{
  "schema": "vedokrok.public-item.v1",
  "release_id": "MHC-RPUB-20261011-9e4b62c5",
  "url": "/knowledge/train-the-boundary-with-a-near-miss",
  "id": "MHC-D-RESEARCH-9208",
  "version": "0.2.0",
  "title": "Train the boundary with a near-miss",
  "summary": "A near-miss can show exactly where the boundary sits.",
  "kind": "protocol",
  "body": "Pair a correct example with a close nonexample that keeps irrelevant context similar but changes one requirement. Name what flips the classification, get feedback, then test a fresh case. Controlled studies suggest close nonexamples can sharpen boundary learning; a 2026 follow-up found matching irrelevant features reduced practice time and errors in its task.",
  "limits": [
    "Evidence is strongest in controlled concept-learning tasks with small samples and arbitrary visual concepts. Close nonexamples can require more time and errors. Matching improved efficiency in one 2026 study, and feedback mattered there. For safety-critical or regulated categories, use authoritative definitions and expert review."
  ],
  "topics": [
    "union-case-comparison-for-transfer-and-discrimination"
  ],
  "intents": [],
  "evidence": [
    {
      "claim": "A meta-analysis of 57 experiments and 336 tests found that case-comparison activities produced better learning outcomes than sequential cases, single cases, nonanalogous cases, traditional instruction or control on average (d = 0.50, 95% CI [0.44, 0.56]).",
      "role": "supports",
      "note": "The synthesis pooled varied tasks, domains and comparison conditions. The average effect does not imply that any two cases placed side by side will improve learning.",
      "locator": "Abstract",
      "source_id": "RS-5FFF4E5CE8ED18B6C9D1"
    },
    {
      "claim": "Across three geoscience experiments, learners identified faults better when they compared fault and non-fault cases rather than viewing cases separately; early comparison of visually similar contrasts also outperformed dissimilar contrasts.",
      "role": "supports",
      "note": "The experiments taught one spatial geoscience concept. They support close contrasts as a discrimination tool but do not establish that maximum similarity is always desirable.",
      "locator": "Abstract",
      "source_id": "RS-1BB824C5534FF98E1FF6"
    },
    {
      "claim": "Across six 2025 experiments, practice with close-in nonexamples that differed from examples on fewer required features produced more robust classification of untaught examples and nonexamples than practice with far-out nonexamples in most tested conditions.",
      "role": "supports",
      "note": "The experiments used small single-case samples, mostly visual or arbitrary concepts, and a specific differential-reinforcement procedure. They do not show that near-miss examples are best for every kind of concept.",
      "locator": "Abstract; General Discussion",
      "source_id": "RS-225E1B484E87E080A82B"
    },
    {
      "claim": "The same 2025 experiments found a practical cost: close-in nonexamples generally required more practice time or trials and produced more errors than easier far-out contrasts, while new irrelevant features at test also disrupted conceptual learning.",
      "role": "limits",
      "note": "The size and importance of the added practice cost varied across experiments and participants.",
      "locator": "General Discussion",
      "source_id": "RS-225E1B484E87E080A82B"
    },
    {
      "claim": "In a 2026 follow-up, presenting close-in examples and nonexamples with the same irrelevant features reduced errors and practice time compared with unmatched pairs, while both practice conditions produced similar final conceptual learning.",
      "role": "supports",
      "note": "Experiment 1 had five completers learning arbitrary visual concepts. Matching is therefore a promising efficiency tactic, not an established universal design rule.",
      "locator": "Abstract; Experiment 1; General Discussion",
      "source_id": "RS-5903DDC50BC85621F14D"
    },
    {
      "claim": "In the 2026 study, extended exposure to matched or unmatched examples and nonexamples without differential feedback did not produce stable conceptual learning in the tested task.",
      "role": "contextualizes",
      "note": "This result concerns the study's arbitrary visual concepts and reinforcement-based procedure. Rule instruction and other learning formats can work differently.",
      "locator": "Experiment 2; General Discussion",
      "source_id": "RS-5903DDC50BC85621F14D"
    }
  ],
  "use_when": [
    "You know obvious examples but still confuse borderline cases that differ on one decisive cue."
  ],
  "avoid_when": [
    "Evidence is strongest in controlled concept-learning tasks with small samples and arbitrary visual concepts. Close nonexamples can require more time and errors. Matching improved efficiency in one 2026 study, and feedback mattered there. For safety-critical or regulated categories, use authoritative definitions and expert review."
  ],
  "example": "For an approval rule, compare two similar requests where only one changes the legal entity. Name the difference, then classify a new request.",
  "check": "You can classify a fresh near-miss and explain the decisive cue without relying on irrelevant context.",
  "source_ids": [
    "RS-5FFF4E5CE8ED18B6C9D1",
    "RS-1BB824C5534FF98E1FF6",
    "RS-225E1B484E87E080A82B",
    "RS-5903DDC50BC85621F14D"
  ],
  "published_at": "2026-10-01T14:27:00Z",
  "steps": [
    "State the rule and the cue you think is required.",
    "Choose a true example and near-miss missing one important requirement.",
    "Keep irrelevant details similar so they do not become the shortcut.",
    "Classify both and name the decisive difference before feedback.",
    "Test a fresh unlabeled case; redesign the pair if it fails."
  ],
  "editorial_overlay": null,
  "editorial": null,
  "notices": [],
  "sources": [
    {
      "id": "RS-5FFF4E5CE8ED18B6C9D1",
      "title": "Learning Through Case Comparisons: A Meta-Analytic Review",
      "url": "https://doi.org/10.1080/00461520.2013.775712"
    },
    {
      "id": "RS-1BB824C5534FF98E1FF6",
      "title": "Finding faults: analogical comparison supports spatial concept learning in geoscience",
      "url": "https://doi.org/10.1007/s10339-013-0551-7"
    },
    {
      "id": "RS-225E1B484E87E080A82B",
      "title": "Using must-have and can-have features to improve conceptual learning",
      "url": "https://doi.org/10.1002/jeab.70037"
    },
    {
      "id": "RS-5903DDC50BC85621F14D",
      "title": "Differential reinforcement using matched examples and nonexamples expedites the practice needed to learn concepts",
      "url": "https://doi.org/10.1002/jeab.70130"
    }
  ],
  "relations": [
    {
      "from": "MHC-D-RESEARCH-9208",
      "to": "MHC-D-RESEARCH-9221",
      "type": "useful_with",
      "url": "/knowledge/compare-two-cases-before-you-write-the-rule"
    },
    {
      "from": "MHC-D-RESEARCH-9208",
      "to": "MHC-D-RESEARCH-9238",
      "type": "use_before",
      "url": "/knowledge/mix-practice-when-the-real-task-is-choosing-the-method"
    }
  ],
  "collections": [
    {
      "id": "RC-36BEDA777EAF62BC47BA",
      "title": "Use near-miss nonexamples to learn where a concept stops",
      "url": "/collections/use-near-miss-nonexamples-to-learn-where-a-concept-stops"
    }
  ],
  "topic_hubs": [],
  "solution_packs": []
}