Sherlockian Way of Thinking
For each method, read the “Scenario” carefully. Under “Your Prompt,” draft the best prompt you can. At the end of the chapter, compare with the sample prompts and design notes, and then revise your prompt once more.
Scenario. You operate an online learning platform. Monthly churn has suddenly risen by 20%. There were no price increases or feature changes. You must determine why.
Your Prompt:
(Write here.)
Design Note.
Direct the model to reconstruct backward from the result: request 3 causal pathways, each as a brief chain (trigger → user experience → decision). Ask for discriminators (what data separates Path A from B) to guide verification.
Scenario. You manage a food-delivery service. After holding at 4.8, the average review score fell to 3.9 over the past two weeks. Diagnose the likely causes.
Your Prompt:
(Write here.)
Design Note.
Require timeline-sensitive sequences (e.g., staffing changes, weather, local events) and customer-journey “mini-scenes.” This enforces causal coherence while making failure modes concrete and auditable.
Scenario. You’re a corporate communications officer. A headline reads: “Midnight Blackout in City Center… Subways Paralyzed, Hospitals Hit Hard.” There is no article text. Reconstruct the likely sequence of events.
Your Prompt:
(Write here.)
Design Note.
Ask for a stagewise progression with cause, time window, and key actors per stage. Retroduction excels when the output is a temporal chain with causal links (not a static list).
Scenario. You know only the last sentence of a short story: “He smiled quietly and hung up.” What must have occurred for this ending to make sense?
Your Prompt:
(Write here.)
Design Note.
Request three prerequisite beats (setup → complication → pivot) that render the final line inevitable. Include emotional transitions to maintain narrative logic, not just plot mechanics.
Sample Prompt.
“Monthly churn on our learning platform rose by 20% month-over-month. Pricing, content, and features did not change. Suggest three plausible causes, and for each, outline a short scenario flow explaining how it could have occurred.”
Design Note.
The result is fixed; the model must construct causal routes that can be checked against data, improving actionability.
Sample Prompt.
“You are the operations lead for a delivery service. The average rating fell from 4.8 to 3.9 within two weeks. Infer three likely causes. For each, sketch a brief, customer-journey ‘mini-scene’ showing what might have happened.”
Design Note.
“Mini-scenes” enforce experiential plausibility and surface diagnostic touchpoints (driver arrival time, food temperature, packaging).
Sample Prompt.
“You are a PR specialist. Based on the headline ‘Midnight blackout in the city center—subways paralyzed, hospitals hit hard,’ infer a four-stage sequence of events. For each stage, include cause, time window, and key actors’ actions.”
Design Note.
Stagewise output ensures chronology, causality, and agency, the core of reliable backward reasoning.
Sample Prompt.
“You are a novelist. If the last line is ‘He smiled quietly and hung up,’ reconstruct the three stages that would naturally lead to this moment. Include the protagonist’s emotional shifts and the conversational context. The ending must feel inevitable.”
Design Note.
By demanding inevitability, you force the model to reconcile character motivation with plot outcomes—rigorous retroductive craft.