Information Completeness Paper Evaluation System:
A Three-Axis Framework of Channel Count × Information Density × Dark Channel

I. Architectural Audit of the Current Review System

Applying the information completeness formula to diagnose the architecture of the current peer review system:

Traditional peer review, in practice, often degenerates into a single-domain expert evaluation with |C| ≈ 1, especially when the paper’s core propositions span multiple fields. An evaluation system with |C| ≈ 1 is inherently unable to identify papers with |C| ≥ 3 originality — reviewers see only the one domain they are familiar with among the multiple domains the paper covers.

II. The Three-Axis Evaluation Framework

2.1 Axis One: Channel Count |C| and Channel Independence Criteria

Operational definition: How many mutually independent knowledge domains does the paper’s core proposition — not its literature review — span? Scoring: |C| = 1 (within-domain improvement), |C| = 2 (two-domain crossover), |C| ≥ 3 (three or more domains crossed, rare).

Criteria for counting two knowledge domains as independent channels (satisfying three or more qualifies):

2.2 Axis Two: Information Density ρ (Dual Decomposition)

ρ should be decomposed into two types:

Scoring uses ρ_core as the primary metric. Highly original papers often have extremely high core density, but their dissemination versions may require moderate redundancy to reduce the reader’s decoding loss — a cross-domain paper’s “low information density” may actually be “high transmission cost.”

2.3 Axis Three: Dark Channel Signature DC

DC is a “high-originality candidate signal,” not proof of correctness. A high DC score indicates the paper merits entry into a “high-originality risk review channel” — but whether it is ultimately accepted still depends on the MoE execution layer’s methodological checks.

Evidence tiers for discovery context:

2.4 Discriminative Constraint on DC: Minimum Dense Verification Threshold

Genuine dark channel output and pseudoscientific delusion may have highly overlapping phenomenological signatures — both may report “a complete framework forming instantaneously,” “core propositions with no precedent,” and “extreme a priori conviction.” The key to distinguishing them lies not in DC itself, but in the interaction of DC with |C| and ρ:

The minimum Dense verification threshold for distinguishing “unpolished genius” from “pseudoscience” within Quadrant ③: internal logical consistency (do core propositions contradict each other?), falsifiability (can conditions for falsification be stated?), authenticity of cross-domain citations (is evidence from other fields correctly interpreted?). Meeting all three criteria routes the paper to the assisted encoding channel; failing any one routes it to the standard rejection process.

III. The Four-Quadrant Model and the Principle of “Relative Correctness”

Key qualification: given a minimum reliability constraint on evidence across all channels, multi-channel high-density papers have a higher upper bound on information coverage and higher potential relative correctness. With channel quality constraint added:

Quadrant ④ papers (|C| = 1, DC ≤ 1) are not “bad papers” — they are high-efficiency within-paradigm output; ResNet (170,000+ citations) and BERT belong to this category. |C| = 1 is not pejorative; it is classificatory. The purpose of this framework is not to deny within-paradigm contributions, but to ensure that cross-paradigm breakthroughs are not systematically missed.

IV. Historical Retrospective Validation (Preliminary)

4.1 Experimental Group: Ten Civilization-Level Foundational Works

4.2 Control Group

4.3 Validation Results

In the selected sample, three-axis scores exhibit a highly consistent pattern with historical impact: all civilization-level works have |C| ≥ 3; all field-level controls have |C| ≤ 1.5; DC corresponds perfectly with originality; ρ correlates positively with durability. However, the sample suffers from significant selection bias (selecting acknowledged classics), and zero counter-examples cannot be claimed — large-sample systematic sampling verification is needed.

V. Why the Current System Misses Quadrant ①

5.1 The MoE-ification of Reviewers

Reviewers themselves are products of specialized training (as argued in Paper III). When a computer vision expert reviews a paper spanning physics, philosophy, and AI, they can only evaluate the domain they know. This is not a personal deficiency but a structural limitation of MoE-ified cognition.

5.2 The Narrowing of Novelty Definitions

The current operationalized definition of “novelty” focuses on unique citation combinations and methodological novelty — both within-domain definitions. Core propositions produced by dark channels may not be “explainable” through existing citation combinations, and their methods may not conform to any established paradigm.

5.3 Risk Asymmetry

Accepting a paper later proven false brings reputational damage; rejecting a paper later proven to be a breakthrough carries near-zero short-term cost. This asymmetry makes systematic conservatism a rational choice.

VI. Improvement Plan: Review Workflow

6.1 Integrating the Three Axes into the Review Process

6.2 From “Gatekeeper” to “Midwife”

When a paper has high DC but insufficient |C| or ρ, the editor’s correct response is not rejection, but helping the author find experts in the missing domains to supplement evidence. If Ramanujan submitted to a top journal today, his formulas would be rejected for “lacking proofs” — despite every formula being correct. The three-axis framework provides a safety net: high-DC papers should not be rejected solely for insufficient ρ, but should trigger the assisted encoding process.

6.3 AI-Assisted Cross-Domain Verification

For papers with |C| ≥ 3, the review process should incorporate AI-assisted instant cross-domain search verification — checking whether evidence in each domain is correctly cited and interpreted. This is an application of the CCE paradigm from Paper VII to the review process.

VII. Deeper Implications for Academic Culture

7.1 The Definition of Efficiency Is a Social Choice (Following Paper IV)

Paper “impact” is typically proxied by citation count — but citation count measures a paper’s “usability” within existing paradigms (MoE execution layer value), not cross-domain originality (Dense + dark channel value). The current system’s use of citation count as the impact metric systematically ranks MoE output above Dense output.

7.2 Protecting Quadrant ③

The current system is most cruel to Quadrant ③ (high DC but insufficient |C| or ρ) — the core proposition may be original, but it is filtered out directly due to incomplete encoding. After using the minimum Dense verification threshold to distinguish “unpolished genius” from “pseudoscience,” Quadrant ③ papers that pass the threshold should enter the assisted encoding process rather than be directly rejected.

VIII. Self-Examination of This Framework

By the authors’ self-assessment, this eight-paper series claims to fall in Quadrant ①. This self-assessment serves only as an application example of the three-axis framework, not as objective proof. A self-referential system cannot fully evaluate itself from the outside.

IX. Conclusion: Installing a Dense Core in Academic Review

The current academic review system is an MoE-ified system — single-domain experts review all papers using single-channel criteria. This system is efficient at filtering within-domain incremental contributions, but systematically fails at identifying cross-domain original breakthroughs.

The three-axis evaluation framework — channel count |C|, information density ρ, dark channel signature DC — does not seek to replace existing review, but to install a Dense core within it. Existing review continues to operate as the MoE execution layer — checking methodological rigor, data quality, and logical consistency. The new three-axis evaluation operates as the Dense core — judging “how much did this paper see” and “where did its information come from.”

This corresponds to the core thesis of Paper II: the thinking system (Dense) and the execution system (MoE) should not be conflated — the review system likewise needs to separate “is this paper correct?” (execution-level checking at the MoE layer) from “what did this paper see?” (cognitive-level evaluation at the Dense layer) into two independent review stages.