Self-Tracking Your Games

Community stats are a starting point; personal stats are ground truth for calibrating strategies against your play. This chapter describes what to log, how, and what to do with the log.

What to capture

MVP schema for data/playlog.csv:

Workflow

1. After every game, append a row to data/playlog.csv.
2. Invoke si-post-game (Claude skill) to parse the row + prompt for learning points. Surfaces unverified items in played-spirit chapters.
3. Weekly, run a win-rate rollup. Wilson 95% CIs on anything with n ≥ 10.
4. Monthly, compare your rates to mindwanderer for the same (spirit × adversary). Gaps indicate playstyle divergence worth investigating.

What to do with the log

Find your strongest matchups

Group by (spirit × adversary × level). For n ≥ 5, compute win rate + Wilson. Your highest-rate tight-CI groups are your practiced edges.

Find your blind spots

Groups where your rate is 10%+ below community AND your n ≥ 10 → you’re under-performing and the signal is real (not tiny-sample noise). Start there.

Calibrate tempo guesses

fear_generated / final_round = average Fear/round. Compare to Tempo expected curves — divergence indicates pace issues.

Detect pivots that worked

In notes, record “pivoted from G2 to G3 at T3 — ended T4 at 6E” observations. Across 10+ games, patterns emerge: which pivots reliably work vs. which fix symptoms but lose the game.

Existing tooling

• si-post-game skill — accepts game facts, appends to playlog, surfaces unverified items from played-spirit chapters.
• si-daily-challenge — rotates spirits by priority list + your recent log (less-played spirits get priority).
• si-live tool — every state edit persists to data/current-game.json; on game end, export the final state into a playlog row.

What not to over-track

Don’t log every decision. Signal-to-noise is bad and you’ll stop within a week. Track end state + 2-3 surprising moments per game — enough to compound over 50 games.

Sample starter CSV

game_id,date,spirits,adversary,level,scenario,boards,expansions,result,final_round,final_terror,fear_generated,blight_placed,notes
2026-04-19-001,2026-04-19,shadows-flicker-like-flame,england,3,,A,base,win,7,3,14,1,"L1 innate fired T3 via Haunts draft; no cliff hit; fear-rush cleaned up T6-7"

See also

• Reading mindwanderer — Wilson CI + interpretation.
• Digital vs. Tabletop — keep separate logs per mode.
• Skills: si-post-game, si-daily-challenge in skills/.