Conformance report

phonometry’s key differentiator is not the feature list but the proof behind it: every metric is implemented from the governing standard’s text, and a numerical conformance report pins each check to a standard, a clause or table, the normative expected value and the value the library actually computes, with the delta and a pass/fail verdict.

The report is an auto-generated document, regenerated by CI on every pull request (the build fails if it drifts from the code), so it is always in sync with the released library. Every row below is rendered at build time from docs/conformance.json, the committed document the checks write; the same rows as a plain table are in docs/CONFORMANCE.md.

• Filter classes: the IEC 61260-1:2014 class verdict per filter architecture, with the measured relative attenuation at the governing band, the class-1 limit it must clear and the margin in dB.
• Frequency weightings: A/C (IEC 61672-1 Table 3) and G (ISO 7196 A.3) deviations from the nominal curves, judged at the governing frequency with the applicable tolerance band and headroom.
• Closest to their published limit: every check knows the limit it is judged against, so the ten that consume the largest fraction of theirs are listed first. A high figure is not a failure; it is the room the check passes with, and it is the number a tolerance table is really asking about.
• One conformance table per domain (levels, psychoacoustics, room and building acoustics, sound power, materials, vibration, uncertainty, …): the standard and clause, the quantity, the normative value, the value the library computes, the published limit, the deviation and how much of the limit it uses. Expected values come from the standards’ own worked examples or from closed forms synthesized to a known result.

Every domain section is collapsible and stays collapsed while all of its rows pass; a section with any failing row opens automatically. Search and the domain filter open whatever answers them. Every row has its own anchor, so a single check can be linked to directly. On narrow screens the wide tables scroll sideways inside their own box.

The registry of checks lives in scripts/conformance_report.py and is run locally with make conformance, which writes docs/conformance.json and renders docs/CONFORMANCE.md from it. Expected values are single-sourced from the same reference tables the test suite enforces, so the report and the tests cannot disagree silently. This page reads the committed document and never generates it, and CI fails if the document drifts from a fresh run of the checks.

For the design philosophy behind this approach, and a worked case study on IEC 61672-1 time weighting, see Why phonometry.

Re-deriving the standards this closely also turns up defects in the published documents themselves: worked examples that contradict their own normative clauses, misprinted constants, broken cross-references. Every confirmed case, with its evidence and what the library does about it, is recorded in the errata registry.