Overview: The Eins System — A Research Measurement System for AI Influence

v1.112026-08-29

This page systematically sets out the positioning, object of measurement, scale design, and verification mechanisms of Eins (Empirical Influence Norm System); the content is grounded in the Eins White Paper v1.11 (2026-08-29 bilingual snapshot) and preserves its calibers throughout. The full text of the white paper (including proofs and appendices) appears in Methodology, and every key figure can be checked item by item in the mechanism-check section of the Index Panel. We hold that the measurement of influence ought to be a form of public knowledge: every definition, parameter, and figure is open for any reader to read, to examine, and to critique — this is at once the methodological discipline of this system and its commitment to the community.

The Research Positioning of the System

Eins (Empirical Influence Norm System) is a research measurement system: it attempts to measure, scientifically, fairly, and in real time, the influence that "AI Originators" — the communities that continuously create, release, and iterate AI models and thereby supply AI productive capacity to society — exert on society; its flagship output is the Eins Index (code 8100). Methodologically, the system presents itself as a construct-validity case rather than a dashboard: an influence index that cannot say what would falsify it is closer to propaganda than to measurement.

Problem Background: The Gap in Public Cognition

In 2026 the general public simultaneously faces two kinds of information: on one side, a stream of model names — GPT, DeepSeek, Claude, Gemini, Kimi, Qwen, among others; on the other, assorted benchmark rankings that purport to evaluate model "capability." Capability leaderboards describe models; product announcements describe vendors; neither describes the structure of influence: which communities are changing how society works, how fast, and in what direction.

Just as a market index lets a lay reader grasp the state of an economy, a well-constructed influence index should let a lay reader grasp the state of the AI landscape — equality of AI concepts and information is itself part of the public-good promise of AI. To the best of the white paper's knowledge, no public index of AI influence addressed to a general audience existed at the time of writing; the index family presented here is an attempt to fill that gap, and it submits itself to methodological scrutiny.

Capability Is Not Influence

The central premise of the system is that capability and influence are different quantities, and that confusing them corrupts public reasoning about AI. The white paper gives three arguments, each accompanied by its most serious counterargument:

Capability can be gamed: whenever a quantity is used as a target, the statistical regularities behind it tend to collapse (Goodhart's law variants, reward over-optimization, benchmark contamination). A capability score is precisely a quantity that can be optimized by engineering effort in a way that need not track whether models are used, discussed, and built upon by society — measuring "capability" measures a necessary input to influence, not influence itself.
Omniscient and value-based assessments are value-contaminated: any attempt to judge which models are "better for humanity" is forced to adopt a standpoint on what society should value. An influence measure should minimize the amount of evaluation it performs and instead record observable social fact: who is used, by how much, by whom, and with what consequence.
Value arises through use, discussion, iteration, and diffusion: a model is released; it is used in products and workflows; it is discussed publicly; it is iterated upon by developers and researchers; it diffuses into edge devices and new applications. The object of measurement is therefore a composite of these four channels rather than a single score of technical ability.

"Capability ⇒ influence" does not hold as a measurement identity: the first FA snapshot in Chapter 11 of the white paper exhibits this structure — the leaders in capability are not the leaders in the ecosystem.

The Object of Measurement: AI Originators

Because capability alone is not influence, and because influence flows from communities that produce and propagate models rather than from models in isolation, the system takes as its measurement unit the AI Originator — a class-level object, deliberately not a company, a brand, or a legal person.

AI Originator: a community that continuously creates, releases, and iterates AI models, thereby acting as a source of intelligence capability and supplying AI productive capacity and technological innovation to society; its operational and technical boundaries need not coincide with its capital, legal, or brand boundaries. (White Paper Definition 3.1)

The choice of unit deserves scrutiny: measure influence at the level of the model and you lose the organization that funds, hosts, and propagates it; at the level of the company you miss open-source collectives and joint initiatives; at the level of the country you average away the very communities whose behavior you want to observe. The first assessment covers twelve communities in the FA fundamentals domain (roster in Appendix C); the roster makes no claim to exhaustiveness: entry into the index is governed by inclusion rules (production-flow, coverage, identity and disclosure predicates, plus the no-backcast discipline), admissions take effect at a settlement and published history is never restated; candidates with insufficient coverage (e.g. ByteDance) are candidates for future inclusion and may first appear in the shadow set and the real-time layer.

The Measurement Scale: Relative · Centered · Real-Time

The scientific contribution of the system is a measurement scale rather than a single number. Influence is measured against a cross-sectional reference set (relative); a conservation law ties the sum of the index to the sum of its fundamental anchors, so transient discussion reorders positions but cannot enlarge the total pool (centered); a collection–pricing–display pipeline publishes at minute level (real-time).

8100 = FA + 3 \times (RI - \operatorname{avg}(RI)) + BR

The centered flagship composition of White Paper Eq. (1) is written Θᵢ(t) = FA_adjᵢ(t₀) + k_ri·(RIᵢ(t) − RĪ(t)) + BRᵢ(t), where RĪ is the cross-sectional mean over the roster and the value of k_ri is deliberately not disclosed. The panel check presents the same centered term with a fixed coefficient of 3 (shown as "3 × (RI − R̄)"); R̄ is how RĪ appears on the panel — the official INDEX_8102, i.e. the mean RI of the 10 orgs on the roster. BR is reserved for membership adjustments and is zero outside membership transitions (the panel counts it as zero and labels it honestly).

Conservation is the load-bearing property of the scale: exact conservation on the adjusted anchor — the sum of the flagship index over the included set equals the sum of FA_adj over the same set at every moment (White Paper Proposition 6.9). What can be summed, compared across time, and decomposed can be audited: a published single number is not a simplification that hides the science, but the smallest object on which an outsider can test the system's invariants.

Composition and Mechanism: FA Slow Anchor / RI Fast Signal

The quote of the flagship index 8100 is composed of three parts, each with a clean semantic boundary and never mixed in computation:

FA — the Fundamental Anchor (index 8101): the slow skeleton of capability, compositing capability, edge-model coverage, ecosystem stock, and commercial accessibility. It is stock rather than flow, with no time decay; it updates only at the two identical-weight settlement windows per day (UTC 04:00 and 16:00, three hours each) and is frozen intraday. The media/attention dimension has been removed from FA — attention alone cannot certify influence.
RI — Real-time Influence (index 8102): the fast signal of discussion and diffusion, centered and signed — heat only reshuffles relative ranks and never enlarges the total pool. Leaderboards first, topics as backfill; events decay by content-tier half-lives (leaderboard/factual/opinionary/emotional); without new stimulus a pulse decays to zero and never settles into FA.
BR — the bridge mechanism family: quote jumps caused by purely mechanical changes (e.g. membership adjustments) are bridged by a finite-term item that must reach zero; real fundamental changes are reflected immediately and are never masked by the bridge.

The algebraic structure of RI is given in White Paper §6.6 (eight-track decomposition, budget constraints); the numerical parameters (half-lives, volatility budgets, etc.) are not disclosed by design under the asymmetric-disclosure policy.

Trust and Verification: Asymmetric Disclosure

The system is only partially reproducible but wholly scrutable: its invariants (conservation, centralization, boundedness, audit inequalities) are stated as propositions with proofs and can be checked against public data, while the fast, easily manipulated components are kept opaque by design. FA methods and reference parameters are released with the paper and on GitHub (Appendix B); RI internals are reserved. Open what is hard to manipulate, close what is easy to manipulate — this is the institutional basis of the system's trustworthiness.

Scope and Non-Goals

The system does not adjudicate model quality, does not rank "best model for X," does not predict future winners, and does not assign responsibility in a governance sense. It describes an evaluation-method layer: it produces no investment information and constitutes no financial information, investment advice, credit assessment, reputation judgment, or basis for personal decisions; the current measurement reflects what has already happened, not a prediction of what will happen.