[feat] 24-b r1: the game-football e2e suite (real scene + real zip, two peers + late joiner)

[AlexZ005]

Lane 29-football: the 24-B remainders (R1 → R2 → R3).

What landed

• R1 2be54c3: tests/e2e/game-football.test.cjs. It loads the real games/football/scene.tpscene from the scenes feed (SCENES_BASE @v2) and installs the real football.zip on two peers plus a late joiner. If the scene or the zip can't be reached, the suite skips with the reason printed; it never fails on the network. It covers the modules test-flight's cases, the HUD screens (menu, playing, pause, over) and the P menu. Game-shell transitions are asserted on the trigger log, never on a perRound flowValue.
  • Counterfactuals:
    • A ball nobody touched, put into the red gate, still scores for blue and credits no player (in-suite).
    • The late joiner with remoteSimulating cleared gets a dead knock probe (in-suite, then restored).
    • With the handshake simulate push disabled in peerHandler, checks 5.3 and 5.8 go red; restored, they pass.
• R2 (D1, the HUD-builder decision): no, nothing committed. I tried a gameShellHud({...}) extraction and authored Towers and Stars Room before and after it (--only towers,stars-room --out .d1/{before,after}, then node scripts/compare-authored.cjs .d1/before .d1/after):
  • Towers came out identical; Stars Room did not (first difference: pause panel w 400 → 380).
  • Stars Room has no menu screen, and its pause screen uses different sizes and spacing from Towers' (400x400 panel with 5 buttons 50px apart, against 380x300 with 3 buttons 52px apart). The Beat contest's menu differs from Towers' in 8 numbers.
  • The only screen with the same shape in two games is "over", and Football's HUD lives in the module (hud.js → football.def.json), so a core helper can't produce it.
• R3 is modules PR [docs] 24-b r3: football row, api.onHit/hitLog digest, DEVX #15 yes modules#6 (docs).

Suites

suite	this branch	base
game-football (new)	102 PASS / 0 FAIL, 71 s	—
game-stars-room	36	36
game-towers	20	20
knock-node	50	50
knock-physics	76 + 2.14 red	76 + 2.14 red (already failing on base)

No core code changed, so the held suites stay at their base counts. svelte-check is 341 errors / 47 warnings with the same list as the base, and the build passes.

Findings (core, not fixed here)

• Two peers can both run the physics simulation. If both press Play inside the sim's start-up window, both pass maybeSimOnPlay's "is a sim already running?" guard before the other's simulate message lands, so both simulate. After B's hit, B sent the ball about 74 moves in 2 s; on A the ball then sat under a permanent hold: external, applyThrow snapped back, and no goal could score. The suite enters Play in order and asserts check 2.2b (B does not start a second simulation). This needs a ticket: a peer that receives simulate while it is simulating should yield by a deterministic rule.
• P pauses everyone. In the Football template the P pause menu is session-wide, because the Key Press pulse replicates. Check 3.10 asserts that behaviour as it is.

Owed

• The headset half of the football flight (VR hands, the physical buttons by ray, a colocated match).

🤖 Generated with Claude Code