Sharpen league and Climb difficulty and give Gore a front door

PROJECT.md

@@ -4888,6 +4888,40 @@ per late opponent check quality ceilings and open-court frequency, with Club Pro
 control. All 131 tests pass with the default seed and eight additional seeds. Build v349 targets
 staging only. Combined difficulty still needs Oliver's playtest; shot statistics are not win rates.
 
+## Session log: THE RETEST — sustained final-league returns, a brisker Climb, Gore at the front (2026-09-07, v350)
+
+Oliver still defends league 5 reliably at level 26; the first 6–7 Climb opponents feel like filler.
+The previous final-league pass raised quality CAPS but left the .45 floor intact. Raising pace
+traits alone would do nothing here: their return-pace multiplier already hits its cap of 1.
+
+- Brute / Director / Champion now return in quality bands .62–.76 / .68–.83 / .74–.90, and use
+  open-court aim on 50% / 70% / 85% of returns. `trySwing` reads the optional foe `qMin` just as
+  it already reads classic's floor. These final-league traits follow the body into cup, training
+  and Climb. Player level/peer gifts, opponent legs, wrong-footing, wounds and cup length stay put.
+- **Climb-only opening**: `gameplay.matchStep` adds four progression steps at opponent one,
+  tapering linearly to zero at opponent eight. Movement (including intercept planning), return
+  pace, width and flub all read it. Actual opponent identity, territory pace and learning still
+  read the real rung. Cup onboarding and its training reps retain their old early curve.
+  Measured on actual update/return paths: the first Climb foe chases at 3.407 vs 2.520 m/s,
+  averages 18.639 vs 14.817 m/s return pace, reaches 2.364 vs 1.595 m return width, and returns
+  1152 vs 1073 of 1200 contact attempts. The existing auto-run reachability soak still passes.
+- Oliver chose the **direct Gore door** over a mid-season teaser: SPLATTER GORE sits below the
+  primary league fixture and starts Gore in one tap with the selected court/kit. OTHER GAMES,
+  STATS and BOARD share one quiet navigation row (44px touch targets); their headline numbers
+  are removed from the splash. Court/kit setup remains accessible through OTHER GAMES.
+- Badge glints peak at the START of their long CSS cycle; the first peak is now .39–.89 s after
+  creation instead of ~5.5–19 s. The three points stagger, repeat slowly and still respect reduced
+  motion. Existing art and typography are reused; no generated assets or dependencies.
+
+Validation: 132 tests pass with the default seed plus 1, 7, 42, 99, 1234, 2026, 8888 and 20260907.
+The return sampler measures mean quality .688 / .756 / .820 and open-corner shares 66.8% / 80.1% /
+90.7% for the three final opponents. New Climb regression exercises real movement and returns,
+including convergence at eight and unchanged cup/training behavior. Browser checked at 390×844
+and 320×568; all splash buttons fit, Gore starts directly, no browser errors in that flow.
+This is stronger measured opposition, not a measured human win rate: league 5 still needs
+Oliver's level-26 phone verdict. Oliver requested merging to main and pushing for his iPhone
+test; this release carries the cache-busting entry version v350.
+
 # OPEN — leftovers, in rough priority order
 
 Nothing here is broken; these are known gaps and unverified risks.