Every post here uses numbers from actual save files rather than a synthetic benchmark, because in this game the difference is enormous: a fresh farm and a level 400 farm are not the same program under load. Where a fix changed nothing for a particular save, that is reported too. What a change does not do is half of the result.
When the profiler lies
Five real cases where a diagnostic tool answered a question I had not asked: an empty reading printed as zero, an experiment the game silently undid, and a baseline that drifted by a factor of 8.6.
The freeze was not the shadows
Players reported 10 fps. Shadows, resolution and triangle count were all innocent. What a three.js scene actually charges for, found by ablation, plus one dead end worth publishing.
Exponential cost, linear benefit
The same balance failure in four separate systems: the price curved upward and the thing it bought did not. Including the upgrade that cost 4.1 billion lira for 3.2% more output.
A multiplier that fed itself
One correct-looking change closed an economic loop and produced saves holding numbers past float precision. Why the syntax check, the save verifier and a three-save balance measurement all passed.
The budget was in the wrong currency
A demand budget promised the same bale twice, so upgrading the factory made the shortage worse. Then the fix removed a player from the game, because my fixture set had no edge profile.
A dead button is worse than a missing one
A 2.5 billion lira building shipped and not one of 326 saves installed it. One flag carrying two meanings swallowed the click in silence.
What this is not
It is not a release-notes feed. Version notes live on their own page and inside the game. This section is for the parts worth reading if you build things: the measurement, the wrong turn, and the number that settled it.
If you play the game rather than build one, the guide is the better page. It covers the economy, the production chain and fleet strategy as they actually work in the current build.