Chapter 1 — The Man Who Was Never Wrong

There was a real man behind Sherlock Holmes, and he was a surgeon.

Joseph Bell taught at the Edinburgh Royal Infirmary, where Arthur Conan Doyle was his clerk. Bell's party trick — and it was, in the lecture theatre, deliberately a trick — was to tell a patient his occupation, his recent travel, and his habits before the patient had finished sitting down. A tan line here, a callus there, a particular way of holding the shoulder. Doyle watched him do it for a year and then spent forty years writing him down. [ESTABLISHED] (Irish Times)

That is the gift the fiction gave us, and it is a real one: attention to the specific. Not attention in general — attention to this callus, on this hand, which is not like other calluses. Almost everything good in investigative practice descends from that habit, and no machine yet built has replaced it.

Then there is the poison.

The universe was rigged

In sixty stories, Holmes is wrong four or five times, and never catastrophically. This is not a statement about Holmes. It is a statement about Doyle, who wrote both the clue and the world it pointed to, and who therefore never had to face the situation every real investigator faces every week: a clue that is perfectly clear, perfectly consistent, and points at an innocent person.

The word for what Holmes does is not deduction. Deduction runs from general premises to a conclusion already contained inside them; if the premises hold, the conclusion cannot fail. Nothing Holmes does has that structure. He observes something surprising and proposes the explanation that would, if true, make the surprise unsurprising.

Charles Sanders Peirce named that move abduction — reasoning from a surprising fact to the hypothesis that would explain it. You come home, the lawn is wet, you conclude it rained. Not certain. Best available. [ESTABLISHED] (PhilArchive; Acta Analytica 2024)

The philosophy literature is unambiguous that detectives are the standard textbook example of abductive reasoning, and that the stories' in-universe use of "deduction" is Doyle's label rather than a claim about logic. [ESTABLISHED] (ResearchGate, The Abduction of Sherlock Holmes)

Why does the word matter? Because a deduction is finished and an abduction is provisional. The first admits no new evidence; the second is obliged to. A detective who believes he is deducing has given himself permission to stop.

Peter Lipton sharpened the point into the most useful distinction in this book. Among competing explanations, some are lovely — elegant, unifying, satisfying to hold — and some are likely — probably true. These are not the same property, and human beings systematically mistake the first for the second. [ESTABLISHED] (Synthese)

Every wrongful conviction and every failed deployment in the chapters ahead is, underneath, a lovely explanation that was not likely, held by people who could not tell the difference because nothing in their process required them to check.

I am not going to develop the logic further here. The companion volume in this library, The Science of Deduction, does that work properly — abduction, base rates, the full anatomy of how a mind fails. This book takes the correction as given and gets on with the job.

Doyle himself did the job, and it looked nothing like Holmes

The part of the legend that deserves more attention than it gets is that Conan Doyle twice worked real cases, and both times the method was slow, documentary, and adversarial rather than brilliant.

George Edalji, a solicitor, was convicted in 1903 of maiming animals. Doyle took the case up and argued that Edalji's severe astigmatism made the crime, committed at night across rough fields, physically implausible. Edalji was released after roughly three years of a seven-year sentence. The public scandal around the case was one of the pressures that produced England's Court of Criminal Appeal in 1907 — an institution, not an arrest. [ESTABLISHED] (Irish Times; ACS Bulletin)

Then Oscar Slater, convicted in 1909 of murdering Marion Gilchrist in Glasgow. Doyle pushed, funded and argued for nearly two decades. The conviction was quashed in 1928. [ESTABLISHED] (same sources)

Nineteen years. There is no chapter in the canon where Holmes takes nineteen years. The real work of exoneration is paperwork, persistence, and public pressure — and Doyle, who invented the fantasy, spent his own investigative life on the unglamorous version.

The method is older than the English, and it was written for people with no training

The oldest surviving forensic manual is Chinese. Song Ci compiled Xi Yuan Ji Lu — "Collected Cases of Injustice Rectified," often rendered The Washing Away of Wrongs — in 1247. It teaches coroners and local magistrates, explicitly assumed to have no formal training, how to distinguish murder from suicide from accident: how a drowning body differs from a body put in water after death, how strangulation marks differ from a hanging, how to read a stab wound. It predates the first European work of forensic medicine by roughly three and a half centuries. [ESTABLISHED] (Wikipedia; Mütter Museum)

Note what Song Ci was doing: not equipping a genius, but raising the floor for the ordinary official who would otherwise guess. That is what a manual is for, and it is what this one is for. The history of investigation is not a history of great minds. It is a history of procedures that made average practitioners less wrong — Vidocq's criminal-record files in Paris from 1811, Hans Gross's 1893 Handbuch fusing law, medicine, chemistry and photography into a single discipline, Bertillon's measurement cards, Vucetich's fingerprint classification. [ESTABLISHED] (Britannica, Hans Gross; Public Domain Review, Vidocq)

Two of the stories you have probably heard about that history are wrong, and both are wrong in the same direction — toward a tidier narrative.

"Every contact leaves a trace" is not a sentence Edmond Locard wrote. It is a later slogan compressing his considerably more hedged, conditional writing across the 1920s and 1930s — sometimes the criminal leaves traces, sometimes he carries them away. Locard never stated transfer as an absolute law, and the absolute version is the one that gets quoted at crime scenes. [ESTABLISHED] (NIH/PMC; Wikipedia)

Fingerprinting's first criminal conviction was Argentine, not British. Francisca Rojas, 1892, Necochea — a bloody print on a doorpost, classified by Juan Vucetich's system and used by the police official Eduardo Álvarez. A full decade before the 1902 English case that Anglophone histories usually name first. [ESTABLISHED] (Gale Academic OneFile)

And a third, told mainly for what it teaches about how such stories form. The Will West case of 1903 — two unrelated men at Leavenworth with supposedly identical Bertillon measurements — is universally told as the moment anthropometry died and fingerprinting won. The historian Simon Cole argues the popular version was shaped and circulated by fingerprinting's own advocates; the real discrepancy included a foot-length difference of around seven millimetres, outside Bertillon's own tolerance, which clerks had glossed over. The failure was human, not methodological. [CONTESTED] (Simon Cole, Suspect Identities, Harvard UP; Hyperallergic)

A profession's founding myths are written by whoever won. Keep that in mind later, when a newer technology arrives carrying its own set of origin stories.

What the fantasy is actually for

None of this is a reason to be embarrassed about loving the stories. The passion they produce is real and it is the correct passion: a preference for finding out what is true over being told. People who come to investigation through Holmes tend to arrive with the right instinct and the wrong model, and the fix is not to lose the instinct.

The fix is to notice what Holmes was actually given that no real investigator has. He was given a world where the clue meant one thing. What he was not given — and what you have — is a machine that can hold ten thousand cases in view at once, so that you never have to bet everything on the one clue you happened to notice first.

That is the trade this book is about. Holmes narrowed brilliantly because he had to. You do not have to, and Part II is about what becomes possible when narrowing stops being forced on you by the limits of a human attention span.

The Move — Name it an abduction, out loud. When you form a theory of a case, write the sentence: "My best explanation is X." Not "the evidence shows X." The phrasing is not humility for its own sake — it is a structural reminder that the conclusion is provisional and owes a debt to every alternative you have not yet killed. Then, before you look for anything supporting X, write the two rival explanations you would have to eliminate. This costs four minutes and it is the cheapest error-prevention available to you.

What would change my mind: if controlled studies showed that experienced investigators who label their inferences as provisional perform no better than those who do not, this Move is ritual rather than method. To my knowledge that specific comparison has not been run — the supporting evidence is indirect, coming from research on falsification prompts rather than from labelling itself. [ABSENT]