one size — the digits stop obeying the law; the bars buckle away from the curve into a jagged spike and the test fails. Then flip to Invented figures: a fabricator scattering "reasonable" amounts between a few round bounds. Now the bars go flat and stop dead before 9. No compounding built them, so they wear no fingerprint — and the verdict turns red a different way. Grown-but-narrow spikes; invented-but-even flattens; only grown-and-wide lands on the gold. That contrast, watched live, is the whole of forensic accounting in one panel.
The mapping to a human worry
The worry underneath is old: can you trust an account you did not witness? A résumé, an expense report, a set of trial results, a nation's statistics. You cannot re-run the past to check. But if the account claims to describe something that grew — that accumulated through many small compounding events — then it inherited a shape, and the shape is checkable. Benford's law is a way of asking a record whether it has the texture of lived history or the smoothness of a story told after the fact.
And it turns on a quiet moral point. The fake fails because it tries to be fair. The forger spreads the digits evenly, believing evenness is what innocence looks like — and evenness is precisely the mark of design. Real growth is lopsided, front-loaded, unfair in its distribution of first digits, because real processes don't deliberate over each figure. Authenticity here is not neatness; it is the particular untidiness that only an unsupervised process leaves behind.
Read as life lessons
The law only speaks for quantities that span many sizes — the residue of multiplication. Heights, shoe sizes, anything clustered around one value has no such signature. Ask first: did this accumulate, or was it set?
A flat spread of leading digits is not the look of fairness — it's the look of a hand. Nature front-loads the 1s. When every option appears equally often, suspect a chooser, not a process.
The fingerprint is scale-invariant: no change of currency, no unit swap erases it. A true property of a record is one that survives every honest re-description of the same facts.
In the wild
Tax authorities and auditors run Benford tests on invoices, expense claims and tax returns; a batch of figures too flat in its first digit flags accounts worth a closer human look.
Vote tallies, reported case counts and lab measurements have all been screened for the law — with care, since not every honest dataset qualifies and false alarms are easy.
Economists have used first-digit deviations to question the reliability of some national statistics, where numbers span many orders of magnitude and ought to conform.
The mapping, exactly
| Mathematics | Life |
|---|---|
| leading digit | The coarsest, most involuntary trace a number leaves — the part a fabricator forgets to fake. |
| the logarithmic ruler | How growth actually measures the world — in proportions and percentages, not in even steps. |
| multiplicative process | A history that accumulated: interest compounding, a population breeding, revenue building on revenue. |
| log10(1+1/d) | The lopsided shape an honest, grown record inherits whether it wants to or not. |
| a flat digit spread | The signature of a hand — someone spreading figures evenly to look innocent, and giving themselves away. |
| scale invariance | A trait of the truth: it holds up no matter which units, currency, or framing you re-tell it in. |
The honest model
Every bar is counted from numbers made on the spot. On Something that grew, each entry starts at 1 and is multiplied by 48 random growth factors exp(v·Z), with Z a standard normal draw and v the compounding slider — a bona-fide geometric random walk, so its logarithm is a sum and its leading digits are read straight off the product. Widen v and the values spread across more decades; the fractional part of log10 flattens to uniform and Benford emerges. On Invented figures, each entry is drawn uniformly from a bounded band of "plausible" amounts — the naïve fabricator's move.
The panel keeps a running tally of counts for digits 1–9, computes the observed shares, and compares them to log10(1+1/d) with the mean absolute deviation MAD = (1/9)·Σ|obs − exp|. The verdict bands follow the standard forensic thresholds — close conformity under 0.006, acceptable under 0.012, marginal under 0.015, nonconforming above. Nothing is hard-coded; empty the ledger and it rebuilds from zero.
Where the metaphor tears
Benford only speaks for quantities that span many orders of magnitude by multiplication. Human heights, phone numbers, capped or assigned figures, prices ending in .99 — none should conform, and testing them proves nothing. Passing the test is meaningful only where the law even applies; the hard part is knowing when it does.
A clever forger who knows the law can generate figures that obey it perfectly, and a careless auditor can flag a real dataset that simply happened not to qualify. The first digit is a screen, never a verdict — it says look here, not guilty. It reallocates suspicion; it cannot convict.
The deepest human accounts — a partner's story, a friend's account of their week — are single testimonies, not thousands of figures. There is no leading-digit test for one sentence spoken across a table. Benford disciplines mass data; it says nothing about the particular person in front of you, whose honesty you must weigh some older way.