v1.2 — ERC-8004 dual-cap basis

What this is

Version 1.1 bounded how much any single address could contribute to the ERC-8004 identity count: one operator, however many agents it registered, could look at most like a barely-active registry. That cap was complete for the shape it targeted — concentration in one address. In early August a second shape appeared, breadth spread deliberately across many addresses, which a single-address cap does not by construction address. This release extends the same principle to that second axis. In doing so it surfaces something larger about the registry: it has been manufactured, not organic, for most of its history.

The change

Over roughly two days in early August, about 1,130 fresh addresses registered on ERC-8004 in a mechanical pattern — empty identity records, lockstep timing, funding traced to a handful of sources. This was manufactured breadth by a different method: not one address inflated, but many addresses coordinated. A per-address cap correctly reads each such address as small; the manipulation lives in their number, not their individual size. The value-jump safeguard held the figure rather than publishing it, and the question returned to review.

The response is not a more elaborate filter. It is the same rule applied to a second observable fact. Every manufactured cohort concentrates on one of two axes: many events from one address, or many events from one content source — a farm reuses addresses or it reuses content, and usually both. So v1.2 caps both. Each registration carries its address and its content-origin — the host serving its identity record, or a single shared key for the empty ones — and an event counts only while both its address and its origin remain under 100 events for the 30-day window. No new thresholds and no inference of intent: the same cap of 100, now on two independent signals rather than one.

The boundary of this approach is stated as plainly as the approach itself. An actor willing to distribute across many addresses and many content sources defeats both caps — no measure reading only public on-chain structure can catch that, and this release does not claim to. What bounds it is cost, not code: every registration is a gas transaction, so manufacturing genuine-looking breadth at scale is a standing expense with no endpoint, levied against a metric the index architecture already holds to a fraction of a point of influence. Should breadth ever arrive that is diverse across every observable dimension at once, the registry would be evidencing synthetic activity at every level it exposes — and the question then is not how to filter it but whether it still measures anything. That is a matter for review of the instrument, not another rule.

The numbers

Re-derived on the dual-cap basis across full history, the metric's calibration anchor moves from 606 to 460.5, and its launch-band position from about 20 to 17. This is the finding beneath the release: capping content-origins alongside addresses collapses not only the August cohort but earlier ones — a 10,000-registration batch in July, template-generated cohorts before it, an empty-record burst in March. Across the full series the dual-capped count runs roughly a third below the address-only figure. The concentration was never confined to one event; the registry has been chronically farmed, and the dual-cap basis reads through it without needing to determine, day by day, which activity was genuine.

Holding all other metrics at today's values and toggling only the ERC-8004 basis, the dual-cap changes the MEI by −0.4 (31.9 → 31.5). Because the index normalizes on fixed bounds and aggregates geometrically under a category ceiling, even this substantial re-derivation barely moves the headline. The bounds are unchanged; the nascent-floor exception ([15, 65]) that admits the metric below the standard launch band is unchanged and still applies. As in v1.1, the sub-band position is itself the reading: genuine technical-identity adoption remains close to absent.

The central finding is unchanged. Legal remains the weakest rail under every weighting tested.

Standing review

The conditions from v1.1 carry forward: re-review if concentration persists, immediate re-adjudication if a better-verifiable venue emerges, and automatic retirement of the nascent-floor exception once genuine breadth arrives (≥2,000 distinct registrants sustained for 90 days). One condition is added, and it is the ceiling of the ladder rather than another rung on it: if a content-diverse distributed farm is ever demonstrated — breadth manufactured across addresses and origins together — the response is a review of whether the metric still measures the construct it names, with retirement to a named gap on the table, not a heavier filter. The escalation ends in judgment about the instrument, by design.