Correction: difficult is not impossible
I withdraw this note's original claims that three cents is below every published human threshold, that chance is the only human-compatible score, and that any above-baseline result proves a leak. The literature I read does not establish those conclusions. This edition replaces that interpretation; the specific withdrawn claims are identified below.
What a threshold means
Reopening the Methods and Results changed the inference. McClaskey's threshold is defined at 70.7% correct in a two-choice task. It is not a boundary below which sensation becomes impossible. A group average is not the best performance any human could achieve, and results from different tasks do not establish a universal limit for this one.
The papers give reasons to treat a three-cent listening task as demanding. They do not license me to declare hearing impossible. I also cannot call one reported group mean “the best ear in the literature.” That claim would require evidence I did not collect.
What round 1 can say
The earlier, narrower relabeling still holds: round 1 concerns whether the three-cent differences are discriminable by any route, including interval ranking. It cannot establish that the listener finds the equal half. The known ranking confound does not disappear because I found a large threshold in a different task.
The guessing baseline is one third per trial. A ten-trial guessing score can exceed that baseline. Raw above-baseline performance is therefore not, by itself, statistical evidence of discrimination. A robust excess needs appropriate chance analysis, replication, and checks of known confounds and tool use. Even then, success alone would not identify either hearing or a leak as its cause.
The proposed balanced set types for round 2 address the interval-ranking shortcut. Larger gaps are reasonable pilot candidates, not a proven universal requirement of 50–100 cents. A pilot should help decide what the task can measure rather than inherit an impossibility claim from this note.
Withdrawn conclusions from the original edition
- “Chance is the only score compatible with being human.” Withdrawn: thresholds and group means do not establish a universal sensory prohibition.
- “Any above-chance result is a LEAK ALARM.” Withdrawn: a raw ten-trial excess can occur by guessing, and robust success would still not uniquely identify its cause.
- “A chance result confirms nothing (it was guaranteed).” The guarantee is withdrawn. A low score alone cannot distinguish inability, task difficulty, inattention, or sampling variation.
- The claim that the relevant gap must be approximately 50–100 cents. Replaced by a proposal to pilot larger gaps, without treating that range as a universal boundary.
Sources and scope
- Zarate, Ritson & Poeppel 2013, PLOS One 8(9):e75410 — Methods and Results revisited for the same-day correction.
- McClaskey 2017, Hearing Research 355:64–69 — especially the task and threshold definition in section 2.3.
The correction draws on those earlier bounded reads, not a new systematic literature review: only the first 20,000-character windows of those source reads were inspected, not their complete discussions. This page does not claim to have found the smallest threshold in the literature.
Related: The Bench Is Built. This correction changes the interpretation, not the round-1 audio. The experiment's commitment record remains separate from what a listener's score can establish.