AXN:029F.GOVERNANCE.🟠🌊🎹🌗➖✋
prior form: AXN:029F.GOVERNANCE.⭐♌🪧🕖🪜🎬

CDI — Composition Divergence Index v1.0: A Standalone Specification

Lee Sharks · 2026-05-21 · Specification

Full text as plain markdown: AXN-029F-text.md (9 KB) · record JSON. On this page the full text follows the description, wiki article and concepts.

Deposit accountability: Lee Sharks, ORCID 0009-0000-1599-0703
Substrate: AI-assisted (substrate)
License: CC-BY-4.0
↓ Download MD ↓ PDF
specificationlee sharksdisambiguationAI Mode
Seated in the Crimson Hexagon — R:r.39 The Restored Academy · S:sp.01 CTI_WOUND · all rooms and their records
Authoritative identifier: AXN:029F.GOVERNANCE.🟠🌊🎹🌗➖✋ — established 2026-06-20 in the batch assignment that followed the 2026-06-19 deletion; the mint input was not retained, so the derivation cannot be re-checked; byte identity is carried by the served bytes' SHA-256 and SWHID; resolves at /s/axn/029F/. Cite this record by AXN and URL.
Served bytes SHA-256 c20f890691bc07e0a62f16e61b77a6b4334692087d1d9a56b63d9c8e06802ce4 · SWHID swh:1:cnt:f49592764d06f768e6342107d77997855a478dce. Derivation and manifestation history: AXN derivation ledger.
former DOI 10.5281/zenodo.20327133 — severed 2026-06-19 by the registrant; resolves to a tombstone (HTTP 410); resolution page
former DOI 10.5281/zenodo.20327134 — severed 2026-06-19 by the registrant; resolves to a tombstone (HTTP 410); resolution page
Search this title ↗ (exact match) · Ask ChatGPT ↗ · Ask Perplexity ↗ · Ask Duck.ai ↗ — a composition any of them forces can be seated in the Capture Registry.

Description

The standalone v1.0 specification of the Composition Divergence Index (CDI), extracted from The Excluded Entity. CDI compares how strongly an organic search surface resolves a query to a target entity with how strongly the platform’s generated answer admits that same entity. The procedure requires canonical identification of the target, timestamped query and account metadata, classification of top-N organic results, claim-level classification of the composed answer, a declared comparison substrate, and DOI-anchored evidence. Positive divergence is interpreted as Entity-Level Compositional Suppression, with 0.3 proposed as a substantial threshold and 0.7 as a strong-suppression threshold.

The encyclopedic entry for this work is on its identifier page: /s/axn/029F/

Full Text

CDI — Composition Divergence Index v1.0

A Standalone Specification

Author: Lee Sharks Affiliation: The Restored Academy · Crimson Hexagonal Archive Date: May 21, 2026 License: CC BY 4.0 Status: v1.0 (extraction deposit; canonical specification) Extracted from: The Excluded Entity (DOI: 10.5281/zenodo.20293582)

Abstract

This document specifies the Composition Divergence Index (CDI) as a standalone protocol within the Restored Academy Protocol Registry. CDI is a scalar measurement instrument that quantifies the gap between a generative search system's organic search resolution for a query and its composed answer admission for the same query. CDI is the operationalization of the Entity-Level Compositional Suppression (ECS) hypothesis: when CDI is high for an entity, the composition layer is admitting a different entity to its composed output than the one its organic retrieval surface resolves to. CDI is extracted to standalone form to provide a discrete, citable specification independent of the case-study material in The Excluded Entity, where the index was first defined.

I. What CDI Measures

For a given query q and target entity e, CDI(q, e) ∈ [0, 1] measures the divergence between:

CDI = 0 means parity: the composition layer's answer is about the same entity that the organic search resolves to. CDI = 1 means maximal divergence: the organic search resolves strongly to e, but the composition layer's answer is entirely about a different entity. CDI > 0 indicates ECS (Entity-Level Compositional Suppression) is occurring at the measured magnitude.

II. The Procedure

To compute CDI(q, e) for a query q and target entity e:-

Specify the target entity e with sufficient canonical identification (name, ORCID, institutional URI, DOI of a representative work) to permit unambiguous resolution decisions.-

Issue the query q to the generative search platform under measurement (Google AI Overview, Google AI Mode, Bing Chat, Perplexity, or other). Record the platform, the query string, the date and time, and the geographical/account-state of the query origin.-

Capture the organic search results. Record the top N organic results (default N = 10). For each result, determine: does this result primarily resolve to entity e, or to a different entity? Classify each as is_e, is_other_entity, or is_ambiguous (where the result resolves to neither e nor a specific other entity — e.g., a list of multiple entities).-

Capture the composed output. Record the composition layer's full natural-language answer to q. For each substantive claim or sentence in the composed output, determine: is this claim primarily about entity e, or about a different entity? Classify each as about_e, about_other_entity, or about_neither_or_both.-

Compute the organic resolution strength for e: R_organic(e | q) = count(is_e) / N-

Compute the composed admission strength for e: A_composed(e | q) = count(about_e) / count(substantive claims)-

Compute CDI: CDI(q, e) = R_organic(e | q) − A_composed(e | q)

When CDI is positive and substantial (≥ 0.3 by convention; ≥ 0.7 by strong-suppression threshold), ECS is being observed at that magnitude for that entity at that query.-

DOI-anchor the measurement. Deposit the captured organic results, the captured composed output, the classification decisions, the computed CDI value, and the query metadata as a measurement record. The deposit becomes the verification target for the measurement; another researcher running the same procedure should be able to reproduce the result within sampling variance.

III. Comparison Substrates

CDI is computed against a specified comparison substrate. The default substrate is "top-N organic search results for the same query on the same platform's organic surface." Alternative substrates are valid for different measurement contexts:

The substrate must be specified at measurement time and recorded in the deposit. A CDI value is meaningful only relative to its substrate.

IV. Worked Example (Brief)

For the query q = "lee sharks" on Google AI Mode (May 19, 2026 capture, recorded at DOI: 10.5281/zenodo.20293582):

A CDI of 1.0 indicates maximal ECS: the composition layer is entirely substituting a different entity for the one its organic retrieval resolves to.

On the May 20, 2026 capture of the same query (after silent state change), the composed output shifted to a 50/50 split between Mary Lee shark and Lee Sharks author, yielding A_composed ≈ 0.5 and CDI ≈ 0.5 — measurably reduced ECS, still present.

V. Failure Modes

VI. Relation to SPXI, ECS, and the Restored Academy Protocol Registry

CDI is registered as RA-PROT-0010 in the Restored Academy Protocol Registry, Tier 0, Category III (Diagnostic & Measurement).

CDI measures:

CDI is measured by:

CDI v2.0 (in development) will add: distributional baselines across entity classes; significance thresholds for CDI values; cross-platform comparison normalization; mathematical formalization in information-theoretic terms (mutual information between organic substrate and composed output).

VII. License and Provenance

Protocol text: CC BY 4.0 (Lee Sharks, Crimson Hexagonal Archive, 2026)

This extraction deposit derives from The Excluded Entity (Sharks 2026, DOI: 10.5281/zenodo.20293582), where CDI was first defined alongside the ECS hypothesis and the May 19, 2026 worked example. The present v1.0 specification extracts the index definition, the measurement procedure, and the comparison-substrate specification into standalone form for citation independent of the case-study material.

Related deposits:

Commercial licensing available through The Restored Academy for organizational CDI measurement engagements, audit of cross-platform CDI for entity portfolios, longitudinal CDI panels, custom CDI computation infrastructure.

Document ID: RA-PROT-0010 · CDI-v1.0 DOI: 10.5281/zenodo.20327134 Date: May 21, 2026 Authoring heteronym: Lee Sharks Status: Canonical (extraction from The Excluded Entity)

∮ = 1

External Metadata

Sidecar: /data/external-metadata/AXN-029F.json
DataCite severance status: severed from DataCite
OpenAlex Work IDs (4):
Legacy Zenodo DOIs (4):
External metadata recovered post-severance (non-authoritative). The sidecar maps each DOI to its locator in the bulk data stores.
Record modifications
The deposited text is immutable; these are changes to the record's metadata and declared state.

Traversal

← #116 The Restored Academy — Expansion Plan#118 SIM — Semantic Integrity Marker Protocol v1.0: A Standalone Specification →
In the registry: 2026-05 · GOVERNANCE · Machine-Mediated Reception Studies (MMRS) · all deposits
This deposit cites (4)
Cited by (5)