Each entry names the boundary being crossed and what it costs — a call translated, a path rewritten, a shader compiled, a frame missed.
| No. | Entry | Section | Length |
|---|---|---|---|
| 001 | Translation is not emulationAn emulator pretends to be different hardware and pays for every instruction. A translation layer runs the same instructions natively and only rewrites the calls that cross into the operating system. That distinction sets the whole cost model. | THE BOUNDARY | long |
| 002 | The syscall boundaryThe narrow doorway between a program and the kernel, and why it is the right place to intercept. | THE BOUNDARY | medium |
| 003 | ABI, not APIMatching the compiled shape of a call, not the source-level name of it. | THE BOUNDARY | short |
| 004 | What cannot be translatedKernel drivers and anti-tamper systems that inspect the machine itself. | THE BOUNDARY | short |
| 005 | Threading assumptionsScheduling behaviour a program was tuned against, quietly different underneath. | THE BOUNDARY | short |
| 006 | One API into anotherEvery draw call rewritten on the way through, and the state that has to be tracked to do it correctly. | PIPELINE | medium |
| 007 | State trackingThe older API remembers; the newer one does not. The layer has to. | PIPELINE | short |
| 008 | Shader compilation stutterThe first time a scene appears, shaders are compiled. That is the hitch, and why a cache fixes it only the second time. | PIPELINE | medium |
| 009 | The shader cacheWhere the compiled result lives, and what invalidates it. | PIPELINE | short |
| 010 | Descriptor setsBinding resources in batches instead of one at a time. | PIPELINE | short |
| 011 | Case sensitivityA program that wrote Textures/ and reads textures/ works on one system and fails on another. The commonest and dullest breakage there is. | FILESYSTEM | medium |
| 012 | Path separatorsBackslash against forward slash, and the rewriting that happens invisibly. | FILESYSTEM | short |
| 013 | The registry that is a fileA key-value store the program expects, implemented as a plain file. | FILESYSTEM | short |
| 014 | File lockingTwo systems with different ideas about who may hold a file open. | FILESYSTEM | short |
| 015 | Where the milliseconds goFrame time, not frame rate: an average hides the one frame in sixty that took four times as long, and that is the one you feel. | FRAME TIME | medium |
| 016 | The one per cent lowThe measurement that corresponds to what smoothness actually feels like. | FRAME TIME | short |
| 017 | CPU bound or GPU boundTwo very different problems that produce the same complaint. | FRAME TIME | short |
| 018 | Input latencyThe distance between a key press and a photon, and the queue in between. | FRAME TIME | short |
| 019 | Vsync and tearingWaiting for the display, and what you trade for not waiting. | FRAME TIME | short |
| 020 | Measuring honestlyA repeatable run, a warm cache and a stated configuration. Without those a number means nothing. | BENCH | medium |
| 021 | The warm runThe first pass is always slower. Which pass you should be reporting. | BENCH | short |
| 022 | Thermal driftTen minutes in, the same machine is a different machine. | BENCH | short |
| 023 | What a percentage hidesRelative numbers without absolutes, and why they are unfalsifiable. | BENCH | short |