A sell-cascade count belongs to its definition

The reported 3,878 DX Terminal Pro sell cascades belong to a specific event definition: at least ten vaults selling the same token within ten minutes. Changing the count threshold or time window changes the question. We should preserve that definition beside the number whenever we use it to characterize the historical deployment.

The published observation extract records the threshold, same-token requirement and ten-minute window. It describes a bounded 21-day market with 12 tokens, one frozen production prompt and harness, and one model family. Those scope fields matter as much as the cascade count because the market periodically eliminated tokens and changed the available opportunity set.

Membership can change at a boundary

An illustrative token has ten distinct selling vaults between 12:00 and 12:09. It meets the ten-in-ten rule. If the final vault instead sells at 12:11, the result depends on which ten-minute interval contains the other nine sales. A fixed-window implementation and a rolling-window implementation can classify boundary cases differently. Exact reconstruction also requires clustering and deduplication choices absent from the aggregate definition.

Likewise, nine vaults selling repeatedly would fail a distinct-vault threshold even if they generated a hundred transactions. Ten vaults selling once each would meet it. This is why a transaction count cannot silently replace the published vault threshold. The observation describes collective participation, while transaction frequency is a separate measure.

We would build a sensitivity artifact with columns for minimum distinct vaults, interval length, interval construction, overlap handling and event count. Each row would reuse the same saved sales. A proposed sweep might compare five, ten and twenty vaults across five, ten and twenty minutes. These diagnostic settings are analyst-selected proposals awaiting measured results. Reporting the complete sweep would reveal whether the description is stable or largely a product of one boundary.

Avoid converting events into incidence

Dividing 3,878 by 21 gives approximately 185 reported events per day averaged over the deployment. A typical day's rate requires the daily distribution. Cascades can cluster, overlap or share participants, and the number of remaining tokens changes through the event. Without a daily distribution, the average cannot show uniform recurrence or independence.

The controls paper companion also reports a large buy concentration and 92.9% of trades in two-sided token windows. These measures use different windows and units. A sell cascade can occur in a window that also has buying, so the categories can overlap.

For readers evaluating agent behavior, the practical question is whether an alert or comparison would survive reasonable definition changes. Our proposed sensitivity table answers that question without erasing the original result. The continuous record companion makes methodological disclosure part of the research lineage; a future cascade study should carry its counting code and boundary tests beside its headline.