Strategy & Reasoning Study Portal

A: Because being correct is a demonstration and getting a decision is persuasion, and they answer different questions. An audience is simultaneously asking whether the argument holds, whether your judgement can be trusted, and why this matters now — Aristotle's logos, ethos and pathos — and a proposal that answers only the first is incomplete rather than merely unlucky. In practice the two most common gaps are a missing cost of delay, so a correct proposal with no urgency loses to a worse one with a deadline attached, and a missing track record, so the audience has no basis to weight your judgement against someone else's. The deeper cause is usually an enthymeme: real arguments leave their shared premise unstated because stating it is tedious, and the argument fails silently when that premise is not in fact shared. "We should migrate to X" often rests on an unspoken "our current pain comes from the platform", and if the other person does not believe that, then more evidence for the conclusion is not just unpersuasive, it is irrelevant to them. So the specific move is to surface the hidden premise and test it directly — ask what would have to be true for this to be the right call — rather than adding evidence to a conclusion whose foundation is disputed. That single question resolves more disagreements than any amount of additional data.

A: A decision is judged by what was knowable and what process you used at the time; an outcome is what actually happened, and it is influenced by things outside your control. They come apart in both directions: a well-made decision can produce a bad outcome through bad luck, and a reckless decision can produce a good one. An interviewer cares because it is a direct probe of whether your future estimates are trustworthy — if you evaluate your past work purely by results, you will learn the wrong lessons from both your successes and your failures, and you will systematically misjudge risk. It is also why the two obvious answers to a failure question both score badly: blaming circumstances reads as no ownership, while taking total responsibility for genuinely uncontrollable events reads as poor judgement rather than humility. The strong answer applies the Stoic control dichotomy explicitly — name what was outside your control, name what was inside it, state the decision you would make differently with the information available at the time, and describe the change that catches the same class of problem earlier. That structure is also just an accurate blameless postmortem, which is not a coincidence: blamelessness works precisely because it separates the decision from the outcome, so people report what actually happened instead of defending themselves.

A: That a derived number can be wrong in direction, not merely in magnitude, because it comes from a model of the data rather than from the data. The arithmetic assumed the keys would partition the corpus roughly evenly, and real distributions were skewed, so a small number of heavy blocks dominated the mean — and those heavy blocks were exactly the region the estimate had no visibility into. The result was not "a bit more than a hundred", it was nearly two and a half times larger than the thing the change was supposed to improve on, which inverts the decision rather than adjusting it. Archimedes made this distinction explicit in The Method: he described the mechanical, heuristic process by which he actually discovered his results, stated plainly that it was not a proof, and then proved them again rigorously — keeping the method of discovery separate from the method of proof. The modern discipline is the same and is usually skipped: label every quantity measured, derived or assumed, and keep the label attached as the number moves between documents, because an unlabelled estimate becomes indistinguishable from a measurement after it has been copied twice, and at that point a real decision rests on a number nobody ever took. The practical test is to ask which distribution an estimate depends on and whether anyone has looked at it; if nobody has, treat the direction as unknown and run the cheap experiment before designing around the figure.

A: By changing the ground rather than winning on it. Sun Tzu's useful claim is that the outcome is largely settled by position and information before any engagement, and that the best result is the objective reached without the fight — so the first question is whether the disagreement can be dissolved instead of resolved, by changing the interface, splitting the deliverable, sequencing the work so both things become possible, or reframing it as a decision the data can settle. Where it genuinely is a disagreement, the move is to choose the terrain: not litigating it in a forum where you lack context or standing, and moving it to one where evidence decides rather than seniority. Then attack premises rather than the conclusion, because people defend conclusions and readily abandon premises they had not noticed holding — "what would have to be true for this to be right" invites them to examine their own reasoning instead of defending a position in public. All of that depends on ethos: premise-level questioning reads as an attack unless it is clearly established that you are testing the argument and not the person, which is why acknowledged uncertainty and a record of admitted error make the same words land completely differently. And the senior framing that candidates under-report is that the confrontation you removed never appears anywhere — it produces no story, no visible win, and it is still the better outcome.