Add team run differential and Pythagorean expectation

[Mattsface]

Summary

• Adds a /run-differential page: per-game run differential trend plus the Pythagorean expected record beside the actual one
• No migration and no new MLB request. Runs allowed is already in the database — it is the opponent's own runs on their row for the same game_pk. A self-join in list_team_season_run_results recovers it
• Win/loss is derived rather than stored: a completed MLB game cannot end tied, so outscoring the opponent is the whole definition. That gives an actual record without waiting on Add game result context and W/L markers to team charts #29
• Pythagorean expectation uses the exponent 1.83 that Baseball Reference publishes against, so the figure can be checked against a public source

Closes #39.

What makes this page different from the other four

It is the only signed metric. Every other per-game value has a floor of zero; this one does not. The chart draws diverging bars split into win and loss traces rather than a marker line, and its y axis uses rangemode="normal" — anchoring at zero would clip every loss off the chart. Its zero line is drawn darker than the grid, because zero is the win/loss boundary here rather than an arbitrary axis end.

It has no MLB reference line, which is not an omission. League-wide run differential is exactly zero by construction: every run scored by one team is a run allowed by another, so the totals cancel. The zero line already on the chart is the MLB average, and the page says so rather than leaving a reader wondering what is missing.

Missing-data handling

The join is a LEFT join on purpose. A single-team import has no opponent rows at all, and an inner join would silently return zero games for it — indistinguishable from a team that was never imported. Instead the unpaired game_pk values are reported, and the page refuses with the established 409 pattern.

The remedy differs from the batting-strikeout and baserunner backfills: nothing is wrong with the team's own rows and re-importing the team cannot help, so the page names scripts/import_league_season.py instead. A partially paired season is refused too — an average over only the paired games would understate runs allowed and produce a differential that looks plausible and is wrong.

The other four pages read only the team's own rows, so they are unaffected by a single-team import and keep working.

Test plan

• poetry run pytest — 1309 tests passing, 98% coverage (new: 110 tests across analytics, repository join, chart contract, schemas, formatting, and the web page)
• poetry run ruff check . / poetry run ruff format --check . clean
• Repository tests cover the join's real failure modes: a third team sharing a date, doubleheaders (keyed on game_pk, not the date), cross-season isolation, and symmetry from both clubs' perspectives
• Verified end-to-end against a seeded 162-game two-club season: 162 games paired, 0 unpaired, 94-68 actual vs 88.0 expected wins, page renders 200 with 94 win bars and 68 loss bars spanning −10 to +11

🤖 Generated with Claude Code

https://claude.ai/code/session_0178J2Rs8QQfS1wTqG98FSK9