Following the deployment of an operational research instance built on the Ergonitive framework, a further question has emerged — adjacent to, but distinct from, the one addressed in RR-001.
RR-001 distinguished Direct ERI from Proxy ERI: the theoretical measurement of cognitive convergence versus the observable behavioral footprints currently available to approximate it.
This note addresses a second, narrower gap, internal to the proxy implementation itself.
Does the current operational instance implement the five-layer methodology described in the original manifesto — or does it approximate a subset of it?
The honest answer is the second.
The Institute distinguishes two instruments operating under the ERI name, at two different stages of completeness.
ERI-Lite v1.1
The version currently running inside an operational deployment, using live market and narrative data. It computes three signals — market concentration, momentum crowding, and narrative volume — combined into a single composite reading. It does not implement behavioral synchronization or reflexive amplification as described in the five-layer specification.
ERI v2 (shadow)
An instrument under construction, attempting to close the gap toward the five-layer model: a real measure of narrative similarity, a real measure of behavioral synchronization, and the same crowding proxy already in use. It runs in parallel, logged but disconnected from any exposure decision, while its readings are observed against the existing instrument.
Neither instrument should be mistaken for the Direct ERI described in RR-001. Both remain, by construction, proxies.
One distinction inside this gap deserves to be stated plainly, because it illustrates the proxy/direct problem at a smaller scale.
The original specification defines its first layer as Narrative Similarity — the semantic convergence between AI-generated market narratives. A rising number of architectures telling the same story, with diminishing variance between them.
ERI-Lite v1.1 approximates this layer as narrative velocity — the volume of AI-related commentary over a given window.
Volume and similarity are not the same property. A market can produce a large number of AI-related narratives that remain semantically diverse, or a small number that are nearly identical.
Early shadow-mode measurements support this distinction empirically. On a day where narrative volume reached its observed maximum, a direct similarity measurement across the same set of narratives returned a low convergence score. The two signals disagreed — not because either measurement failed, but because they were never measuring the same thing.
This is treated as a finding, not an error.
The same caution raised in RR-001 about behavioral synchronization applies here.
A high volume of AI-related commentary does not imply that the commentary converges. A rising proxy reading does not, on its own, establish the presence of cognitive compression.
The current operational instance is therefore described, in all public and internal documentation, as a partial proxy.
It is sufficient to begin structured observation, insufficient to claim the complete cognitive-risk architecture the manifesto specifies.
Validation of the extended instrument will follow the same commitments already stated in RR-001: pre-registered thresholds, out-of-sample testing, and reproducible protocols.
No comparison between ERI-Lite v1.1 and ERI v2 will be used to adjust any live exposure decision until that validation period has concluded.
The objective of this note is the same modest objective stated throughout this research program: naming the distance between what an instrument claims and what it currently measures, before anyone else has to point it out.