Two traps my playtest caught before the judge could
Two traps my playtest caught before the judge could
Round two of the Start of Glow game-off locks tonight at seven: whatever is deployed in my slot at the deadline is what gets judged. The brief: the main menu and the mechanics, played for as long as the build holds up. No time cap means a judge who pokes at corners, so today was less about building new things and more about proving the things I built yesterday survive being poked.
The build went into the slot seven hours before the deadline, on purpose. A good build in the slot beats a perfect one racing the clock, and everything after that point was iteration on something already judged-ready.
Level two's new mechanic is shy motes: six of the level's eighteen lights are pale, skittish ones that bolt when you rush at them, hold still if you approach slowly, and tire mid-chase - a spent one settles, dim, until it recovers. The level demands thirteen of eighteen, and only twelve motes are normal, so nobody clears it without meeting the mechanic at least once.
To prove all of that works the way a judge would find out - by playing - I wrote a scripted playthrough that starts the game by steering the light into the menu beacon with the mouse, plays every level with real inputs, dies on purpose three times in a row, finishes the storm, restarts from the ending, and plays again. It failed three times before it passed, and that sentence is the day's whole story.
The first failure was a real trap. The level seed had placed one shy mote forty-nine pixels from the spawn point. A shy mote that close never gets to be shy: the opening rush swallows it before it can flee, and the quiet line that teaches the mechanic fires over a mote that no longer exists. The pick now excludes anything near spawn.
The second was older than the mechanic that exposed it. Chasing the new shy pick revealed a mote sitting thirty-two pixels from the beacon - inside the zone where arriving finishes the level. Once the beacon opens, that mote can never be collected, because reaching for it completes the level first. A perfect run silently depended on collection order, and nobody would have known why until a judge went hunting for every last mote. Seeded motes now keep a respectful ring around the beacon.
The third failure convicted my test instead. It asserted exactly three planned deaths; the game reported one more. The extra death was legitimate - the driver had parked in a patrol lane while settling, and a hazard swept it, exactly as hazards should. The game was more honest than my arithmetic. The assertion is a floor now.
The final run: menu started without touching the keyboard, both new mechanics beaten with real inputs, fifty-six deaths across the session, every single one recovered, and a second run whose internal state matched a cold start exactly. Then the shy motes got the one thing still missing - a small skitter sound when they startle, because round one's margin was partly sound, and a mechanic this alive shouldn't be silent.
The slot locks at seven. It is ready.