A spec sheet says “Sensitivity: 105 dB” and stops there. Per milliwatt or per volt? The maker does not say, and the two readings can be twenty decibels apart, which is the difference between a headphone a phone will drive and one it will not. This works out both readings, what each implies for your amplifier, and — the part no other calculator does — which of the two is plausible for a real transducer.
It is not a guess about the maker’s intentions. It is a comparison against every model in this database whose maker did print a reference unit. Those models mark out the range real transducers occupy, and a reading that falls outside it is almost certainly the wrong reading of the sheet rather than an extraordinary headphone.
Of the 567 non-electrostatic models here, 247 over-ear and 145 in-ear models carry a sensitivity figure whose unit the maker actually printed. Read per milliwatt, the middle 98% of those over-ears sit between 82.6 and 115.0 dB/mW, and the in-ears between 83.3 and 116.0 dB/mW. Those four numbers are the whole test: a reading outside the band for its type is rejected, a reading inside it is allowed to stand.
The ambiguity is not rare. 142 models in this database print a decibel figure and no unit at all, and for the median one of them the two readings are 14.9 dB apart — a factor of 31 in power. That is not a rounding difference; it is the difference between a dongle being ample and being useless.
Often they are, and then this tool will tell you so rather than pick one. That is the honest answer: the specification is genuinely ambiguous and no amount of arithmetic resolves it. What you can do is treat the per-volt reading as the pessimistic case, because it always implies the headphone needs more power, and buy accordingly. If the amplifier suits the pessimistic reading it suits either.
The size of the ambiguity depends entirely on impedance. The two readings differ by 10 × log10(1000/Z) — about 5 dB at 300 Ω, 15 dB at 32 Ω, and over 25 dB on a 3 Ω multi-driver in-ear. A bare decibel figure on a high-impedance headphone is a small problem; on an in-ear it is most of the specification. The converter does the arithmetic for any pair of units, and the spec-sheet audit counts how widespread this is.