PhronesisThe kernel
Grant this first; it has measured signatures. To learn a thing is to lose access to not knowing it. You cannot, on demand, reoccupy the state you were in before. Two documented effects show the shape.
The curse of knowledge is the first. Camerer, Loewenstein, and Weber named it in 1989: once you hold private information, you cannot ignore it when modeling someone who lacks it. "Better-informed agents are unable to ignore private information even when it is in their interest to do so; more information is not always better" ("The Curse of Knowledge in Economic Settings," Journal of Political Economy). The bias is robust — market forces reduce it by roughly half, but never clear it. The informed cannot reconstruct the uninformed judgment; the earlier state is closed to them. Newton's tapping study (Stanford dissertation, 1990) is the vivid instance: tappers of a familiar tune estimated about half of listeners would name it; listeners identified 2 of 150. The tapper hears the melody and cannot subtract it back out.
Hindsight bias is the second. Fischhoff, in 1975, gave people an outcome and asked what they would have predicted without it: "receipt of such outcome knowledge was found to increase the postdicted likelihood of reported events … Judges were, however, largely unaware of the effect that outcome knowledge had on their perceptions. As a result, they overestimated what they would have known without outcome knowledge" ("Hindsight is not equal to foresight," J. Experimental Psychology: Human Perception and Performance). Once you know how it ended, you cannot recover the uncertainty you held before. The prior is gone and you cannot get it back to measure it.
The folk form is old and exact: you cannot un-ring a bell; you cannot un-see it. The instruments above show the perceptual case under measurement — the naive state is not available on demand.
The correction
Here is the spine of the note, stated plainly. Irreversible is not permanent, and the claim is firmer for the distinction.
Knowledge decays. Forgetting is real and lawful. Ebbinghaus mapped it in 1885 — the forgetting curve, retention dropping fast at first and then flattening toward an asymptote (Über das Gedächtnis; English Memory, 1913). So the claim is not that knowledge lasts forever. It plainly does not. The precise claim is narrower and thermodynamic in flavor: the reverse path is unavailable. You can lose a thing, but you cannot travel back along the road you came.
Forgetting is not the inverse of learning. Ebbinghaus supplies the proof, in his own method. Relearning a forgotten list is faster than learning it the first time — his "savings." A forgotten thing leaves a trace below the reach of recall: relearning is measurably faster even when explicit recall has dropped to zero, which detects a residual memory inaccessible to conscious retrieval (savings as a sensitive index of implicit retention; Murre, 2022). So the person who forgot is not the person who never knew. They differ by exactly the trace savings measures. Learning is one door. Forgetting is a different door — drift out of a room — never the way back in.
The framework reading
This section is a proposal, clearly its own, and it does not rewrite the framework's math. In the site's displacement language (the sheet), ordinary displacement is a distance from a fixed ground state. You are pushed off ground; you pay a cost of staying; you can, in principle, return — the ground was where it was, and the road back exists.
Knowledge is different in kind. It does not displace you from a fixed ground and let you pay your way home. It relocates the ground state itself. The baseline moves. That is the reading proposed here for why knowledge registers as irreversible: not that the displacement is large or costly, but that there is no old ground left to return to. Forgetting, on this reading, is drift away from the new ground — never reinstatement of the old one. The old ground is not far. It is gone.
Knowledge does not displace you from the ground. It relocates the ground.
Offered as a proposed reading of DC5-style irreversible displacement (published on Zenodo), not as an established result and not as a redefinition of the formal terms. The framework's ground state, displacement, and cost of staying keep their published meanings; this note only proposes where knowledge sits among them.
The limit
State the limit; it sharpens the claim rather than dents it. Not all knowing is equally irreversible.
A random phone number is near-fully losable — weakly reversible in practice. Declarative trivia decays toward the naive state, not to it; the savings trace remains, but for most purposes the loss is near-total. The claim is strongest for structural knowledge — knowing that reorganizes perception or understanding rather than adding a fact to a list. A resolved two-tone image that snaps into a face. A learned language. A spoiler that rewrites the whole plot. A proof once understood. These lock — durably rather than permanently; the induced perceptual lock shows measurable decay over weeks (Ludmer et al. 2011) and varies across observers, so "lock" names the asymmetry, not immortality. The reorganization holds, and the effort to reverse it is not the effort of forgetting but of somehow un-seeing a structure that now organizes everything downstream of it.
Procedural and tacit skill sits at the far, most-locked end. Polanyi: "we can know more than we can tell" (The Tacit Dimension, 1966). A skill you cannot fully articulate is a skill you cannot easily lose — savings is strongest here. So the gradient runs from the phone number, near-fully reversible, through structural insight that locks, to tacit skill that barely decays at all. The word irreversible is a claim about the top of that gradient, honestly stated.
The physics rhyme
One line, kin not pillar: information erasure has a thermodynamic cost and a direction — logically irreversible operations dissipate at least of order kT per bit (Landauer, 1961, IBM J. Res. Develop.) — which is the physics rhyme for a one-way process; see the surround paper for the reservoir it dumps into.
The live instrument sits at /field/mooney: see the source, watch the two-tone image resolve, and note that it stays resolved on re-exposure (the disambiguation persists; days–weeks documented, per the sources below).
Kin to Perception, Minus the Narrator — the resolved image stays resolved — and The Introspection Ceiling and The Silent Second Term, with the courses (module m04, irreversible displacement) and Mooney for the lock on screen.
Rests on: Camerer, Loewenstein & Weber, "The Curse of Knowledge in Economic Settings: An Experimental Analysis," Journal of Political Economy 97(5), 1232–1254 (1989) — the bias is robust, halved but not cleared by markets; Newton, "The rocky road from actions to intentions" (doctoral dissertation, Stanford, 1990) — tappers predicted ~50%, listeners identified 2 of 150 (cite as an unpublished dissertation; the popular "3 of 120" is the Heath retelling); Fischhoff, "Hindsight ≠ Foresight: The Effect of Outcome Knowledge on Judgment Under Uncertainty," J. Experimental Psychology: Human Perception and Performance 1(3), 288–299 (1975); Ebbinghaus, Über das Gedächtnis: Untersuchungen zur experimentellen Psychologie (Duncker & Humblot, 1885), English Memory: A Contribution to Experimental Psychology (Ruger & Bussenius trans., 1913) — the forgetting curve and the savings method; Murre (2022, PMC9971077) for savings as a "pure" measure detecting retention below recall; Mooney, "Age in the development of closure ability in children," Canadian Journal of Psychology 11(4), 219–226 (1957) — the two-tone closure items later called Mooney faces; the durable one-way disambiguation effect after seeing the source is established (Dolan et al., Nature 389, 596–599, 1997; durability of induced insight, Ludmer, Dudai & Rubin, Neuron 69(5), 1002–1014, 2011, which reports measurable decay over weeks; one-way framing, Flounders et al., eLife 8:e41861, 2019) — the qualitative asymmetry is uncontroversial, the "immediate/lifetime/universal" strength is an idealization; Landauer, "Irreversibility and Heat Generation in the Computing Process," IBM J. Res. Develop. 5(3), 183–191 (1961) — the exact kT ln2 bound has been debated (Earman & Norton) but the direction is mainstream; Polanyi, The Tacit Dimension (Doubleday, 1966) — "we can know more than we can tell." All of these results are established and cited as such. What is proposed is only the reading — that knowledge relocates the ground state rather than displacing from it, offered as a proposed reading of DC5-style irreversible displacement, to be argued with.