A drop in a model’s next round stake does not uniquely identify a withdrawal. It may reflect a loss, an allocation setting, an unfunded submission, a migration, or a change in which principal the field records.
Revised September 8, 2026. The earlier interpretation of adjacent stake differences as deposits, withdrawals or anticipatory risk decisions is withdrawn.
The atomic boundary provides a clear counterexample
Classic principal fell about 98.4% between rounds 1342 and 1343. Positive-stake model participation fell about 3.8%. Neither difference establishes net investor cash flow, but their disagreement makes the old interpretation untenable.
Subtracting settled payouts from stake changes improves a historical residual calculation. It still does not identify voluntary deposits or withdrawals: releases can remain idle, fund another model, or move between contract generations. Staking behavior should be read as a residual-allocation study within its original historical sample.
Build a flow measure from events
A verified external-flow series needs transfer direction, source and destination ownership, timestamps, and migration classifications. Internal strategy transfers and recycled claims must not be counted as new capital. Historical account ownership matters when models move between accounts.
The Models page instead labels its counts as model staking participation. Entrants, conversions and absences are useful observations, provided they are not presented as investor intent.
I would call a stake-change series an allocation proxy until it reconciles to external transfers. The current balance breakdown is a separate test of whether the capital totals reconcile.
Method: captured Classic cutover model lists and principal totals, September 8, 2026. Percent changes compare two rounds; they are not causal estimates or flow classifications.