Game Progress and Measuring the Right Thing

Seeing a character move through an actual game is one of the more satisfying parts of this project. The Hollow Knight work reached controlled gameplay in King's Pass, with tests that followed the scene and checked what the game was doing. That was a useful application result, even though a complete human play session, audio, saves, and the rest of the game still needed broader acceptance.

The route itself needed more care than simply holding a direction. A camera movement can make the player look stationary, and a character near the edge of a frame can be difficult to locate. The development work used the scene and the captured images to distinguish a real stall from a change in how the game was being observed.

Performance work also needed better controls. If two guests or a build are competing for the same machine, an apparent slowdown may have very little to do with the code being compared. The project accumulated rules about one guest at a time, recording load, and comparing the same backend and workload. Those conditions are part of the result, not details to leave out of the report.

Memory growth was another example. A long game run can show a problem without identifying which component owns it. Reproducing growth outside the game gives us a smaller case to examine, and it avoids blaming the title for behavior introduced by the translation stack or the harness.

The same period included CPU and translation research, along with packaging and test-isolation work. I'm interested in optimizations that survive a real comparison, but I don't want a faster number produced by changing the workload, lowering the quality, or losing part of the measurement.

So the takeaway from this part of the journal is specific: there was real, bounded gameplay progress, and the measurement process became more demanding. It wasn't a declaration that the game was fully supported or that Ferrum had reached a general performance target. Those are larger claims that need their own evidence.