How the numbers are actually produced.

Room mode formulas

Axial, tangential, and oblique room modes are computed from the standard rectangular-room modal equation:

f = (c / 2) × √((n_x/L)² + (n_y/W)² + (n_z/H)²)

where c is the speed of sound, L/W/H are the room's length, width, and height, and n_x/n_y/n_z are mode-order integers. This is textbook rectangular-room acoustics, not a proprietary model — see Room Modes Explained for the full walkthrough. It assumes a sealed rectangular room with rigid boundaries; real rooms with irregular shapes, open doorways, or unusually soft/hard surfaces will deviate from the calculated frequencies to some degree.

RT60 assumptions (Sabine equation)

RT60 = 0.161 × V / A

where V is room volume in m³ and A is total absorption in metric sabins (surface area × absorption coefficient, summed across every surface). This is the Sabine equation, the standard starting-point reverberation-time formula. Its own well-documented limitation, carried straight from the calculation engine's code comments rather than softened here: Sabine assumes a statistically diffuse sound field with fairly evenly distributed absorption. Small rooms — the overwhelming majority of home studios — are not acoustically diffuse, especially at low frequencies, so Sabine becomes progressively less reliable below a room's Schroeder frequency, and less accurate generally as average absorption coefficients get very high or very low. Treat the RT60 number as a genuinely useful starting-point estimate, not a measured result.

Speaker-placement heuristics

Recommended listening position uses a commonly-cited starting point of roughly 38% of room length from the front wall — see The 38% Rule Explained for the full reasoning and its real limits. Recommended speaker-to-wall distance starts around 0.6–1.2 m, scaled with room size, specifically to move the resulting speaker-boundary-interference (SBIR) null out of the most critical midrange — see How Far Should Studio Monitors Be From the Wall? for the f ≈ c/(4d) relationship behind that. Every one of these is explicitly labeled a heuristic starting point in the calculation engine's own code, not a guarantee — real placement should still be verified by ear and, ideally, measurement.

Audio loudness measurement

Integrated, short-term, and momentary loudness are measured per ITU-R BS.1770-4 — K-weighting followed by 400ms-block gated measurement, exactly as described in What Is LUFS?. True peak is estimated via 4x oversampling, per the same standard's true-peak guidance — see True Peak Explained. A useful, independently verifiable sanity check: a full-scale 1kHz sine wave measures at very close to -3.7 LUFS integrated under this algorithm, which is a widely quoted reference point for correct BS.1770 implementations.

Frequency-band analysis

Frequency-band energy is computed from the track's power spectral density, split into a fixed set of bands, each reported as a fraction of the total energy across the analyzed range. It's a measurement of where energy actually sits in your file — never a claim about a "correct" balance to hit, and never phrased as a mixing instruction (see the honest phrasing rule in Reference Mastering). What you do with that measured balance is still a mixing decision.

Stem separation limitations

Stem Splitter uses a deep-learning source-separation model, trained on isolated multi-track stems, to estimate the vocal/instrumental (or vocal/drums/bass/other) components of a finished mix. It is a statistical estimate, not a physical un-mixing of the original session — see How to Remove Vocals From a Song for the full explanation. Dense arrangements, heavily doubled or effected vocals, and unusual instrumentation separate less cleanly than a straightforward vocal-over-band recording — that's a real, acknowledged limit of the technology, not a quality bug specific to this site's implementation.

Mastering limitations

The mastering chain applies loudness normalization (bounded-iteration, converging toward each profile's target LUFS) and true-peak-limited peak control, targeting the specific LUFS/true-peak pair documented for each profile in How Loud Should a Master Be?. It is an automated chain reacting to measured loudness and peak characteristics — it does not make creative EQ or dynamics decisions the way a human mastering engineer listening to the track would, and it can't fix underlying issues in the mix it's given (see the honest framing in that article). It's a genuinely useful, real signal chain — not a replacement for a professional mastering engineer on material where that level of judgment matters.