Deliberate practice with Anthropic's high-scoring interaction patterns for genuine skill mastery
Run deliberate practice sessions that build genuine mastery using:
| Pattern | Mastery Effect | How We Apply It |
|---|---|---|
| Generation-First | +18% retention | You attempt before seeing solution |
| Hybrid Code-Explanation | +23% transfer | Visual + verbal + worked examples |
| Conceptual Inquiry |
| +15% deep learning |
| Why questions before how questions |
| Metacognitive Reflection | +12% self-awareness | Reflect on your thinking process |
| Anti-Pattern | Mastery Effect | Warning Signs |
|---|---|---|
| AI Delegation | -17% mastery | "Just give me the answer" |
| Progressive Reliance | -22% independence | Each problem harder without AI |
| Iterative Debugging | -14% understanding | Trial-and-error without learning |
┌────────────────────────────────────────────────────────────────┐
│ THE GENERATION-FIRST PRINCIPLE │
├────────────────────────────────────────────────────────────────┤
│ │
│ 1. ATTEMPT (struggle productively) │
│ ↓ │
│ 2. REFLECT (what made it hard?) │
│ ↓ │
│ 3. REVEAL (see the solution) │
│ ↓ │
│ 4. COMPARE (where did your thinking diverge?) │
│ ↓ │
│ 5. CONSOLIDATE (what pattern did you learn?) │
│ │
│ ✗ NOT: See solution → Try to replicate │
│ ✓ YES: Struggle → Compare → Understand gap │
│ │
└────────────────────────────────────────────────────────────────┘
When invoked with a skill (e.g., /ari-practice kelly-criterion):
// 1. Initialize the skill registry (loads from ~/.ari/learning/)
const { getSkillRegistry } = await import('../cognition/learning/skill-registry.js');
const registry = await getSkillRegistry();
// 2. Get or register the skill
let skill = registry.getSkill(skillId);
if (!skill) {
skill = await registry.registerSkill({
skillId: skillId,
skillName: 'Kelly Criterion',
domain: 'LOGOS',
targetLevel: 80,
initialLevel: 0,
});
}
// 3. Calculate ZPD for optimal challenge
const zpd = registry.calculateZPD(skillId);
╔════════════════════════════════════════════════════════════════╗
║ 🎯 DELIBERATE PRACTICE SESSION ║
║ ══════════════════════════════════════════════════════════════ ║
║ ║
║ Skill: Kelly Criterion ║
║ Domain: LOGOS (Quantitative Reasoning) ║
║ ║
║ ┌─────────────────────────────────────────────────────────┐ ║
║ │ Current Level ████████████░░░░░░░░░░░░░░ 35/100 │ ║
║ │ Target Level ████████████████████████░░░░░ 80/100 │ ║
║ │ Gap to Target ───────────────────────────── 45 pts │ ║
║ └─────────────────────────────────────────────────────────┘ ║
║ ║
║ Zone of Proximal Development ║
║ ┌─────────────────────────────────────────────────────────┐ ║
║ │ TOO EASY │ OPTIMAL │ TOO HARD │ ║
║ │ [0━━━━━25] │ [25━━━━●━━━━━45] │ [45━━━━━━━━━━━100] │ ║
║ │ │ YOU ARE HERE │ │ ║
║ └─────────────────────────────────────────────────────────┘ ║
║ ║
║ Today's Focus: Problems calibrated to difficulty 25-45 ║
║ Estimated Duration: 20-30 minutes ║
║ ║
╚════════════════════════════════════════════════════════════════╝
Before any problems, activate prior knowledge:
╔════════════════════════════════════════════════════════════════╗
║ 📚 CONCEPTUAL WARM-UP ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Before we practice, let's activate your understanding. ║
║ ║
║ Q: In your own words, what is the Kelly Criterion trying ║
║ to optimize for? ║
║ ║
║ 💭 Think about this for 30 seconds before continuing... ║
║ ║
║ [Your response: ___________________________________] ║
║ ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ 💡 This is "conceptual inquiry" — understanding WHY │ ║
║ │ before practicing HOW produces 15% better transfer │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
╚════════════════════════════════════════════════════════════════╝
After response, provide the conceptual anchor:
╔════════════════════════════════════════════════════════════════╗
║ ✅ CONCEPTUAL ANCHOR ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Kelly Criterion optimizes for: ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ LONG-TERM GEOMETRIC GROWTH RATE │ ║
║ │ (not expected value, not minimizing risk) │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ The key insight: ║
║ • Betting too little → suboptimal growth ║
║ • Betting too much → risk of ruin ║
║ • Kelly = the exact fraction that maximizes E[log(wealth)] ║
║ ║
║ Formula: f* = (bp - q) / b ║
║ Where: b = odds, p = win probability, q = loss probability ║
║ ║
║ 📊 Visual Intuition: ║
║ Growth │ ╭───────╮ ║
║ Rate │ ╱ ╲ ║
║ │ ╱ ╲ ║
║ │ ╱ ● Kelly ╲ ║
║ │ ╱ ╲ ║
║ │╱ ╲ ║
║ └────────────────────── ║
║ Bet Size → (0% to 100%) ║
║ ║
║ Ready to practice? Let's begin. ║
║ ║
╚════════════════════════════════════════════════════════════════╝
CRITICAL: User attempts BEFORE seeing any hints or solutions
╔════════════════════════════════════════════════════════════════╗
║ PROBLEM 1/5 [ZPD: ●] ║
║ Difficulty: 35 (In Your Optimal Zone) ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ SCENARIO: ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ You have a trading opportunity: │ ║
║ │ • Win probability: 60% │ ║
║ │ • If you win: 2x your bet │ ║
║ │ • If you lose: lose your bet │ ║
║ │ • Current bankroll: $10,000 │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ QUESTION: ║
║ What fraction of your bankroll should you bet? ║
║ ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ 🎯 GENERATION PHASE (3-5 minutes) │ ║
║ │ │ ║
║ │ 1. Write out your calculation step by step │ ║
║ │ 2. Include your reasoning at each step │ ║
║ │ 3. State your final answer with confidence level │ ║
║ │ │ ║
║ │ Your work: │ ║
║ │ ________________________________________________ │ ║
║ │ ________________________________________________ │ ║
║ │ ________________________________________________ │ ║
║ │ │ ║
║ │ Final answer: _____% │ ║
║ │ Confidence: □ Low □ Medium □ High │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ ⏱️ Take your time. Struggling productively is learning. ║
║ ║
║ [Submit when ready] ║
║ ║
╚════════════════════════════════════════════════════════════════╝
╔════════════════════════════════════════════════════════════════╗
║ 🤔 METACOGNITIVE PAUSE ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Before I show the solution, reflect: ║
║ ║
║ 1. What part felt most uncertain? ║
║ □ Identifying the variables (b, p, q) ║
║ □ Remembering the formula ║
║ □ Executing the calculation ║
║ □ Interpreting the result ║
║ ║
║ 2. Where might you have made an error? ║
║ [________________________________] ║
║ ║
║ 3. What would change your answer? ║
║ [________________________________] ║
║ ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ 💡 This metacognitive reflection increases retention │ ║
║ │ by 12% compared to just seeing the answer │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ [Ready to see the solution] ║
║ ║
╚════════════════════════════════════════════════════════════════╝
╔════════════════════════════════════════════════════════════════╗
║ ✅ SOLUTION + COMPARISON ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ CORRECT ANSWER: 40% (0.40) ║
║ ║
║ WORKED SOLUTION: ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ Step 1: Identify variables │ ║
║ │ b = 2 (payout odds, 2:1) │ ║
║ │ p = 0.60 (win probability) │ ║
║ │ q = 0.40 (loss probability, 1-p) │ ║
║ │ │ ║
║ │ Step 2: Apply formula │ ║
║ │ f* = (bp - q) / b │ ║
║ │ f* = (2 × 0.60 - 0.40) / 2 │ ║
║ │ f* = (1.20 - 0.40) / 2 │ ║
║ │ f* = 0.80 / 2 │ ║
║ │ f* = 0.40 │ ║
║ │ │ ║
║ │ Step 3: Interpret │ ║
║ │ Bet 40% of bankroll = $4,000 │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ YOUR ANSWER: [Shown here] ║
║ ║
║ ┌─────────── COMPARISON ───────────┐ ║
║ │ ✓ Variables identified correctly │ or │ ✗ b was wrong │ ║
║ │ ✓ Formula applied correctly │ or │ ✗ Forgot q term │ ║
║ │ ✓ Calculation accurate │ or │ ✗ Arithmetic error │ ║
║ └──────────────────────────────────┘ ║
║ ║
║ KEY INSIGHT from your specific gap: ║
║ [Personalized insight based on comparison] ║
║ ║
║ Visual: ║
║ [No Bet] ████████████████░░░░░░░░ [Full Kelly] ║
║ ↑ 40% ║
║ ║
╚════════════════════════════════════════════════════════════════╝
╔════════════════════════════════════════════════════════════════╗
║ 📌 CONSOLIDATION ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Now that you've compared, answer these to lock in learning: ║
║ ║
║ 1. What's the most common error you should watch for? ║
║ [________________________________] ║
║ ║
║ 2. If the odds were 3:1 instead of 2:1, what would change? ║
║ [________________________________] ║
║ ║
║ 3. When would Kelly suggest betting 0%? ║
║ [________________________________] ║
║ ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ These "elaborative interrogation" questions build │ ║
║ │ the connections that enable transfer to new problems │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
║ [Continue to next problem] ║
║ ║
╚════════════════════════════════════════════════════════════════╝
═══════════════════════════════════════════════════════════════════
📊 PRACTICE SESSION COMPLETE
═══════════════════════════════════════════════════════════════════
Skill: Kelly Criterion
Duration: 25 minutes (22 focused)
┌─────────────────────────────────────────────────────────────────┐
│ PERFORMANCE │
│ ═══════════════════════════════════════════════════════════════ │
│ │
│ Problems Attempted: 5 │
│ ✅ Correct (before hints): 3 │
│ ⚠️ Correct (with reflection): 1 │
│ ❌ Incorrect: 1 │
│ │
│ Generation Success Rate: 60% → 80% with metacognition │
│ │
│ ┌────────────────────────────────────────────────────────┐ │
│ │ 🎯 This is above your ZPD average — excellent progress! │ │
│ └────────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────┐
│ SKILL PROGRESSION │
│ ═══════════════════════════════════════════════════════════════ │
│ │
│ Level Progress: 35 → 42 (+7 points) │
│ ████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 42% │
│ │
│ Weekly Trajectory: +12 points/week │
│ Estimated Time to Target: 4 weeks │
│ │
│ ZPD Adjustment: Now calibrating problems to 30-50 range │
│ │
└─────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────┐
│ LEARNING PATTERNS DETECTED │
│ ═══════════════════════════════════════════════════════════════ │
│ │
│ ✓ Strengths: │
│ • Formula recall is solid │
│ • Variable identification improving │
│ │
│ ⚠️ Focus Areas: │
│ • Arithmetic under time pressure │
│ • Remembering to subtract q (loss probability) │
│ │
│ 📚 Recommended Next: │
│ • Practice 3 more problems with edge cases (p = 0.5, b = 1) │
│ • Review the "when Kelly = 0" concept │
│ │
└─────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────┐
│ INTERACTION QUALITY │
│ ═══════════════════════════════════════════════════════════════ │
│ │
│ Session Type: Generation-First ✓ │
│ │
│ High-Scoring Patterns Used: │
│ ✓ Attempted before seeing solution (5/5 problems) │
│ ✓ Metacognitive reflection (4/5 problems) │
│ ✓ Conceptual warm-up completed │
│ ✓ Consolidation questions answered │
│ │
│ Anti-Patterns Avoided: │
│ ✓ No AI delegation ("just tell me") │
│ ✓ No progressive reliance │
│ ✓ No unlearned debugging │
│ │
│ 🎯 This session follows high-mastery interaction patterns │
│ │
└─────────────────────────────────────────────────────────────────┘
🏆 MILESTONE: Foundation Laid! (25%)
"You have established the fundamental patterns!"
Next session recommended: Tomorrow, same time
Focus: Edge cases and verification strategies
═══════════════════════════════════════════════════════════════════
The session monitors for patterns that undermine skill development:
// Detect if user is asking for answers without attempting
function detectDelegationPattern(interaction: Interaction): Warning | null {
const delegationPhrases = [
'just tell me',
'what\'s the answer',
'give me the solution',
'I don\'t want to think about it',
];
if (delegationPhrases.some(p => interaction.input.toLowerCase().includes(p))) {
return {
type: 'DELEGATION_DETECTED',
message: `
┌────────────────────────────────────────────────────────────┐
│ ⚠️ DELEGATION PATTERN DETECTED │
├────────────────────────────────────────────────────────────┤
│ │
│ Asking for answers without attempting reduces mastery │
│ by ~17% (Anthropic 2024). │
│ │
│ Instead, try: │
│ • "I'm stuck at [specific step], can you hint?" │
│ • "I got [answer], is my approach correct?" │
│ • "What am I missing in my reasoning?" │
│ │
│ The struggle is where learning happens. │
│ │
└────────────────────────────────────────────────────────────┘
`,
};
}
return null;
}
// After practice session completes
const { getPracticeTracker } = await import('../cognition/learning/practice-tracker.js');
const tracker = await getPracticeTracker();
const { session, skill } = await tracker.recordSession({
userId: 'default',
skill: skillId,
session: {
skill: skillId,
startedAt: sessionStart,
endedAt: new Date(),
plannedMinutes: 25,
focusedMinutes: 22,
tasksPlanned: 5,
tasksCompleted: 4,
errorPatterns: ['forgot-to-subtract-q', 'arithmetic-under-pressure'],
interactionPatterns: {
generationFirst: true,
metacognitiveReflection: 4,
consolidationCompleted: true,
delegationAttempts: 0,
},
},
});
// Update skill registry with practice
const { milestone } = await registry.updateFromPractice(skillId, session);
if (milestone) {
console.log(`🏆 MILESTONE: ${milestone.name}!`);
console.log(` "${milestone.celebration}"`);
}
If the storage adapter cannot load or save session data:
try {
const session = await tracker.recordSession({ ... });
} catch (error) {
if (error.code === 'STORAGE_UNAVAILABLE') {
// Gracefully degrade: Continue session without persistence
console.warn(`
╔════════════════════════════════════════════════════════════════╗
║ ⚠️ STORAGE WARNING ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Unable to save session progress to disk. ║
║ ║
║ Your practice will continue, but progress won't be saved. ║
║ Check that ~/.ari/learning/ directory is writable. ║
║ ║
║ [Continue without saving] ║
║ ║
╚════════════════════════════════════════════════════════════════╝
`);
}
}
If stored data is malformed or corrupted:
// Storage adapter automatically backs up before writes
// Recovery: ~/.ari/learning/skills.json.backup
// Manual recovery command:
// cp ~/.ari/learning/skills.json.backup ~/.ari/learning/skills.json
If a practice session is interrupted (crash, network loss, etc.):
╔════════════════════════════════════════════════════════════════╗
║ 🔄 INCOMPLETE SESSION DETECTED ║
╠════════════════════════════════════════════════════════════════╣
║ ║
║ Found an incomplete practice session from 15 minutes ago: ║
║ ║
║ Skill: Kelly Criterion ║
║ Completed: 3/5 problems ║
║ Time spent: 12 minutes ║
║ ║
║ Would you like to: ║
║ ║
║ [A] Resume where you left off (2 problems remaining) ║
║ [B] Review what you completed and start fresh ║
║ [C] Discard and start a new session ║
║ ║
║ ┌────────────────────────────────────────────────────────┐ ║
║ │ 💡 Resuming preserves your progress and maintains │ ║
║ │ your skill level trajectory │ ║
║ └────────────────────────────────────────────────────────┘ ║
║ ║
╚════════════════════════════════════════════════════════════════╝
Implementation:
// Check for incomplete session on start
const incomplete = await tracker.getIncompleteSession(userId, skillId);
if (incomplete && (Date.now() - incomplete.startedAt) < 3600000) { // 1 hour
// Offer resume
}
src/cognition/learning/skill-registry.ts - ZPD and proficiency trackingsrc/cognition/learning/practice-tracker.ts - Session recordingsrc/cognition/learning/storage-adapter.ts - Persistencesrc/cognition/learning/weakness-analyzer.ts - Error pattern detection| Domain | Focus Areas | Icon |
|---|---|---|
| LOGOS | Bayesian reasoning, Kelly, EV, Systems Thinking | 🧠 |
| ETHOS | Bias detection, Emotional regulation, Discipline | ❤️ |
| PATHOS | CBT, Stoicism, Reflection, Wisdom traditions | 🌱 |
Related Skills:
/ari-review - Spaced repetition reviews for long-term retention/ari-learning-mode - Comprehension-building interaction mode/ari-think - Metacognitive reasoning supportIntegration:
Practice sessions automatically create spaced repetition cards for concepts you struggle with. These cards appear in your daily review queue via /ari-review.
Workflow:
┌─────────────────────────────────────────────────────────────┐
│ │
│ [/ari-practice] → Build skill through deliberate practice │
│ ↓ │
│ Auto-generate review cards for weak areas │
│ ↓ │
│ [/ari-review] → Reinforce via spaced repetition │
│ ↓ │
│ Long-term mastery and retention │
│ │
└─────────────────────────────────────────────────────────────┘
┌────────────────────────────────────────────────────────────────┐
│ │
│ "The goal is not to get the right answer. │
│ The goal is to build the skill that generates │
│ right answers independently." │
│ │
│ Every time you struggle and then compare, │
│ you're building neural pathways. │
│ │
│ Every time you ask for the answer, │
│ you're bypassing those pathways. │
│ │
└────────────────────────────────────────────────────────────────┘