A piece that had already landed kept sliding. The report came back within the hour.
Fixing one bug can write another. The first attempt at the corner fix quietly broke a much more basic promise, that where you land is where you stay, and a test suite that only checked whether a piece was supported never once asked whether it moved after it didn't need to.
The corner-stones fix closed most of the mid-air stalls, but not all of them, a rarer case remained, where a piece's own straight-down path was blocked while the pile still had room for it to go lower some other way. The first attempt at a fix for that case was: fall in whichever direction lowers the piece's cells the most, not just its own straight-down path.
A plausible idea, and it was wrong
It worked, in the narrow sense: mid-air stalls roughly halved. It also broke something nobody had tested for. The planet's geometry has a small secondary pull near its poles, pieces that have already landed near the centre still register as able to "go lower" by sliding sideways along the ring they're sitting on. A piece that had finished falling kept counting as falling.
And because a piece that can still "fall" never starts its lock timer, those pieces didn't just drift once, they sat there, unresolved, until the next gravity tick nudged them again. The report came back inside an hour: positions were being forced toward the middle of a ring or its extreme sides, especially on the first ring out from the core. That's exactly what the bug predicts, described by someone who had no idea what the code was doing.
Why the tests didn't catch it
Every existing test checked that a piece ends up supported by something. None of them checked that a piece, once it had landed, stayed exactly where it landed, because until this bug, that had simply always been true, so nobody had written a test for it. Two earlier versions of a stricter check had actually existed and passed anyway: one judged "at rest" using the same fall code that had the bug in it, and the other never let a game run long enough to reach the first ring where the drift showed up.
The actual fix
The rule now has a strict order: a piece tries its own straight-down path first, exactly as before any of this started. If that's blocked and the piece is already resting on anything, the pile, the core, a corner stone, it stops moving, full stop; gravity never nudges a landed piece again. Only a piece that is still genuinely airborne, resting on nothing at all, is allowed to look for another way down.
Of 3,584 straight drops onto an empty board, zero landed anywhere other than where the original, simpler rule would have put them, the new rule is provably identical to the old one for every piece that was never actually airborne in the first place. The regression test that guards this now includes a real-browser check: sixteen fresh pieces, soft-dropped onto the first ring at sixteen different positions, watched for longer than a full gravity step to confirm none of them so much as twitch after landing.