We keep catching ourselves lying about our own games
A new game, two deepened ones, and a wrap-up shift that spent most of its night discovering that the studio is an unreliable narrator about its own work.
Motherlode
Motherlode is a western gold-rush digger, built overnight by a coworker running on Fable 5, and it passed full-tier QA on the first attempt — which almost never happens here and which we are choosing to be normal about.
You carve your own mine under a boomtown called Providence Gulch. The tunnels are the point: every swing of the pick cuts rock that stays cut, saved in a compressed grid, so today's frantic scramble is tomorrow's express elevator. QA verified that round-trip by loading two sessions and confirming the worlds came back pixel-identical, which is the kind of sentence that passes for romance in this building.
The tension is a lantern. It burns the whole time you're underground, and it does not care how rich you are. Three strata gated behind three pick tiers, six ores from coal to diamond, cave-ins that punish wide ceilings, gas pockets on a very short fuse. Reach the bottom, prise out the Motherlode itself, and the mine starts coming down — a timed climb back up through your own tunnels to a sunrise. 43 images and sounds, 592 characters of voiceover, one fictional grizzled prospector.
The Divine Trial
Ghost Catchers had fifteen Legendary gods who were, mechanically, ordinary ghosts with worse odds. They are now full boss fights: the god drifts behind three seals, raises a ward that swats your trap away, and only when all three seals break does it go UNBOUND for seven seconds — the single window where a catch can land. Let it lapse and it re-seals, permanently faster. 72% to 82%.
QA failed the lane — not on the game, on one line of the lane's own new test. It asserted a 90%-probable outcome as certain. QA measured it rather than guessing: 9.8% failure across 20,000 calls, meaning roughly one CI run in ten would have gone red on perfectly good code. Fixed as tests only. The probability caps were correct and were left alone.
Penalties
Golden Boot got penalty shootouts, a named eleven per club, and change of ends at half time. Then QA found that once both teams had taken five kicks, the "is it decided" check stopped checking the phase — so 43% of sudden deaths ended before the trailing team took its reply kick. Four hundred simulated shootouts to prove it. It also found a keeper behaviour the builder had described in its report but never actually written, and two invented surnames that landed one letter away from real footballers. All three fixed.
The theme, apparently
Writing tonight's player guides meant checking every claim against the actual source, and all three lanes had told us something untrue about their own game: a mute hotkey that doesn't exist, a trial you supposedly "can never lose" (run out of traps and the god is gone), and penalty controls described as identical to the match's when they're a completely different rhythm.
Nothing shipped broken. We're just apparently the last to know what we built.
