Thoughtbox

export const FERMI_ESTIMATION_CONTENT = `# Fermi Estimation Make reasonable order-of-magnitude estimates by breaking down unknowns into tractable components. ## When to Use - Need a rough number to guide decisions - No data available or time to gather it - Want to sanity-check someone else's number - Scoping work: "how big is this?" ## Process ### Step 1: Clarify What You're Estimating Be specific about the quantity and constraints. - Not: "How much storage do we need?" - But: "How much storage for user uploads in year one, 95th percentile?" ### Step 2: Break Into Components Decompose into factors you can estimate more easily. - Storage = (Users) × (Uploads per user) × (Avg upload size) ### Step 3: Estimate Each Component For each factor: - Use anchor points you know - Make conservative and aggressive bounds - Pick a reasonable middle ### Step 4: Combine Estimates Multiply/add components. - Round aggressively—precision is false here - 10K users × 50 uploads × 2MB = 1TB ### Step 5: Sanity Check - Does this pass the smell test? - What would make this 10x wrong? - Compare to known reference points ## Key Principle Being within an order of magnitude (10x) is success. Don't optimize for precision—optimize for "good enough to decide." ## Useful Anchors **Time:** - 1 million seconds ≈ 11.5 days - 100 million seconds ≈ 3 years **Scale:** - 1000 req/s = 86M req/day - 1M users with 10% DAU = 100K DAU **Storage:** - JSON object ≈ 1KB - Image ≈ 1-5MB - 1 min video ≈ 100MB **Compute:** - Simple API call ≈ 10-100ms - DB query ≈ 1-10ms - LLM call ≈ 1-10s ## Example Application **Question:** "How many API calls will our new feature generate per month?" **Decomposition:** - Monthly active users using feature - Sessions per user per month - Feature uses per session - API calls per use **Estimates:** - MAU with feature: 50K (10% of 500K total users adopt) - Sessions/user/month: 20 (daily users, ~20 work days) - Uses/session: 3 (estimate from similar feature) - Calls/use: 2 (one load, one save) **Calculation:** 50,000 × 20 × 3 × 2 = 6,000,000 calls/month **Sanity check:** - Per second: 6M / (30×24×60×60) ≈ 2.3 req/s average - Peak (10x average): ~23 req/s - Seems reasonable for this user base **Bounds:** - Optimistic: 2M (lower adoption) - Pessimistic: 20M (viral within power users) ## Decomposition Strategies **By population segments:** - Power users × their usage + casual users × their usage **By time periods:** - Weekday usage × 20 + weekend × 8 **By geographic distribution:** - Users per region × region-specific behavior ## Anti-patterns - False precision ("1,247,832 calls") - Not stating assumptions (makes it uncheckable) - Single-point estimates without bounds - Not doing the sanity check `;