si-post-game — Log Game + Surface Learning

Normalize Brett's shorthand game report, append to playlog, and give him 1–2 concrete takeaways to feed tomorrow's challenge.

Data Sources

1. ~/repos/spirit-island/data/playlog.csv — append here.
2. ~/repos/spirit-island/src/spirits/{tier}/{slug}.md — the spirit chapter for this game (for common-mistakes + tempo-profile lookup).
3. ~/repos/spirit-island/src/adversaries/{slug}.md — adversary chapter (if full-written).

Input Normalization

Brett's input will look something like:

"Just played Shadows vs Sweden L3. Won on T8, fear 12, blight 4. Almost lost T5 because I forgot to defend."

Parse into:

• date (today if not given)
• player_count (1 unless he says otherwise)
• spirits (shadows-flicker-like-flame)
• aspects (none unless he mentions)
• adversary (sweden)
• level (3)
• scenario (none unless he mentions)
• board (unknown unless he mentions A/B/C/D/E/F)
• result (win / loss)
• rounds (8)
• fear_level_end (12 — this is the fear pool total, not Terror level)
• blight_end (4)
• notes (the free-text commentary verbatim)

Ask for anything missing only if it's load-bearing (result, spirit, adversary, level). Don't pester for board letter or fear pool if he didn't mention them.

Append to playlog

Append one row to ~/repos/spirit-island/data/playlog.csv. Use the schema in the CSV header.

Surface 1–2 learning points

1. Read the spirit chapter's Common Mistakes section.
2. Match Brett's notes + game state against the mistakes list. For example:

• If he mentions "forgot to defend" and the chapter has a "forgot-to-defend" mistake → surface that.
• If rounds > the chapter's expected game length and he lost → surface "tempo issues" with a pointer to the Tempo Profile table.
• If blight_end is near cascade and he lost → surface blight-management mistake.

1. Limit to 2 points. More = noise.

Output Format

Logged: {date} {spirits} vs {adversary} L{level} — {result} T{rounds}

## What likely cost/made the game

1. **{Learning point 1}** — {one sentence; link to chapter section}
2. **{Learning point 2}** — {optional; only if there's a second signal}

## Rolling stats
- Games logged: {N}
- Win rate overall: {X%}
- Win rate vs {this adversary} at L{level}: {X% (n=Y)}

## Next session suggestion
{One line — either "repeat to cement the lesson" or "rotate to {spirit/adversary} via `si-daily-challenge`"}

Rules

• Never editorialize beyond what the chapter says. If the chapter doesn't call out a mistake, don't invent one.
• Include the note field verbatim in the CSV row so future sessions can reference Brett's own framing.
• If Brett's note contradicts what you'd surface, ask once: "You said X; the chapter flags Y — which felt like the actual issue?"
• CSV appending: use >> append, not overwrite. Preserve the header row.

Edge cases

• Multi-spirit game: spirits field gets comma-separated slugs (shadows-flicker-like-flame,thunderspeaker).
• Aspect used: aspects field gets the aspect name.
• Multi-session game (rare): treat each session as one row.
• Chapter is a stub: skip the mistakes lookup; output "chapter not yet written — just logged." Still surface tempo/blight observations from general fundamentals.