Methodology · rubric v1
The methodology: our rubric, weights and rounding, in public
A rating you can’t audit is an adjective wearing a number. This page is the audit: the dimensions, the weights, the rounding rule, and the limits of what a documentation-based review desk can honestly claim.
The arithmetic
Every tool receives 0–10 in five dimensions. Scores are multiplied by fixed weights, summed, multiplied by ten and rounded half-up to a Standard Score out of 100. The same table appears on every review with that tool’s numbers filled in — if our maths ever disagrees with our prose, the maths is the review and the prose owes you a correction.
| Silence weight 0.35 | How much noise actually disappears, and what survives of the voice |
|---|---|
| Toll weight 0.20 | What it costs your machine and your patience: hardware demands, setup friction |
| Wallet weight 0.20 | Value against price, free tier included |
| Fit weight 0.15 | Platforms covered and how naturally it sits in a call-day workflow |
| Candor weight 0.10 | How honestly the vendor documents limits, pricing and privacy |
What we verify, and how
Prices, free-tier limits, platform support and hardware requirements are checked against vendor pages and help centres, and every page prints its verification date. When a vendor changes a price, the score can change — that is the system working, not failing. Facts on this site were last re-verified July 30, 2026.
What we refuse to claim
We do not run an acoustics laboratory, and we decline to invent one. Where a judgment can only be settled by your room, your accent and your hardware — exactly how a suppressor treats your voice — our reviews say so and point you at the free tier to run the test that matters. Docs-based scoring has limits; pretending otherwise is how this category got so adjectival.
Independence, mechanically
Some reviewed tools pay referral commissions; some pay nothing and outscore ones that do. The wall is mechanical rather than moral: scores are computed from the rubric table before any recommendation box is written, and the funding page lists every commercial relationship. NVIDIA and SteelSeries pay us nothing; both sit above Descript, which links through an affiliate route. The ledger does not know who pays.
Versioning
This is rubric v1, in force since launch. If weights ever change, the rubric gets a new version number, every score is recomputed, and the change is logged here — silently re-marking old homework is not correction, it is laundering.