# Interpret estimates

## Read the basis with the number

helx combines compiler output, simulator counters and platform parameters. These are different kinds of evidence. Preserve their provenance when reporting a result.

| Output | What it tells you |
| --- | --- |
| Compiler analysis | The compiled model and its compiler-reported resource requirements |
| FVP counters | Execution in the configured simulator environment |
| Energy estimate | A calculation using counters and the platform energy model |
| Platform provenance | The revision and confidence status of underlying values |

Energy outputs depend on calibration and should be treated as experimental unless your evaluation explicitly establishes otherwise. Do not present estimates as measured silicon results.

## Provenance status

- **Confirmed:** values reviewed or measured by their owners.
- **Fitted:** values tuned to reproduce reference measurements.
- **Unverified:** specification values, estimates or placeholders without that confirmation.

The bundle records the least trusted status it depends on in `provenance.values.status`. A confirmed parameter does not turn a simulated experiment into a physical measurement.

## Compare like with like

Keep the model, compiler/runtime, platform revision, clock and memory configuration explicit. Change one variable at a time. Cross-check estimates against measurements on the intended hardware before using them for a product decision.