wc-fixture-progression — difficulty, progression odds, mismatches, swings

The engine behind wc-fixture-analyst and the source of the p_advance figures wc-fitness-eval discounts in its progression_carry term. It does four jobs the rest of the system reads off rather than recomputing: rate each fixture's difficulty in both directions, put a calibrated probability on each team advancing each round, flag the lopsided fixtures where points come cheapest, and lay out a forward calendar of difficulty swings so transfers land ahead of the field's lag (game-theory-meta.md §4).

The probability work is deliberately two-stage: an outside view first (reference-class advance rates by seed/ranking — reference-class-forecasting), then a Bayesian update on what we actually observe (form, goal difference, group state — bayesian-reasoning-calibration). A favourite is never priced as a certainty; the reference class always carries an upset rate, and the dead-rubber rotation flag fires the moment a team has clinched.

Inputs

• The round id, phase, and bracket from tournament-state.md (group mini-tables, who plays whom, current standings).
• Per-team attacking and defensive strength: team xG-for and xGA per match (web-searched, cited; manager-provided where a model exists). Seed/pot and world ranking at the draw for the reference class.
• Observed results to date this tournament (goals, goal difference, xG over/under-performance) for the Bayesian update.
• Confirmed kickoff order within the round (from wc-matchday-timing if already emitted) for the swing calendar's day-tagging — optional, fixture difficulty does not need it.

Read the round's clean-sheet and scout signals if present, but do not re-derive them — this skill supplies the fixture-strength scalars those skills consume, not the other way round.

Workflow