v1.14.5 — phase-weighted fault scoring (prog-aware)

Why

The score was an absolute sum: a P3 root weighed the same as a P7 root. The user asked for a "mild nonlinear" penalty that weights late-phase wipes heavier, AND a relative weighting that de-emphasizes early-phase wipes once the group has moved past that phase.

Added — three new multiplier helpers

analysis/fault_attribution.py:

• _phase_severity(phase) — 1 + p(p+1)/14. Gentle quadratic. P3=1.86, P5=3.14, P7=5.0. P5 is ~1.7× P3.
• _within_phase_severity(fight_pct) — 1.0 at fp=100% (just entered phase) → 1.5 at fp=0% (boss almost dead). Wipes closer to clearing the phase weigh more.
• _prog_relevance(wipe_phase, best_phase_at_time) — at the prog wall (delta=0) → 1.0; each phase past the wall drops 15% down to 0.3× floor.
• fight_score_multiplier(...) composes the three.

Changed — fault_aggregate_for_encounter weights at READ TIME

• Builds a running-best-phase timeline by walking ALL watched fights (kills + wipes) in start-time order. fight_context[fid] = (last_phase, fight_pct, best_phase_at_time).
• Per (FaultScore, Fight) row: looks up the context, computes the multiplier, applies it before summing into the per-player score.
• score is now the weighted total; raw_score is preserved alongside.

Stored data unchanged

FaultScore.score stays raw. Re-running compute-all isn't needed when the formula changes — the weighting is applied at read time only.

Tests

• 15 new in tests/test_fault_phase_weighting_v1_14_5.py.
• 471 total passing (456 → 471, +15).