[
  {
    "id": 1,
    "name": "Dual Process Classifier",
    "category": "cognitive_models",
    "difficulty": 3,
    "language": "python",
    "description": "Build a classifier that takes a description of a human decision and determines whether it was primarily System 1 (fast, intuitive, automatic) or System 2 (slow, deliberate, analytical). Input: a scenario string. Output: dict with 'system' (1 or 2), 'confidence' (0-1), 'reasoning' explaining which cognitive markers were present (heuristics, emotional valence, time pressure, complexity). Include edge cases where both systems interact."
  },
  {
    "id": 2,
    "name": "Working Memory Simulator",
    "category": "memory_architecture",
    "difficulty": 4,
    "language": "python",
    "description": "Simulate human working memory with Miller's 7±2 capacity limit. Build a WorkingMemory class that can hold items, chunk related items together (reducing slots used), decay items over time if not rehearsed, and prioritize emotionally tagged items. Methods: add(item, emotional_weight=0), rehearse(item), retrieve(), dump() showing current state with decay levels. Demonstrate how chunking allows more information in fewer slots."
  },
  {
    "id": 3,
    "name": "Cognitive Bias Detector",
    "category": "decision_making",
    "difficulty": 3,
    "language": "python",
    "description": "Given a decision scenario with context and outcome, identify which cognitive biases are at play. Support at least 8 biases: confirmation bias, anchoring, availability heuristic, sunk cost fallacy, dunning-kruger, bandwagon effect, framing effect, hindsight bias. Input: scenario dict with 'context', 'decision', 'outcome'. Output: list of detected biases with confidence scores and explanations of how each manifests in the scenario."
  },
  {
    "id": 4,
    "name": "Attention Allocation Engine",
    "category": "attention",
    "difficulty": 4,
    "language": "python",
    "description": "Model selective attention as a resource allocation problem. Build an AttentionEngine that receives a stream of stimuli (each with salience, novelty, relevance_to_goal, emotional_intensity) and allocates a finite attention budget across them. Implement: bottom-up capture (high salience/novelty forces attention), top-down control (goal-relevant items get priority), and attentional blink (brief refractory period after processing high-intensity stimuli). Return attention distribution per timestep."
  },
  {
    "id": 5,
    "name": "Mental Model Builder",
    "category": "reasoning",
    "difficulty": 5,
    "language": "python",
    "description": "Implement a system that builds and updates mental models from observations. A MentalModel holds beliefs as key-value pairs with confidence levels. When new evidence arrives, it updates beliefs using a simplified Bayesian approach. The model should detect when new evidence contradicts existing beliefs (cognitive dissonance) and measure the 'surprise' level. Track model revision history. Demonstrate with a scenario where initial beliefs about a system are progressively corrected by evidence."
  },
  {
    "id": 6,
    "name": "Cognitive Load Estimator",
    "category": "hci",
    "difficulty": 3,
    "language": "python",
    "description": "Estimate cognitive load for a user interface task. Input: a task description with number of choices, information density, time pressure, familiarity level, and number of mode switches. Output: estimated cognitive load score (0-100) broken into intrinsic load (task complexity), extraneous load (poor design), and germane load (learning effort). Classify as underload/optimal/overload. Apply Hick's Law for choice time and Fitts's Law principles for interaction complexity."
  },
  {
    "id": 7,
    "name": "Learning Curve Modeler",
    "category": "learning",
    "difficulty": 3,
    "language": "python",
    "description": "Model how humans learn a skill over time. Implement three learning curve models: power law of practice, exponential learning, and S-curve (logistic). Given performance data points (attempt_number, performance_score), fit each model and determine which best describes the learning pattern. Calculate: current skill level, predicted mastery time, learning rate, and the plateau point. Handle the spacing effect — distributed practice should show better retention than massed practice."
  },
  {
    "id": 8,
    "name": "Decision Fatigue Tracker",
    "category": "decision_making",
    "difficulty": 3,
    "language": "python",
    "description": "Model decision fatigue over a day. Build a DecisionTracker that logs decisions with their complexity (1-10) and emotional weight (1-10). Track cumulative decision load. Predict quality degradation based on ego depletion theory — after threshold load, decisions become either impulsive (shortcut to default) or avoidant (defer/skip). Model recovery through rest periods. Output current decision capacity as percentage and recommended action (proceed/simplify/rest)."
  },
  {
    "id": 9,
    "name": "Expertise Reversal Detector",
    "category": "hci",
    "difficulty": 4,
    "language": "python",
    "description": "The expertise reversal effect: instructions that help novices can hurt experts, and vice versa. Build a system that takes a user's skill level (novice/intermediate/expert) and an instructional design (scaffolding level, detail level, example count, redundancy level) and predicts whether the design will help or hinder that user. Model the crossover point where guidance becomes a burden. Output: effectiveness score, recommendation to increase or decrease scaffolding, and the estimated optimal instruction level for this user."
  },
  {
    "id": 10,
    "name": "Distributed Cognition Mapper",
    "category": "cognitive_models",
    "difficulty": 4,
    "language": "python",
    "description": "Model distributed cognition — how thinking is spread across people, tools, and environments. Input: a team scenario with agents (people), artifacts (tools, documents, screens), and information flows between them. Build a CognitionMap that identifies: where knowledge lives (which agent or artifact), single points of failure (one person holds critical info), cognitive bottlenecks (too much flowing through one node), and redundancy (same info in multiple places). Output a health score and vulnerability report."
  },
  {
    "id": 11,
    "name": "Metacognition Monitor",
    "category": "metacognition",
    "difficulty": 5,
    "language": "python",
    "description": "Build a metacognition system — thinking about thinking. A MetaMonitor wraps around a problem-solving process and tracks: what strategy is being used, whether progress is being made (stuck detection), confidence calibration (does stated confidence match actual performance), when to switch strategies, and what was learned from the attempt. This is the most self-referential task: build a system that monitors its own cognitive process. Log the monitoring itself as data."
  },
  {
    "id": 12,
    "name": "Cognitive Architecture Comparator",
    "category": "cognitive_models",
    "difficulty": 4,
    "language": "python",
    "description": "Compare three major cognitive architectures: ACT-R (production rules + declarative memory), SOAR (problem spaces + chunking), and Global Workspace Theory (broadcast + competition). For a given simple task (like making a sandwich), simulate how each architecture would process it — what gets activated, what sequence of operations occurs, where bottlenecks appear. Output a comparison table showing strengths and weaknesses of each for that task."
  },
  {
    "id": 13,
    "name": "Interface Affordance Analyzer",
    "category": "hci",
    "difficulty": 3,
    "language": "python",
    "description": "Analyze UI affordances using Gibson's and Norman's theories. Input: a list of UI elements with properties (shape, color, size, position, label, state). For each element, determine: perceived affordance (what users think they can do), actual affordance (what they can do), signifier quality (how well the element communicates its function), and mapping quality (how naturally the control relates to its effect). Flag mismatches between perceived and actual affordances as usability risks."
  },
  {
    "id": 14,
    "name": "Embodied Cognition Simulator",
    "category": "cognitive_models",
    "difficulty": 5,
    "language": "python",
    "description": "Simulate embodied cognition — the theory that thinking is shaped by the body. Model an agent navigating a space where physical states affect cognitive performance: temperature affects working memory, hunger reduces executive function, physical posture influences confidence levels, and movement aids creative thinking. The agent must solve problems while managing their physical state. Track how bodily conditions change cognitive output quality. Show that the same problem yields different solutions based on physical context."
  },
  {
    "id": 15,
    "name": "Theory of Mind Engine",
    "category": "social_cognition",
    "difficulty": 5,
    "language": "python",
    "description": "Build a Theory of Mind system — modeling what another agent believes, wants, and intends. Given a scenario with two agents where Agent A has information Agent B does not (asymmetric knowledge), model: what Agent B believes (possibly false), what Agent B would predict Agent A will do, and what Agent A knows about Agent B's beliefs (second-order). Handle false belief scenarios (where one agent's model of the world is outdated). This is the foundation of empathy, deception detection, and social reasoning."
  }
]