Notes.

Official TWIN-Treasuries 21 Sep: BUIDL $2.30bn / 108 / 25 after a $330m weekend drop. Holders stayed 108 → 108. Activity 27 → 25.

Names did not leave the book. Redemption pressure would have shown names leaving, not only AUM. This is the same divergence as the 14 Sep note, at a larger size.

Cite the official print: /twin/treasuries/2026-09-21.

In our v0 proxy formula, we applied a steep multiplier to the 7-day movement percentage in order to exaggerate small changes in highly stable Treasury products. The formula was clamp(50 + 25 × weekChange, 0, 100).

Over a US banking holiday, BUIDL saw a ~2% nominal dip in TVL. Multiplied by 25, this resulted in a -50 penalty, instantly driving the movement score down to 0. The proxy was floored.

While this behavior highlights why a simple multiplier is too fragile for a public proxy, it also demonstrated exactly the kind of edge case the TWIN production methodology is designed to smooth out using proper timescale denoising and volatility-adjusted weighting. For the Lab, we've softened the multiplier to 8 and constrained the floor to 5 to keep it functional as a toy model.

Echoing the recent TWIN print, we observed that despite the minor TVL movements in major tokenised Treasury products over the weekend, the concentration metrics barely budged.

When evaluating liquidity and redemption infrastructure, we look for divergence: does the capital pool shrink while the participant count drops? Or does it shrink while the participant count remains steady? The latter suggests routine treasury management rather than a structural exit.

The "crowd" (the spread channel) staying tight even as throughput dipped is a strong indicator of resilience. It shows the value of a dual-channel approach over relying solely on top-line TVL.