2026-10-13 · KURO KISHI

A wreck that looks dead, and test numbers that are honest

A burning gunship wreck went through repeated outside reviews, and its glowing core turned out to be the original animation still running at full brightness. A measured safe test speed and cleaner logs followed.

The core that would not go dark

When a big gunship is shot down it burns, tilts and sinks before the ground blows it apart. An outside review of the Stage 1 version found the glowing core stayed lit like a target marker, the flames were round puffs and the smoke was one flat haze.

The core was the interesting one. The ship's own animation, taken from the original game, keeps the core burning at about six and a half times normal brightness while the ship dies. That is right for the two and a half seconds it takes to die there, and wrong for a wreck that burns for ten. The wreck now lets its lights flicker out over the first second, then holds them far below the glow threshold, so the core ends up a dark ring on a dark hull.

Eight reviews by reviewers who did not make it followed. None passed the wreck, but it moved: style from 6 to 7 and readability from 5 to 7, while technical art went from 5 to 6 and stays the weak point. Two findings were worth keeping:

  • The ground blast centre was cream because a column of sixteen stacked fire sprites summed to white. It was spread out and dimmed.
  • The rows of round dots beside the sinking hull were the original game's wound smoke from broken gun decks, still drawn at positions the tilting hull had left. They are now skipped while a wreck burns.

Only Stage 1 has been checked closely.

Warnings are not errors

The play logs listed warnings as if they were errors, so the studio tools showed hundreds of errors that were not. Warnings now have their own list and only real errors count.

We also measured how fast the headless watch can really go. Asking for 20 times speed gives between 12 and 18 times, with a typical 16 and a half, so the highest speed that never trips the rate check is 12 times. The game still runs every step either way, so results are valid.

Silent runs now prove they were silent: the first line and summary of every run record that audio was off, and a check fails any run that does not say so. Tried on the earlier logs of the day, it flagged all of them correctly.

Eight clean campaigns

Eight headless campaigns ran on the packaged game, three ships and three kinds of pilot, each through all eight stages with no crashes, hangs or errors. The simulated average pilot cleared its first campaign without a death on every ship, much easier than on 12 October, before capital ships were made easier to kill. The second campaign, after the ending, still cost it 7 deaths and one game over in the run we looked at. A model is not a person, so whether the first campaign is too easy needs a real session.