OpenIDEA (@OpenIDEAae) header
OpenIDEA (@OpenIDEAae) profile picture

Most failures are decision failures. One essay a week on the discipline of deciding well: judgment, choice, strategy. The Decision Papers — a book in progress.

Joined May 2025
Photo by OpenIDEA (@OpenIDEAae)Photo by OpenIDEA (@OpenIDEAae)
Authority belongs to the jurisdiction where the question, the risk and the skill sit together, not to rank. Each consequential call therefore needs one rightful owner, named before the pressure arrives.
5
Case 5: Crime Scene Averaged Into a Rate The first report came from a nurse: an infusion pump had shut down mid-dose, and no alarm had sounded. The manufacturer checked its fault records. Over three years the failure rate had stayed well below the recall threshold, so the incident looked like the rare event that rate predicted. The team proposed a software update and a notice to clinicians. On the numbers it had, that was sensible. Two weeks later, a second report arrived from another hospital, then a third. A component supplier had changed in the same quarter the failures began. The historical record described a casino with stable odds. This was a crime scene: a one-off change, incomplete data, too few cases to estimate a rate. The team had reasoned as if the past described the odds, and never asked whether the ground itself had changed. The better move was to treat each incident as evidence to investigate, not an observation to average, and to act on the most cautious reading until the supplier question was settled. Call it the average-the-one-off error: a crime scene cannot be averaged into a rate.
16
Case 4: Casino Mistaken for Courtroom The insurer’s warranty clause was written in the language of certainty: the product will not fail above a stated rate. The rate came from five years of claims data on home appliances. The actuaries checked it twice, and legal turned it into a contract term. Sales grew quickly, because the warranty looked like a guarantee. Then a component supplier changed its process, and the defect rate doubled within a quarter. The historical figure had been a fair estimate of a casino with stable odds. The supplier change altered those odds without warning. The contract still described the old odds as if they were law. The company now owed refunds it had not priced, and it could not change the clause without admitting that the rate had been a bet, not a rule. A statistic that held for five years had been written down as a fact that would always hold. The insurer had treated a casino as a courtroom, and the courtroom was now enforcing yesterday’s odds. Call it the casino-courtroom error: odds that repeat are not rules that hold.
32
The Terrain Card, before anyone presents: 1. Name the ground: courtroom, casino, or crime scene, and the confidence it allows. 2. Analogy or rule of thumb? Call it a hypothesis, and ask whether the bet can be undone. 3. Each new caveat is a cost, not a comfort.
11
Every new caveat is a cost, not a comfort. A model that needs three fresh excuses to survive contact with reality is not sturdier. It is more fragile.
36
Case 3: The Complexity Check The ridership forecast had missed three times in a row, and each miss came with a new explanation. The first was blamed on fuel prices, the second on a rail strike that ended before the forecast window opened, and the third on remote working, which the model never included. Each brought a new adjustment. The model now had twenty-three inputs and a band wide enough to contain almost any outcome. A junior modeller asked the question nobody had asked in three years: how many inputs improve accuracy on data the model was not tuned to? The answer was four. The other nineteen had been added after each miss. They explained the past beautifully and predicted the future no better than a trend line. Every new qualifier had been sold as rigour. Each one was another assumption that had to be true, and most had quietly stopped being true. The model had not grown sturdier with each excuse. It could survive any outcome, which meant it could tell the council nothing about the one it would face. Call it the complexity check: every new qualifier is a cost, not a comfort.
46
Objection: if every decision needs a terrain ceremony, we drown easy choices in process. True for most decisions. The terrain call is a circuit breaker, not a gate on every door. It earns its place when the ground is contested, or the room is about to pick the wrong mode.
9
Case 2: The Smuggling Check The roadmap slide had one line of evidence: the rival did it, and the rival is winning. A mid-sized software company wanted an AI tutor for its learning platform. A competitor had launched a conversational assistant six months earlier and was adding users fast. The product lead proposed copying the design: the onboarding, the personality, and the bundled pricing. Their success came from this design, so ours would too. The analogy looked clean. Both sold to schools, and both lived on subscriptions. Nobody checked whether the rival’s growth came from its design, or from a distribution partnership with a large district network, or from a licence bundle the company did not have. Eighteen months and a large engineering budget later, the assistant shipped. Engagement was flat. Nobody could tell whether the design had failed or the analogy had never applied, and the bundled pricing was already public, so the bet could not be quietly withdrawn. Call it the smuggling check: an analogy is a hypothesis, and a bet you cannot undo should never rest on one.
1
39
Three kinds of ground, three levels of certainty. Courtroom: closed systems, where rules hold. Casino: repeatable odds, where patterns predict. Crime scene: one-off and unrepeatable, where only a hypothesis fits. Know which one you are on before the slides open.
1
37
Mode mismatch: the wrong kind of reasoning for the problem, because nobody chose the right kind. The most common version is a crime scene mistaken for a casino. Revenue drops, last quarter’s trend gets projected, and a confident forecast describes a future that no longer exists.
15
Three kinds of ground. A courtroom: closed rules, where certainty is allowed. A casino: repeatable odds, where patterns predict. A crime scene: one-off and noisy, yielding a suspect to test, never a verdict. The ground, not the speaker’s confidence, decides what survives.
20
Before any forecast, ask one question: has this exact situation repeated often enough to trust a pattern? If the honest answer is no, you are standing on a crime scene. A crime scene yields a suspect to test, never a verdict to announce.
34
Case 1: crime scene mistaken for casino The revenue line dropped 11% in the second quarter, and by Thursday the finance team had a forecast on the wall. The company sold project software to mid-sized firms and had grown steadily for six years. The dip was the first of its kind. A new competitor had launched, a pricing change had gone live, and the largest reseller had rewritten its terms. Five explanations were on the table, and each came with a story. The analyst did not argue with any of them. She did what the spreadsheet invited: she fitted six years of history to a seasonal trend and projected a steady recovery. The board approved a hiring plan on the strength of it. The quarter did not recover. The forecast had treated a one-off as a repeat, with four causes, none of them tested, each changing the ground under the model. Six years of history had trained the team to treat revenue as a casino. This quarter was a crime scene, and nobody had said so. Call it the crime-scene mistake: a one-off forecast delivered with casino confidence.
36
Water shapes its course according to the nature of the ground over which it flows. Sun Tzu Reasoning works the same way. Name the ground before the argument.
33
In the meeting, a senior voice asks for the ROI, and a real question mark gets dressed up as a hard number. Someone else asks for the trend, and a one-off is forced to behave like a pattern. The essay above asks the question nobody did: what ground are we standing on?
21
Case 5: heuristic trap Every regional sales director at the chain carried the same rule, and it had earned its place. Discount at quarter-end, because the warehouse needs to clear and customers come back. In the core market it had worked for years. Quarter-end sales jumped, stock cleared, and bonus targets were met. The rule travelled. A new team in a price-sensitive market applied it exactly as written. Within two quarters, margins were negative. Customers there had learned to wait for the sale, so full-price sales dried up and the discount became the normal price. Nobody had reasoned badly. The team had followed a rule of thumb that had worked somewhere and trusted it everywhere. A heuristic carries no record of the conditions that made it true. It compresses one setting into a habit, and the habit never checks the setting before it acts. The fix was simple: record the conditions under which each rule was learned, and test it in the new market before scaling it. A rule that was gold in one setting can bankrupt you in another.
39
Which logic? And how much weight has it earned? A room that never asks will keep mistaking a confident guess for a settled fact.
30
The objection: labelling every sentence would paralyse a room. Agreed, which is why you save it for arguments carrying real weight. Then say the mode and what it buys you: this holds while the rule holds, or this is the leading explanation, awaiting a test.
9