Core Definitions

These three definitions are locked across all Zorthex outputs. They cannot be modified between reports, between versions, or between clients. Any change requires a new version number and full dataset recalculation.

t_start Source Policy — Levels A–D

Every t_start used in a Zorthex report must be justified with a documented primary source. Sources are classified into four defensibility levels. The level does not determine the choice — the date closest to the actual operational moment prevails. The level determines how defensible that choice is.

Not acceptable as t_start sources: Wikipedia, undated web articles, AI-generated summaries, secondary aggregators, sources without a verifiable publication date.

Google Trends CSV Policy — Snapshot-Locked Data

Google Trends produces normalized scores (0–100) relative to peak search volume within the selected time period and geography. The same query run on different dates or with different windows produces different absolute values. Zorthex handles this through snapshot-locking.

Cluster Renormalization & the Rock Rule

Because Google Trends normalizes every series against the maximum of the downloaded window, a large recent surge in attention rescales the entire history of a term downward. A phenomenon that was already structurally active years earlier can appear, in a freshly downloaded CSV, to have "broken out" only recently. This is the cluster renormalization effect, and it is most pronounced for terms caught inside a synchronized attention wave (in this dataset, a broad August 2025 cluster across multiple domains).

The rock rule is why the dataset separates stable structural cases from observation-stage cases on the basis of where the peak sits in time, not on the raw score alone. It is also why a second and third source are required: an independent signal that anchors a phenomenon earlier than its Google Trends breakout is direct evidence that the recent surge is a renormalization artifact, not a genuine first emergence.

Three-Source Verification

Google Trends alone is insufficient. Each case in the v2.0 dataset is positioned using three independent attention signals, each measuring a different kind of attention. They are not averaged into a single number; they are read together.

What the second source corrects

Wikipedia pageviews are the corrective to Google's renormalization. For phenomena such as generative AI and deep learning, Wikipedia shows high and constant readership since 2023 — revealing them as mainstream and structural even where a freshly downloaded Google Trends series, rescaled by a later surge, would understate them.

The Dual-Velocity finding

For B2B-infrastructure and policy-driven phenomena, Wikipedia and Reddit signals are systematically lower than Google Trends scores. This divergence is declared as a finding, not corrected as an error: it reflects the structural difference between operational attention (active search by professionals who must use the tool) and informational or community attention (the mainstream public, not yet reached). Attention moves at two velocities, and Zorthex frequently intercepts a phenomenon while it is still confined to technical departments — before it becomes mainstream news. Cases where this pattern holds are flagged as dual-velocity in the dataset.

The 25/100 Threshold

The threshold of 25/100 on Google Trends (Worldwide, normalized) is the activation point for structural classification. It is empirically set based on sensitivity testing across the dataset.

How it was tested

Tested at 20, 25, and 30 across the historical dataset. At 25, the separation between confirmed structural phenomena and confirmed transient peaks is most stable. At 20, confirmed bubbles pass the threshold. At 30, confirmed structural phenomena like Bitcoin fail the test.

Classifications

A fourth state, PRE-dataset / non-stationary, is used for cases whose origin precedes the Google Trends data window (2004) or whose curve is dominated by an exogenous shock (e.g. a pandemic). These are retained for completeness but excluded from L averaging.

4-Regime Taxonomy

Beyond classification, each case is assigned a regime describing the shape of its diffusion: Policy-Trigger (slow ramp, breakout on a regulatory trigger), Institutional Mass (multi-year invisibility then institutional adoption), Market-Narrative (speculation- and narrative-driven cycles), and Shock/Spoke (single-event spike then collapse). The regime is a structural property of the thematic area: knowing the area lets one read the likely regime by analogy — a descriptive read of shape, never a prediction of an individual case.

Products — What Each Promises

Zorthex offers three distinct products. Each delivers a different type of output with a different data standard. The distinction between AI-estimated and manually-verified outputs is the foundation of this architecture.

How AI Is Used in This System

Zorthex uses Claude (Anthropic) as the analytical engine for the free app and as a support tool for report drafting. This is declared explicitly and always.

On "Powered by Claude"

Zorthex declares its use of Claude as a matter of transparency, not marketing. Claude is the tool. Zorthex is the method. This distinction is intentional and permanent.

Declared Limitations

These are not disclaimers. They are the most operationally honest part of this document.

Version Changelog