Lab / SEP 10 2026 / experiment

The A/A control that said 19.99%

Measuring a configuration against itself turned up a 19.99% separation under the default .NET runtime. With tiering off it fell to 0.06–0.21%.

Measurement methodology, Controls, Build configurationWeft

Machine: UNKNOWN — not stated in the source
Commit: UNKNOWN
A/A separation, default tiering: 19.99%
A/A separation, DOTNET_TieredCompilation=0: 0.06% – 0.21%
In-process coefficient of variation: falls 7–10×

Before changing anything, I measured the same configuration against itself. That is the whole idea of an A/A control, and it is the only way I know of to find out what my own noise floor is.

Under default tiering the pooled A/A control separated its two arms by 19.99%. With `DOTNET_TieredCompilation=0` the same control separated them by 0.06% to 0.21%. In-process coefficient of variation falls by 7–10×.

What this does not mean

This is a variance collapse, not a throughput win. Nothing got faster. A reader who takes it as a speed result has read it backwards, which is why I am leading with the variance framing rather than burying it.

The number that matters is the 19.99%. With default tiering, a real 20% effect and no effect at all are indistinguishable by this harness. That is a statement about the instrument first and about the workload second.

Arm	Pooled A/A separation
default tiering	19.99%
`DOTNET_TieredCompilation=0`	0.06% – 0.21%

The same configuration measured against itself. These are separations, not speeds — there is no ratio to state and the source records no direction for this entry.

**Environment.** .NET 10. The run itself is undated; it is recorded in `weft-STATUS-2026-09-10.md` (2026-09-10). Machine: UNKNOWN.

**Methodology.** Pooled A/A control: one configuration measured against itself. The reading is the separation between the two arms, so a smaller separation means a quieter instrument rather than a faster one. In-process coefficient of variation is the second reading, and it falls 7–10×.

This is the least Weft-specific entry I have and the most reusable one. Any .NET benchmark run with default tiering and no A/A control is measuring its own noise, and it has no way to know it.