Statistics: Why and How

Spirit Island is a game of probabilities layered under strict mechanics. You can win by feel — most experienced players do — but at higher difficulty levels the feel-based approach hits a ceiling. Beyond that ceiling, counting starts paying.

This part gives you the mental models, the stat sources, and the reusable reasoning tools for playing the underlying math as a first-class part of your game.

What’s in this part

• Reading mindwanderer — how to interpret the community win-rate dashboard without misreading aggregate data. Includes Wilson-CI primer.
• Digital vs. Tabletop — why the Steam app’s win-rate distributions don’t map 1:1 to your physical table.
• Fear Card Expected Value — per-expansion Fear pool distribution + Monte Carlo per-game output.
• Event Deck Risk Profiles — balance + top spirit/invader-favoring events.
• Deck Expansion Impact — headline pool sizes + simulated per-game outcomes across all 5 decks × 6 expansion combos.
• Expansion Dilution Claims — variance analysis testing “B&C swingy, JE balances, NI dilutes” hypothesis.
• Blight Deck Severity — severity-proxy distribution.
• Token-Effect Dilution — per-token × per-deck × per-combo frequency.
• Aspect Power Deltas — how much each Aspect shifts its base spirit’s win rate.
• Major vs. Minor Draft Rates — when Minor-heavy beats Major-shopping.
• Self-Tracking Your Games — what to log, how to use si-post-game, the basics of a personal win-rate history.

The core statistical lenses

Four lenses run through every chapter:

1. Confidence, not point estimates. A win rate of 60% across 10 games isn’t 60% ± 0 — the Wilson 95% interval for 6 wins in 10 is about (31%, 83%). Every rate here reports a CI where data supports one.
2. Conditional on expansion pool. The Minor deck at base-only is not the Minor deck at all-expansions. A “33% chance of drawing Moon-heavy” means different things in each pool.
3. Per-turn distributions, not per-game averages. “Averages 6 fear/game” hides whether that’s 6 evenly or 2+0+0+4+0. Shape is the signal.
4. Exchangeable, not independent. Each deck is a shuffled stack drawn without replacement, so draws are exchangeable, not independent: each unseen card is equally likely to be next, but every card you see changes the odds for the rest. The Event deck is not reshuffled (the Steam engine never does), so this holds for the whole game; a Power deck reshuffles its discard only when it runs out. The Turn-1 Event is drawn and discarded without resolving, and it is public, so it already shrinks the unseen Event pool by one. Fear Cards are earned, not drawn on a schedule: one card per 4 Fear per player, with the Terror Level rising when a divider is reached. Dice-style “independent per turn” math gives the wrong odds for both decks.

Data provenance

• Wiki-parsed card data — data/decks/{minor,major,fear,event,blight}.json, parsed by scripts/wiki-fetch.py.
• Monte Carlo simulations — scripts/deck-simulate.py runs 5 000+ trials per expansion combo; outputs at data/references/deck-stats/sim_*.json.
• Community win-rate data — mindwanderer. Spirit Island Digital games only, analyzed with regression models. Directional; subject to selection bias. Treated accordingly.