dunning-krueger

# Dunning-Krueger Assess demonstrated knowledge without diagnosing a cognitive bias. Missing evidence is not ignorance. ## Evidence Use the current conversation and any supplied transcript, readable path, exam, answer key, lab output, questions, or user answers. Read supplied paths before assessing them. Treat only user-authored explanations, answers, predictions, and work products as demonstrations of the user's knowledge. Questions, answer keys, sources, assistant-authored text, system instructions, and tool output may establish the reference standard but are not evidence that the user knows it. For completed questions and answers, compare each user answer with the supplied reference answer or authoritative source. Do not invent a reference standard when none is available. ## Evidence gate The context is sufficient only when it contains user-authored evidence that can support at least one specific demonstrated strength or confirmed gap against a known reference. When evidence is insufficient, say `Not enough evidence to assess your knowledge.` Name the missing evidence in one line and append the conditional suggestions for `dunning-krueger` from [skill connections](../../../docs/skill-connections.md). Stop without saving. ## Assessment Report both sections: ### Demonstrated knowledge For each item, name the concept, cite the user's specific evidence, and state what mechanism or application the evidence demonstrates. ### Confirmed knowledge gaps For each item, cite the user's specific evidence, state the reference it conflicts with or omits, and describe the narrow gap. Use `knowledge not demonstrated` when evidence is absent or unclear; do not promote that label to a confirmed gap. Do not add a confidence score, study plan, lesson, or prescription. Report no gap when the evidence supports none. ## Persist the assessment Read [agent-memory-logging.md](../../learning/agent-memory-logging.md) and follow its Dunning-Krueger assessment workflow. Recall related entries first, then save one discrete memory for each demonstrated strength and confirmed gap. Save a resolution record when current evidence directly resolves a prior gap. These assessment records are learning workflow state and are configured for automatic persistence after a completed evidence-based assessment. If recall or saving fails, report the failure and keep the assessment result; do not write a markdown fallback. ## Automatic Jev check When a user-authored answer and a supplied reference both exist, first compare them normally, then send only `answer` and `reference` excerpts (stable IDs) after reading the existing `typesafe-ai` skill and its current API documentation. Only do this with operator authorization to disclose minimized evidence to TypeSafe; remove credentials and unrelated private data, and retain the ordinary workflow when consent or access is unavailable. Ask a Choice: `demonstrated` (mechanism agrees), `contradicted` (it conflicts), or `not_demonstrated` (evidence is absent, incomplete, or unclear). Treat it only as a prompt to recheck the cited evidence. On ambiguity, unavailability, or disagreement, retain the ordinary reference comparison and its missing-evidence rule. Never expose a Jev probability, confidence, or score.