Chapter 3: Sherlock Holmes Makes a Diagnosis

What this chapter does. The deductive method in project management. How to learn to see the invisible: seven diagnostic signs of a dying project, each verified against real catastrophes that could have been prevented.

Southsea (Portsmouth), 1 Bush Villas, Spring 1886

Arthur Conan Doyle sits in the waiting room of his medical practice in Southsea, a suburb of Portsmouth. He is twenty-six. Edinburgh is behind him, the diploma is in hand, the practice is set up. Income the first year: £154; the practice will grow to £250, and the eight-year ceiling will stop at £300. Not poverty, but not a Harley Street surgeon either. Modest provincial work, and by evening — silence.

In that silence Doyle writes stories for cheap magazines. Around three pounds a story from London Society — a supplement, not a windfall. What weighs more is what will happen at the end of the year. In three weeks that spring of 1886 he will write a short novel about a consulting detective named Sherlock Holmes: A Study in Scarlet. Ward Lock & Co will buy the manuscript on the twentieth of November of the same year for £25 — all rights. Twenty-five pounds, one payment, no royalties. So the most profitable literary character of the twentieth century enters literature.

Doyle writes with an eye on his Edinburgh teacher, Professor Joseph Bell. Bell would diagnose a patient before the patient opened his mouth: from gait, from clothing, from the skin of the hands. Six years later, in 1892, Doyle would write to Bell: "It is most certainly to you that I owe Sherlock Holmes… round the centre of deduction and inference and observation which I have heard you inculcate I have tried to build up a man."¹ᵃ

So a character is born who, one hundred and forty years later, will explain why most corporate projects fail before they are officially announced.

Doyle does not know he is inventing a methodology for the diagnosis of systemic failure. He writes a short novel in three weeks and sells it for twenty-five pounds.

Baker Street, 221B — Sunday morning

"You see, but you do not observe," Sherlock Holmes will say to Dr. Watson in "A Scandal in Bohemia" five years later, in 1891.

"The distinction is clear. For example, you have frequently seen the steps which lead up from the hall to this room."

"Frequently." "How often?" "Well, some hundreds of times."

"Then how many are there?" "How many? I don't know."

"Quite so! You have not observed. And yet you have seen. That is just my point. Now, I know that there are seventeen steps, because I have both seen and observed."¹

Holmes is the third literary archetype our investigation needs: the method of recognition. Verne gave us the pattern of creation; Shelley the pattern of flight; Holmes gives the instrument without which the first two are useless.

Because the most dangerous stage of a disease is not the agony. It is the period of false wellness, when everyone is convinced the patient is healthy.

The previous chapter showed what happens when a creator abandons his creation. FBI Sentinel, Healthcare.gov, BBC DMI: in each case the signs of catastrophe were visible long before the explosion, but someone saw them, and no one observed.

This chapter is about observation. About how to tell a healthy team from a dying one, a living process from a ritual, adaptation from imitation. Seven diagnostic signs — Holmes's seventeen steps — that everyone sees and no one counts.

Deduction Against Intuition

Holmes was not a genius in the Romantic sense. He was a systematic observer. His method is not a flash of insight but a disciplined sequence: gather the data, eliminate the impossible, and whatever remains, however improbable, must be the truth.²

"I never guess," Holmes says in The Sign of the Four. "It is a shocking habit — destructive to the logical faculty."²

Compare this with how decisions get made in corporate projects: the manager's intuition, the committee's consensus, extrapolation from last quarter, the loudest voice in the room. Anything except systematic observation.

Holmes offers the alternative: trust neither the reports nor the metrics nor the presentations. Observe what people do, not what they say. The first principle of diagnostics: data are collected at the scene of the crime, not in the detective's office.

Here are seven signs Holmes would notice on the first round.

Sign I: The Dog That Did Not Bark

In the story "Silver Blaze," Holmes calls the inspector's attention to "the curious incident of the dog in the night-time." The inspector objects: "The dog did nothing in the night-time." Holmes: "That was the curious incident."³

The dog did not bark because it knew the thief. The most important signal was not in what happened, but in what did not happen.

In corporate diagnostics, silence becomes the most alarming symptom. When no one raises problems at retrospectives; when code reviews contain no serious remarks; when every sprint closes "on plan," and development velocity stays as steady as the heart-monitor of a corpse.

Literature knew this green dashboard a century and a half before the first sprint. The plot of Dead Souls (1842) Pushkin gave to Gogol out of a real story: in the Transnistrian town of Bendery, for years, not a single death was registered. The trick was simple — the names of the dead were not struck from the rolls but reassigned to the newly arrived runaway serfs. The statistics showed an immortal population; in the column for "loss," a spotless zero. That is exactly what a team looks like when it has learned not to enter failure into its reporting: the metric proclaims perfect velocity because everything that would spoil the picture has been carefully removed. Chichikov bought souls that did not exist, on a registry where they were still listed as living, and the whole mechanism held together because the paper had long ago stopped agreeing with the life. A dashboard that has glowed green for half a year is Chichikov's census: a registry of the living in which half are no longer breathing.

The same dog that did not bark in the stables at King's Pyland stays silent at the stand-up. Holmes points Inspector Gregory to "the curious behavior of the dog on the night of the crime," and Gregory is puzzled: "The dog? But it did nothing." "That was the curious thing," Holmes replies. Eighteen months without a single blocker on the board — that is the hound that has kept its voice: not proof that all is calm, but proof that whoever carries the anxiety has chosen silence. A team in which no one has brought bad news for half a year has either reached perfection (a probability Holmes would rate soberly) or has learned that a blocker is a confession of incompetence and has learned to suffer quietly. In either case, the diagnostician should be alarmed not by what is heard, but by what is not.

Healthy teams argue. They break sprints. They raise red flags. Silence means one of two things: the team has achieved perfection (probability near zero), or people have stopped believing they will be heard.

Knight Capital, August 1, 2012. One of the largest market makers on the NYSE. At 9:30 a.m. Eastern the firm activated a new trading algorithm. In forty-five minutes the system took erroneous positions worth about seven billion dollars. Pretax losses came to $440 million; total write-downs brought the figure to roughly $460 million.⁴ᵃ The firm lost more than all its cash on hand — $365 million on the books the night before — which is why it needed $400 million in rescue financing over the next seventy-two hours.⁴

The SEC investigated and published its findings in 2013. The key detail: at deployment, one of eight servers did not receive the update. On that server the old code remained — a test module called "Power Peg," unused for nine years, since 2003, but never deleted from the system. When the server received trading signals, Power Peg began executing them under the logic written for the test environment. The result: chaotic trades at market prices in enormous volumes.⁴

The dog that did not bark: no one asked why nine-year-old dead code was still in production. Everyone saw Power Peg in the codebase, and no one observed.

The SEC established that Knight Capital lacked adequate deployment-control procedures and did not test the full system before launch.⁴ In Holmes's language: the crime was not committed by the algorithm. The crime was committed by silence — years of it — about the technical debt everyone saw and no one named.

Sign II: The Mismatch of Evidence

Holmes never forms a theory before he has the facts. "It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts."⁵

In corporate diagnostics this shows up like so: the dashboard is green, but the users complain; velocity rises, but time-to-delivery of features lengthens; the team reports high morale, but the best people are leaving.

When the metrics contradict the reality, Holmes always trusts the reality.

Target Canada, 2013–2015. In March 2013 Target Corporation, the second-largest discount retailer in the United States, entered the Canadian market. In the first year 124 stores opened; by the end of 2014 the chain had grown to 133 locations. The combined losses of the Canadian venture exceeded seven billion US dollars, including a pretax write-down of $5.4 billion.⁶

By January 2015 Target Canada was closed. Completely. All 133 stores. 17,600 employees lost their jobs.⁷

The Canadian Business investigation found that the product-information system (SAP) contained data so dirty that stores could not order goods correctly. Sizes, prices, descriptions — all of it error-riddled. Shelves stood empty or filled with the wrong stock, and customers came, saw the emptiness, and left forever.⁸

The mismatch of evidence: corporate reporting showed that all 133 stores were open and functioning. Technically that was true — the stores stood, the registers worked, the staff came in. But the shelves were empty because the system that should have filled them was lying about what stock existed, in what quantity, at what price.

Holmes would say: you are looking at the dashboard, and you are not observing the shelf. The metric "133 open stores" is a clue that leads to a dead end. The real clue is Aunt Janine from Montreal, who came for towels and left with nothing.

Target Canada lost seven billion dollars not because management was incompetent, but because the reporting system was telling a story that did not match physical reality, and no one wanted to go to the store and look at the shelves. Data replaced observation, and the dashboard replaced Baker Street.

Sign III: Conway's Law as Forensic Evidence

In 1968 Melvin Conway published an observation that became a law: "Organizations which design systems are constrained to produce designs which are copies of the communication structures of these organizations."⁹

Holmes would have appreciated the elegance. If you want to understand what is wrong with a product, do not look at the code. Look at the organization chart. The code is only the shadow the organization casts on the monitor.

Conway's Law is not a theory. It is a piece of forensic evidence — a fingerprint pointing without error at the perpetrator.

If your application programming interface consists of seven incompatible endpoints, there is a strong chance the company has seven teams that do not speak to one another. If the user journey passes through three different interfaces with three different designs, three different departments run them, with three different heads. If deployment requires coordinating five repositories, look for five managers, each defending his territory.

The product becomes a mirror of the organization, and a broken mirror unfailingly reflects a broken organization.

Back to Target Canada. The SAP system, deployed to run the Canadian stores, was configured by a team in Minneapolis for a Canadian reality that team did not understand. The Canadian tax regime, the bilingual labeling, the metric system — all of it was "just configuration" for people who had never bought milk in liters.⁸

Conway's Law predicted the failure. An American organization built a system that reflected an American understanding of the Canadian market. The organization's structure American, the data's structure American, the understanding of the customer American. Melvin Conway could have delivered the diagnosis in 1968, forty-five years before the first store opened.

Sign IV: The Missing Patient

Holmes always begins with the question: who is the victim? Not "what went wrong," but "who suffered." This is not sentiment. It is method: the victim points to the motive, and the motive to the perpetrator.

In corporate diagnostics the victim is the user, and the most alarming sign arrives exactly when the user stops being talked about.

Listen to the conversations on your team. If the words "user," "customer," "buyer" turn up less often than "process," "framework," "methodology," the patient has gone missing, and the team is treating itself, having forgotten the sick.

Gerald Weinberg showed as early as 1971 that if developers do not meet users, they inevitably build a system for themselves.¹⁰ Not out of malice — out of absence of data. Holmes cannot solve a case without visiting the scene of the crime. A developer cannot build a product without seeing the user.

Lidl and SAP: a five-hundred-million-euro system for a nonexistent user. In 2018 the German discounter Lidl halted the implementation of an SAP inventory-management system after seven years of work and expenditures the press estimated at €500 million.¹¹

The reason, per Handelsblatt (July 2018): Lidl was trying to make SAP work by Lidl's purchasing logic, while SAP demanded that Lidl restructure itself around SAP's logic. Seven years of compromises that satisfied neither the system nor the business.¹¹

The missing patient: the Lidl warehouse worker who decides each morning how many yogurts to order. That person is the end user of a half-billion-euro system, and nobody asked him.

Instead, they asked the SAP consultants (what can the system do?), the finance directors (what is the budget?), and the IT architects (what is the architecture?) — all the right questions, addressed to the wrong people. Holmes never begins an investigation by interrogating the suspect's attorneys. He goes to the scene of the crime and speaks with the witnesses.

Sign V: The Evidence of Time

Holmes records not only what happened but when. The chronology is the skeleton of the case.

In project diagnostics, time tells more than any report. How much time passes from idea to prototype, from prototype to user, from bug discovery to bug fix?

The DORA (DevOps Research and Assessment) study, published by Google Cloud in 2023, classifies teams into four performance levels. Elite teams deploy on demand, several times a day; their lead time — from commit to production — is less than one day. Low-performing teams deploy once a month or less; lead time is one to six months; recovery from an incident takes a week or more.¹²

The difference is not a percentage. It is an order of magnitude.

But the most telling metric is not the deploy speed. It is the time to recover from a failure. Elite teams recover in an hour. Low-performing teams take a week or more.¹²

Holmes would say: the speed at which an organization corrects its errors tells you more than the speed at which it creates something new. Creation is ambition. Correction is character.

Knight Capital recovered in forty-five minutes, and too late — $440 million was lost beyond recall (pretax; the final write-down reached roughly $460 million). Healthcare.gov took two months to recover. BBC DMI never recovered. The recovery timeline is a reliable indicator of organizational health.

If your team needs a week to fix a critical bug, that is not a technical diagnosis. It is an organizational one. Somewhere between discovery and fix sit people who approve, sign off, plan, defer, and wait — instead of fixing.

Sign VI: The Watson Effect

Watson is not merely a narrator. He is a mirror in which Holmes sees his own blind spots. Watson asks "stupid" questions that turn out to be the most important; he notices the human where Holmes sees only the logic.

Every team needs a Watson — a person unafraid to say, "I don't understand why we are doing this." Not out of incompetence. Because if he does not understand, probably nobody understands, and the rest are too embarrassed to admit it.

When the Watson leaves a team — the person who asks the awkward question — what remains is an echo chamber. Everyone agrees, everyone is certain, and nobody checks the base assumptions.

Frederick Brooks described the effect in The Mythical Man-Month: "Adding manpower to a late software project makes it later."¹³ But Brooks's less-quoted observation matters more: large teams lose the capacity to communicate, because the number of channels grows as n(n − 1)/2.¹³ A team of five has ten channels. A team of fifty has 1,225. And at 1,225 channels, no one can be Watson to everyone.

In a small team, Watson is natural. Anyone can ask, "Why?" In a large organization, "why" requires a meeting, a presentation, a business case, and approval. By the time the question finally gets asked, the answer is already stale.

Target Canada was lost in its 1,225 channels. Knight Capital was lost in the silence of a team that saw the dead code and did not ask the question. Lidl was lost in a seven-year project where nobody asked the warehouse worker, "Does this work for you?"

Sign VII: Physical Evidence — the git log

Holmes trusts evidence, not testimony. Testimony is subjective, biased, forgetful. Evidence is objective.

In a software project there is one piece of evidence that cannot be faked: the git log.

The commit history tells a truth no stand-up will tell — who really writes the code, how large the changes are, how often master gets broken, who reviews and whether reviews happen at all. None of it needs to be asked of people. It is visible in a journal that has not learned to answer the way the manager wants.

Commit frequency is the easiest to read. If the last commit in a critical module is dated six months back, the module is dead, whatever the quarterly report says about it. If, on the other hand, the commits come every hour with messages like "fix," "fix2," "fix final," "fix final final" — that too is a diagnosis, just the opposite one: the team is in a panic. Commit size adds the next layer. Average commit of 2,000 lines is not iteration; it is a big bang disguised as regular work. A commit of three lines means the reverse: the team is afraid to change anything serious. Healthy size sits between the extremes.

Authorship. If 80 percent of commits belong to one person, the bus factor is one. That person leaves and the project stops. Not because the others are incompetent — because they have stopped touching the code they consider "someone else's."

Commit messages. "WIP," "asdf," "tmp," "please work" — a cry for help encoded in metadata. Compare with "Refactor auth module to support OAuth2 flow." The second team knows what it is doing and why.

Holmes would not guess whether the project was healthy. He would open the git log and know the answer in five minutes.

Against Diagnostics

A thesis to enrage the management consultants: most diagnostic frameworks are part of the disease, not the cure.

Maturity models (CMM, CMMI), Agile assessments, organizational-readiness scorecards — they all propose one approach: rate yourself on a scale of one to five, identify areas for development, put together an improvement plan.

Holmes would laugh. Asking a sick organization to diagnose itself is like asking the suspect to conduct his own interrogation.

Organizations lie to themselves not out of malice but out of systemic distortion. Leadership overestimates health, because otherwise action would be required. Middle management understates problems, because otherwise it gets blamed. Rank-and-file employees stay quiet, because otherwise they get fired. Each level filters the information, and by the time it reaches the decision-maker, only a groomed shadow of reality remains.

Nokia fell exactly this way. Vuori and Huy called it "distributed attention."¹⁴ When an organization is so large that no single person sees the whole picture, responsibility dissolves. Everyone owns a fragment; nobody owns the whole.

Holmes never relied on the client's self-assessment. He observed. He counted the steps, noticed the mud on the boots, paid attention to the dogs that did not bark.

Real diagnosis has to be external and hard, based on facts, not on impressions.

From Diagnostics to Treatment

Three chapters have accumulated three different instruments. Verne showed how to build living systems through increments, feedback, and adaptation. Shelley showed what happens to those systems when they are abandoned. And Holmes gave the way to recognize the disease before the autopsy.

But a diagnosis is not a treatment. Knowing the project is sick is necessary and not sufficient. Holmes solved crimes; he never prevented them. He arrived after the deed was done. His method works for the observer who reaches the crime scene, not for the builder who wants there to be no crime.

The next act asks a different question: not "how did we get here," but "how do we get out." Not the literary archaeology of failures, but the living practice of teams that manage to stay alive in the conditions killing their neighbors. We leave Baker Street and go where teams have learned to observe themselves.

To be continued. In the next chapter: a librarian from Buenos Aires who described your sprint estimation in 1941, sixty years before it was invented. A lottery in which what is drawn is not money but productivity. A theater in which OKRs play themselves. And teams so absorbed in measuring speed that they forgot to look at where they were driving. Holmes counted seventeen steps. Borges will count your story points.

Footnotes:

¹ Arthur Conan Doyle. "A Scandal in Bohemia" (1891). The Strand Magazine. Canonical English text used directly (the dialogue on the seventeen steps is one of the classic examples of Holmes's method). The Russian edition quotes these lines in Russian translation; the English edition restores Conan Doyle's original.

¹ᵃ Doyle, Arthur Conan. Letter to Joseph Bell, May 1892. Verbatim from the original English (attested via ACD encyclopedia and University of Edinburgh archive references). Full quotation: "It is most certainly to you that I owe Sherlock Holmes… round the centre of deduction and inference and observation which I have heard you inculcate I have tried to build up a man."

² Doyle, Arthur Conan. The Sign of the Four (1890). Spencer Blackett, London. Canonical English text: "I never guess. It is a shocking habit — destructive to the logical faculty." (Ch. 1.)

³ Doyle, Arthur Conan. "Silver Blaze" (1892). The Strand Magazine. Canonical English: "the curious incident of the dog in the night-time." (Full exchange with Inspector Gregory, verbatim from the story.)

⁴ SEC Administrative Proceeding File No. 3-15570, October 16, 2013. "In the Matter of Knight Capital Americas LLC." Canonical English text of the SEC enforcement action used directly for procedural detail; author's summary of the Power Peg incident preserved in body. Fine imposed: $12 million. Deployment-control deficiencies and the reactivation of dormant test code documented in the enforcement order.

⁴ᵃ Loss figures per Knight Capital 8-K filing (August 2, 2012) and subsequent SEC filings: $440 million pretax trading loss; total write-downs and remediation costs reached approximately $460 million. Firm's cash on hand the night before: $365 million (per SEC filings). Emergency rescue capital: $400 million from Jefferies-led investor group (August 6, 2012).

⁵ Doyle, Arthur Conan. "A Scandal in Bohemia" (1891). The Strand Magazine. Canonical English text: "It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts." (The line often misattributed to A Study in Scarlet, in which Doyle uses a shorter variant of the same thought.)

⁶ Target Corporation Annual Report 2014 and The Globe and Mail, January 15, 2015. Total Canadian-venture investment exceeded $7 billion USD; pretax write-down reached $5.4 billion. Canonical English text used directly from public financial disclosures.

⁷ Canadian Press, January 15, 2015. "Target Canada closing: A timeline of the U.S. retailer's disastrous Canadian venture." The 17,600-employee figure comes from Target Corporation's official statement.

⁸ Castaldo, Joe. "The Last Days of Target Canada." Canadian Business, January 2016. Detailed investigation of the SAP-system data problems. Canonical English text of the reporting used directly.

⁹ Conway, Melvin. "How Do Committees Invent?" Datamation, April 1968. Original English formulation of Conway's Law used directly: "Organizations which design systems are constrained to produce designs which are copies of the communication structures of these organizations."

¹⁰ Weinberg, Gerald. The Psychology of Computer Programming (1971). Van Nostrand Reinhold. Observations on the relationship between developers and users. Author's paraphrase from the RU master; canonical English text in Weinberg's book, especially Chs. 3–4.

¹¹ Handelsblatt, July 2018 (primary source for the €500 million figure and the eLWIS project chronology); Computerwoche, 2018; Lebensmittel Zeitung, 2018. Canonical English text of the reporting used directly where available; German-language originals cited for the primary figures. The eLWIS project halted July 2018 after seven years and approximately €500 million (≈ $580–600 million). Lidl reverted to the previous Wawi system.

¹² Accelerate: State of DevOps Report. Google Cloud / DORA, 2023. Canonical English text used directly. Classification of teams into four levels: Elite, High, Medium, Low. Deployment frequency, lead time for changes, mean time to recover, and change failure rate — the four metrics.

¹³ Brooks, Frederick. The Mythical Man-Month: Essays on Software Engineering (1975). Addison-Wesley. Canonical English quotation: "Adding manpower to a late software project makes it later" — Brooks's Law. Communication-channel formula n(n − 1)/2 discussed in Chapter 2.

¹⁴ Vuori, Timo O., and Quy Nguyen Huy. "Distributed Attention and Shared Emotions in the Innovation Process: How Nokia Lost the Smartphone Battle." Administrative Science Quarterly, Vol. 61, No. 1 (2016), pp. 9–51. Canonical English text of the study used directly. Study of the Nokia dissolution and the mechanisms of organizational attention.