{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"slug":"choosing-the-awkward-case","title":"The example that earns its place","url":"https://addaly.com/learn/prompting/choosing-the-awkward-case","position":21,"of":72,"minutes":8,"answer":"An example is only worth its space if a careful person could have got it wrong, so build your set from the real items where you yourself hesitated.","keyIdea":"An example is only worth its space if a careful person could have got it wrong, so build your set from the real items where you yourself hesitated."},"body":{"format":"markdown","text":"## Obvious examples teach nothing\n\nHere is a set somebody wrote to classify support messages:\n\n```\n\"Where is my parcel?\"            → delivery\n\"My parcel hasn't arrived.\"      → delivery\n\"I still haven't got my order.\"  → delivery\n\"Can you refund me?\"             → refund\n```\n\nFour examples, three of which are the same example. The model already knew that \"where is my parcel\" is about delivery; it did not need three demonstrations of the easy case, and it received no information about anything that is actually hard.\n\nAn example is only worth its space if a competent reader could plausibly have got it wrong.\n\n## Where the information lives\n\nThink about your task and find the places where two reasonable people would disagree. That is where your examples go.\n\nFor a support classifier, the disagreements are:\n\n- The parcel arrived broken and the customer wants money. Delivery or refund?\n- The customer is angry but is asking a normal question. Complaint or query?\n- The message is in two languages and half of it is about something else.\n- The message says \"cancel my order\" but it has already shipped.\n\nFour examples, each resolving one genuine ambiguity. That set carries roughly four times the information of the first one, in the same space.\n\n## The three examples worth having\n\nIf you only have room for a handful, pick these:\n\n**1. The boundary case.** The one that sits closest to the line between two answers. Include the label *and*, where it helps, a one-line reason: `→ refund (classify by what they are asking for, not what went wrong)`. A stated reason generalises to cases you did not show.\n\n**2. The case you keep getting wrong.** Go through the outputs that annoyed you. The failure you have corrected three times is a failure your prompt does not cover. Turn it into an example: the input that produced the bad output, plus the output you wanted.\n\n**3. The one that stops the over-produced answer.** Every classifier has a default it reaches for. If everything ambiguous is coming back as `other`, include an ambiguous case labelled something else, so `other` stops looking like the safe choice.\n\n## Where to find real awkward cases\n\nNot by imagining them. Imagined edge cases are systematically wrong, because you imagine the ones you have already thought about, which are the ones your instructions already cover.\n\nTake fifty real items: real messages, real invoices, real rows. Label them yourself, quickly. The ones where you hesitated for more than two seconds are your example set. Your hesitation is the measurement.\n\nThis takes twenty minutes and it is the single highest-value twenty minutes in this course. It also produces something else you will want later: a small set of items with known answers, which is a test set, and lets you check whether any change to your prompt actually helped.\n\n## What to do when you cannot decide either\n\nSometimes you hesitate because the rule does not exist yet. The broken-parcel-refund case has no answer until somebody decides.\n\nThat is not a prompting problem. It is a policy gap that your prompt has revealed, and the right response is to make the decision, write it down in one line, and put it in the prompt as a rule with an example attached. Teams often discover in this exact moment that two people have been handling the case differently for a year.\n\n## The maintenance\n\nAn awkward case stops being awkward once your prompt handles it. It does not stop being useful (it is what keeps handling it), but you should review the set when something changes: a new product line, a new policy, a new failure that keeps appearing.\n\nFour to six examples, each earning its place, reviewed every few months. That is a working example set, and it will outperform any twenty-example collection assembled from cases nobody was unsure about.\n\n## One warning about hard cases\n\nA set made entirely of edge cases has its own distortion: it implies that edge cases are normal. If four out of four examples are awkward, ordinary items start attracting exotic treatment, and you will see the model hedging on a message that needed one word.\n\nSo the working shape is mostly-hard-with-an-anchor: three or four boundary cases and one plainly typical one, placed last. The typical example costs you almost nothing and it re-establishes what the ordinary case looks like right before the model answers."},"check":{"question":"An operations manager builds a five-example prompt for sorting invoices into \"pay now\", \"query\", and \"hold\". After a month, most errors are invoices that should be queried but come back as \"pay now\". What should she add?","options":[{"text":"Two more clear examples of \"pay now\", so the boundary between it and \"query\" is sharper.","correct":false,"tempting":"Adds weight to the category that is already over-produced, which makes the imbalance worse."},{"text":"An instruction to be more cautious and query anything uncertain.","correct":false,"tempting":"A vague disposition rather than a boundary; \"uncertain\" is exactly what the model cannot assess reliably."},{"text":"Examples of every category in equal numbers, to balance the set.","correct":false,"tempting":"Balance helps, but blind balance does not target the specific confusion that is producing the errors."},{"text":"Real invoices that she hesitated over and that turned out to be \"query\", each with a one-line reason for the label.","correct":true}]},"prev":{"slug":"zero-one-and-few","title":"Zero, one, or five: how many examples","url":"https://addaly.com/learn/prompting/zero-one-and-few"},"next":{"slug":"order-balance-and-the-hidden-pattern","title":"What your examples teach by accident","url":"https://addaly.com/learn/prompting/order-balance-and-the-hidden-pattern"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'The example that earns its place', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/choosing-the-awkward-case (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The example that earns its place","courseTitle":"Getting Real Work out of a Model","module":{"index":3,"slug":"teaching-by-example","title":"Teaching by example"},"answer":null,"keyIdea":"An example is only worth its space if a careful person could have got it wrong, so build your set from the real items where you yourself hesitated.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"obvious-examples-teach-nothing","runs":[{"text":"Obvious examples teach nothing"}]},{"t":"p","runs":[{"text":"Here is a set somebody wrote to classify support messages:"}]},{"t":"code","lang":"","label":"Text","text":"\"Where is my parcel?\"            → delivery\n\"My parcel hasn't arrived.\"      → delivery\n\"I still haven't got my order.\"  → delivery\n\"Can you refund me?\"             → refund"},{"t":"p","runs":[{"text":"Four examples, three of which are the same example. The model already knew that “where is my parcel” is about delivery; it did not need three demonstrations of the easy case, and it received no information about anything that is actually hard."}]},{"t":"p","runs":[{"text":"An example is only worth its space if a competent reader could plausibly have got it wrong."}]},{"t":"h2","id":"where-the-information-lives","runs":[{"text":"Where the information lives"}]},{"t":"p","runs":[{"text":"Think about your task and find the places where two reasonable people would disagree. That is where your examples go."}]},{"t":"p","runs":[{"text":"For a support classifier, the disagreements are:"}]},{"t":"ul","items":[[{"text":"The parcel arrived broken and the customer wants money. Delivery or refund?"}],[{"text":"The customer is angry but is asking a normal question. Complaint or query?"}],[{"text":"The message is in two languages and half of it is about something else."}],[{"text":"The message says “cancel my order” but it has already shipped."}]]},{"t":"p","runs":[{"text":"Four examples, each resolving one genuine ambiguity. That set carries roughly four times the information of the first one, in the same space."}]},{"t":"h2","id":"the-three-examples-worth-having","runs":[{"text":"The three examples worth having"}]},{"t":"p","runs":[{"text":"If you only have room for a handful, pick these:"}]},{"t":"p","runs":[{"text":"1. The boundary case.","b":true},{"text":" The one that sits closest to the line between two answers. Include the label "},{"text":"and","i":true},{"text":", where it helps, a one-line reason: "},{"text":"→ refund (classify by what they are asking for, not what went wrong)","code":true},{"text":". A stated reason generalises to cases you did not show."}]},{"t":"p","runs":[{"text":"2. The case you keep getting wrong.","b":true},{"text":" Go through the outputs that annoyed you. The failure you have corrected three times is a failure your prompt does not cover. Turn it into an example: the input that produced the bad output, plus the output you wanted."}]},{"t":"p","runs":[{"text":"3. The one that stops the over-produced answer.","b":true},{"text":" Every classifier has a default it reaches for. If everything ambiguous is coming back as "},{"text":"other","code":true},{"text":", include an ambiguous case labelled something else, so "},{"text":"other","code":true},{"text":" stops looking like the safe choice."}]},{"t":"h2","id":"where-to-find-real-awkward-cases","runs":[{"text":"Where to find real awkward cases"}]},{"t":"p","runs":[{"text":"Not by imagining them. Imagined edge cases are systematically wrong, because you imagine the ones you have already thought about, which are the ones your instructions already cover."}]},{"t":"p","runs":[{"text":"Take fifty real items: real messages, real invoices, real rows. Label them yourself, quickly. The ones where you hesitated for more than two seconds are your example set. Your hesitation is the measurement."}]},{"t":"p","runs":[{"text":"This takes twenty minutes and it is the single highest-value twenty minutes in this course. It also produces something else you will want later: a small set of items with known answers, which is a test set, and lets you check whether any change to your prompt actually helped."}]},{"t":"h2","id":"what-to-do-when-you-cannot-decide-either","runs":[{"text":"What to do when you cannot decide either"}]},{"t":"p","runs":[{"text":"Sometimes you hesitate because the rule does not exist yet. The broken-parcel-refund case has no answer until somebody decides."}]},{"t":"p","runs":[{"text":"That is not a prompting problem. It is a policy gap that your prompt has revealed, and the right response is to make the decision, write it down in one line, and put it in the prompt as a rule with an example attached. Teams often discover in this exact moment that two people have been handling the case differently for a year."}]},{"t":"h2","id":"the-maintenance","runs":[{"text":"The maintenance"}]},{"t":"p","runs":[{"text":"An awkward case stops being awkward once your prompt handles it. It does not stop being useful (it is what keeps handling it), but you should review the set when something changes: a new product line, a new policy, a new failure that keeps appearing."}]},{"t":"p","runs":[{"text":"Four to six examples, each earning its place, reviewed every few months. That is a working example set, and it will outperform any twenty-example collection assembled from cases nobody was unsure about."}]},{"t":"h2","id":"one-warning-about-hard-cases","runs":[{"text":"One warning about hard cases"}]},{"t":"p","runs":[{"text":"A set made entirely of edge cases has its own distortion: it implies that edge cases are normal. If four out of four examples are awkward, ordinary items start attracting exotic treatment, and you will see the model hedging on a message that needed one word."}]},{"t":"p","runs":[{"text":"So the working shape is mostly-hard-with-an-anchor: three or four boundary cases and one plainly typical one, placed last. The typical example costs you almost nothing and it re-establishes what the ordinary case looks like right before the model answers."}]}],"check":{"question":[{"text":"An operations manager builds a five-example prompt for sorting invoices into “pay now”, “query”, and “hold”. After a month, most errors are invoices that should be queried but come back as “pay now”. What should she add?"}],"options":[{"text":[{"text":"Two more clear examples of “pay now”, so the boundary between it and “query” is sharper."}],"correct":false,"tempting":[{"text":"Adds weight to the category that is already over-produced, which makes the imbalance worse."}]},{"text":[{"text":"An instruction to be more cautious and query anything uncertain."}],"correct":false,"tempting":[{"text":"A vague disposition rather than a boundary; “uncertain” is exactly what the model cannot assess reliably."}]},{"text":[{"text":"Examples of every category in equal numbers, to balance the set."}],"correct":false,"tempting":[{"text":"Balance helps, but blind balance does not target the specific confusion that is producing the errors."}]},{"text":[{"text":"Real invoices that she hesitated over and that turned out to be “query”, each with a one-line reason for the label."}],"correct":true,"tempting":null}],"order":[0,1,2,3],"why":null},"prev":{"slug":"zero-one-and-few","title":"Zero, one, or five: how many examples","minutes":8},"next":{"slug":"order-balance-and-the-hidden-pattern","title":"What your examples teach by accident","minutes":9}}}