THE LAB · THRESHOLD INSTRUMENT
What this instrument cannot do.
Every ladder in the threshold test was measured against the audio it is made of, and 9 of those measurements came back imperfect. They are listed here in the words of the pipeline that found them. None of them is a bug that was left unfixed — each one is a place where the physics of the material, not the code, sets a limit on what can be claimed.
This page exists because of a specific decision: compression levels are labelled in kilobits per second, which is exact on every passage, and that label was only allowed on the condition that the variation behind it be stated in public.
Pool rendered 2026-08-19. Acoustic measurements of audio files — no respondents are involved, so there is no data-source badge on this page.
| LADDER | UNIT | RUNGS, GENTLEST FIRST |
|---|---|---|
| Pitch drift | cents | 3.1 · 4.4 · 6.3 · 8.8 · 12.5 · 17.7 · 25 · 35.4 · 50 · 70.7 · 100 |
| Timing smear | ms | 12.5 · 15.7 · 19.8 · 25 · 31.5 · 39.7 · 50 · 63 · 79.4 · 100 |
| Compression · pb1 | kbps | 160 · 128 · 112 · 96 · 80 · 64 · 48 · 40 · 32 |
| Compression · pb4 | kbps | 192 · 128 · 96 · 80 · 64 · 56 · 48 · 40 · 32 |
| Compression · pb6NOT PRESENTED | kbps | 112 · 80 · 64 · 56 · 48 · 40 · 32 |
A bitrate is exact; what it destroys is not
A compression level is labelled in kbps because a bitrate is the same number on every passage. The damage it does is not. These are the widest spreads measured across the passages serving one level — which is why a compression threshold is always reported with the recording it was measured on.
COMPRESSION DAMAGE · pb1 · 96 kbps
A lossy level is a BITRATE, exact on every window — but the DAMAGE it does is not. On pb1, 96 kbps measures 3.14-4.387 dB (1.397x) across the 9 windows serving it, the widest of this source's 9 levels. A threshold reported as "96 kbps on pb1" is a fact about this listener AND this material, not about the listener alone.
COMPRESSION DAMAGE · pb4 · 96 kbps
A lossy level is a BITRATE, exact on every window — but the DAMAGE it does is not. On pb4, 96 kbps measures 1.431-2.86 dB (1.999x) across the 6 windows serving it, the widest of this source's 9 levels. A threshold reported as "96 kbps on pb4" is a fact about this listener AND this material, not about the listener alone.
COMPRESSION DAMAGE · pb6 · 112 kbps
A lossy level is a BITRATE, exact on every window — but the DAMAGE it does is not. On pb6, 112 kbps measures 0.774-1.445 dB (1.867x) across the 9 windows serving it, the widest of this source's 7 levels. A threshold reported as "112 kbps on pb6" is a fact about this listener AND this material, not about the listener alone.
Two rungs that are not really two
The ladder is built so each rung is a fixed proportion harder than the last, and a rung has to be at least 1.15x its neighbour to count as a separate step. On these passages a pair of neighbours came out closer than that. The ladder still runs in the right direction here — the two rungs are simply not distinguishable from each other on this material.
COMPRESSION DAMAGE · pb1 · pb1@53 · 40 kbps
On pb1@53s, 48 kbps and 40 kbps measure 8.736 and 9.893 dB — 1.132x apart, below the 1.15x that makes two levels distinguishable. The ladder is monotone here but these two rungs are not separable on this window's material.
COMPRESSION DAMAGE · pb4 · pb4@318 · 80 kbps
On pb4@318s, 96 kbps and 80 kbps measure 2.86 and 3.234 dB — 1.131x apart, below the 1.15x that makes two levels distinguishable. The ladder is monotone here but these two rungs are not separable on this window's material.
COMPRESSION DAMAGE · pb4 · pb4@318 · 128 kbps
On pb4@318s, 192 kbps and 128 kbps measure 1.17 and 1.29 dB — 1.103x apart, below the 1.15x that makes two levels distinguishable. The ladder is monotone here but these two rungs are not separable on this window's material.
COMPRESSION DAMAGE · pb6 · pb6@0 · 80 kbps
On pb6@0s, 112 kbps and 80 kbps measure 1.445 and 1.522 dB — 1.053x apart, below the 1.15x that makes two levels distinguishable. The ladder is monotone here but these two rungs are not separable on this window's material.
The same rung, measured differently by different passages
Pitch and timing rungs are labelled with the manipulation itself, so they should measure the same wherever they are applied. These are the levels where they did not, to within the allowance the instrument gives.
PITCH DRIFT · 3.1 cents
The same level measures 2.7-3.07 depending on which window serves it (1.137x), consuming 91% of the 1.15x the instrument allows. A threshold reported at this level carries that much window-to-window variation.
Below what the ruler can read
Each family is measured with an analysis routine that has its own noise floor. A rung whose predicted magnitude sits under that floor is still reported, and it is the bottom of what this instrument can honestly resolve.
PITCH DRIFT · 3.1 cents
This level is rendered as a ramp peaking at 95% of its parameter, so it PREDICTS 2.94 cents of peak detune — below the 3-cent parameter floor the cents ruler was measured against. It is the bottom of what this instrument can report.
A source we do not present
One recording is rendered, validated, and deliberately withheld. pb6 serves the pitch and timing families, but its compression ladder spans only 3.5x across seven rungs — the narrowest in the pool. Simulated sessions put its fitted threshold 0.67 ladder steps off at the session length compression uses, and the best it ever reaches, at roughly 32 minutes of listening, is a third of the information per minute that pitch delivers. It is the worst use of a listener's time in the product, and no amount of extra time fixes it. The clips stay in the pool; the sessions do not use them.
Methodology · The Prestige Test · The Delicacy Trials · The Threshold Test · The Ranking Test