This course excludes Kelly-style sizing because its inputs are too easy to overestimate and its outputs are easily misread as real-money advice. A mathematical formula does not remove model risk. Estimated edge, outcome probability, independence, and payoff assumptions can all be wrong, especially in a small, selected, non-stationary dataset.
Instead, the paper engine uses fixed fictional point caps chosen for test coverage, not growth. The cap is identical across comparable cases, small relative to the fixed starting-point budget, and bounded again by topic concentration. It has no PKR conversion and does not adapt upward after favorable outcomes.
Study the failure chain. A forecast looks calibrated in one period. The researcher treats the score as a precise probability advantage. Correlated events are counted as independent. Quoted prices are treated as obtainable fills. Costs and latency are understated. A sizing formula amplifies every upstream error. The resulting paper curve can look smooth until one shared assumption fails.
Risk engineering asks different questions: How sensitive is the conclusion to probability error? What happens if outcomes in a theme fail together? How much of the result comes from a few cases? Does the system survive stale data and missing labels? The correct response to uncertainty is smaller experimental exposure or abstention, not a more elaborate sizing equation.
Use scenario tests with fixed point caps: baseline, probability shifted adversely, costs doubled, correlated cluster failure, and longest outage. Report maximum drawdown, unresolved concentration, and cases blocked by limits. These are diagnostics, not forecasts.
Document the exclusion in SAFETY.md and the schema. Reject fields named kelly, bankroll_fraction, account_equity, or currency-based size. A code review should see that the system cannot smuggle monetary optimization through configuration.
The decision is permanent for this curriculum release and is covered by regression tests.
🇵🇰 Pakistan Angle
Learners may encounter social posts presenting aggressive sizing as a shortcut to income. This course does not reproduce that path. It teaches estimation error, validation, and event-sourced accounting through fictional points. Do not borrow, deposit, or connect financial accounts for any exercise.
Hands-On Exercise
Create a synthetic case set where estimated probabilities are five percentage points too confident and several outcomes share one theme. Compare fixed one-point scenarios with an uncapped hypothetical only as a pre-supplied chart—do not implement a sizing formula. Explain which assumptions drive the divergence and add schema rejection tests.
Completion Rubric
- The exclusion is justified through estimation and dependence risk.
- Paper limits are fixed, fictional, and currency-free.
- Adverse probability, cost, cluster, and outage scenarios are reported.
- Forbidden monetary-sizing fields fail schema validation.
- No formula or instruction enables real-money sizing.
Sources
Key takeaway: When probability and dependence estimates are uncertain, a sizing formula can amplify error; fixed paper limits keep the lesson about engineering.