The claim
A crystal grown slowly, close to equilibrium, is faceted. Its shape is reproducible and does not depend on the seed’s outline. A crystal grown fast, at high supersaturation, loses its flat faces to an instability: bumps grow faster than the surface around them, and the crystal branches into dendrites. Their form records fluctuations in the growth conditions.
An exchange with a chat assistant can run at very high supersaturation. Every reply offers more connections, terms and proposals than the turn asked for. The translation exchange shows what the branching looks like:
- In 51 minutes on 2026-10-10 (U49–U56, workbench 2026/10/10/040457), the thesis moved twice, to CLAIM-133 and then to the first form of CLAIM-139. At U50 a passing remark about FID was escalated into a THEORY, a new metric, three LITs, a CLAIM, a QUESTION and chapter placements, until the owner pulled it back at U51.
- Each of the two crystallized arguments (U38, U56; TERM-052) came at the end of a run of fast turns. The first lost material when it was drafted into C7 (CLAIM-324). The second, A203, left out Tannaka reconstruction (QUESTION-036) and CLAIM-125’s two open problems, and nothing in the exchange says whether they dissolved as the argument matured or were not carried forward.
The record acts on growth by anchoring, not only by pausing. Under the owner’s reframing (CLAIM-185), what lets structure grow in order is a persistent anchor for it to grow on. Drift then happens where material arrives faster than it is anchored: proposals that are never coded have to be restated, and restating is where redefinition creeps in. Filing between turns, the second-occurrence rule (CLAIM-176), and the other triggers of persistence each raise the anchoring rate. Pace matters because it sets how fast material arrives, and the ratio is what decides whether the argument faces over or branches.
The ratio can be measured. In a reconstructed exchange, count the turns in a stretch against the entries filed for it, and set that against the re-derivations and unannounced changes of thesis the workbench records for the same stretch.
What it does not say
- It does not say fast exchanges are bad. Dendrites grow fast, and the branching is where new material is found. The claim is about stability, not yield.
- It does not say the physics carries over. Morphological instability is an analogy here, and the claim is that drift tracks pace. That can be tested without the physics.
- One exchange, and no slow one to compare it with. The comparison that would test it does not exist yet.
Note of 2026-10-11 (later): settling does not rule out bursts
Generalized stability theory (LIT-931, LIT-929, and Trefethen et al., LIT-932, read on 2026-10-14 in NOTE-831, NOTE-842 and NOTE-839; THEORY-323, THEORY-325, THEORY-304) adds a qualification. When a system’s operator is non-normal, it can be stable in every mode and still amplify a disturbance by a large factor for a while before the disturbance decays. Trefethen et al. conjecture (“we speculate”), from a 2 × 2 model, that shear flows judged stable by their eigenvalues turn turbulent because a transient amplification hands the disturbance to nonlinear dynamics, which bootstraps it.
Read across, an argument that settles in the end can still show a burst of drift after a disturbance, and asymptotic settling is not evidence against the bursts. A record’s dependencies are directed: claims rest on claims. The record takes directed, asymmetric coupling to be a source of non-normality; that is its own reading, not the papers’ (Farrell & Ioannou’s source is a background shear, though their 2 × 2 example is a one-way coupling and fits it). If it holds, a change to a claim many others rest on could be amplified through them before the argument settles. Two consequences for this claim:
- Drift should be measured over the short run after a disturbance, not only as an average. An average can hide the burst.
- An argument under development is the nonautonomous case (LIT-929). Its structure changes as it is worked on, so there are no fixed modes to appeal to even in principle. Worse, time variation of a non-normal operator can make the system asymptotically unstable while it is stable at every instant, removing settling altogether, not only adding bursts. In the paper’s deterministic examples variation much faster than the transient growth is harmless, and damping stabilizes; anchoring is a candidate for damping.
The bearing is the assistant’s reading. Neither paper is about arguments.
Note of 2026-10-30: what holds it back
The test its own text names has not been run. No stretch of exchanges in the record has its turns per filed entry (CASE-183’s measure) set against the drift the workbench records for it, and there is no slow or differently anchored exchange to compare with the translation exchange. ARG-028, which would carry the measures to this claim, is Proposed.