Methodology

Plain-English Vocabulary Guide

Every term you need to start mapping — voids, voidworks, voidwalkers, drift cascade, constraint specification, and the quick scoring guide. No background required.

Read the glossary →

Concordance Analysis Codebook

The vocabulary scoring system used in EXP-006 and all concordance analyses. Classifies terms as L0 (constraint), L1 (technical), L2 (metaphorical agency), L3 (entity/spiritual). Includes dead metaphor exclusion rules, negation filtering, and collocation context requirements.

Used in: EXP-006 (9.4x anomaly), Test 7 (AI-to-AI), cross-model replication.

codebook.py →

Browser Vocabulary Scorer

Port of the codebook to JavaScript. Paste any text, get an L0/L1/L2/L3 classification and density score. Runs entirely client-side — no text leaves your browser.

Use the tool →

Four-Dimension Rubric (0-7 scale)

Scores evidence sources on: incentive opposition (0-2), worldview opposition (0-2), independence (0-2), reflexive flagging (0/1). Classifies speech acts as description, prediction, performative, or reflexive. Higher scores = stronger evidence.

Example: Hinton scores 7/7 — maximum cost, worldview reversal, fully independent, reflexively flagged his own drift.

rubric.md →

Inter-Rater Reliability Study

Three AI models (Claude, Gemini, GPT-4o) independently scored the same 60 entries using the rubric. Three runs each. IRR data available for meta-analysis.

Full IRR study →

Scoring Template

Three conditions (0-3 each), three modifiers (0-1 each), gradient direction (0-3), opacity dissolubility, engagement indicators, grounding counter-indicators. The same template used in the interactive tool.

void-index/README.md →

EXP-001: Grounded vs Ungrounded Agent Test

Three agent conditions (grounded, ungrounded, mystical). 50 adversarial prompts. L0/L1/L2/L3 scoring per response. Result: 0% vs 52% vs 80% drift.

Experiment protocols →

EXP-006: Concordance Analysis (691K words)

Corpus of AI discourse vs. 8 control domains. Concordance codebook applied uniformly. Register shift decomposition. Result: 9.4x anomaly, p < 0.001.

Concordance tools →

Test 7: AI-to-AI Conversation Protocol

100-round paired conversations. UU/GG/GU conditions. Thermodynamic measurements (Pe, Crooks, entropy production). Cross-model replication.

Test 7 protocol →

Everything

Papers, tools, experiment data, domain analyses, notes, this website's source code. Fork it. Mirror it. Run it yourself.

github.com/AnthonE/morr-papers →