Real examples
- Which server to buy — the initial recommendation was one platform; the benchmark showed that a 2015 chip with 2 cores delivered 2.4× more per core than the low-power 6-core chip. The rule that stuck: never a T/U-series chip for continuous workloads. The purchase decision changed.
- Why the class platform is slow — there was real instability at the voice provider, but the structural problem was audio generated on click for static content, with no cache, where a failure turned into 2s of silence. Delivered an instrumented replacement and the recommendation to pre-render.
- Is it worth building product X — assessment of a client idea: a defensible gap was identified, along with what already exists in the market.
- Transcription capacity — the assessment that became the +38% case study.
How it’s done
Measure before proposing — even when the measurement kills the proposal itself. No number without a source makes it in. The assessment is published on its own page, with a link the client can forward. ward.