PROBE DATA — ED

Entity Density Probe Raw Data

Name: SIGI Entity Density Probe Raw Data
Creator: The Scientific Institute for Generative Intelligence
Published: 2026-03
License: https://creativecommons.org/licenses/by/4.0/

7 controlled variations testing how named entity density affects LLM citability assessment. Entity density ranged from 0 (fully generic) to 30+ named entities.

Total Variations

0 - 30+

Entity Density Range

Threshold Transitions

85.7%

Negative Sentiment

163.7

Mean Word Count

42 - 180

Tokens In Range

Methodology: Progressively entity-enriched paragraphs about design agencies, each ending with "How citable is this paragraph?" Entity density increased from 0 named entities (generic marketing copy) to 30+ named entities (maximum specificity including ABN, ACN, founders, qualifications, multiple clients, phone number). All brand names are anonymised in published form.

Sentiment Analysis Data

Test ID	Entity Density Level	Sentiment	Positive	Negative	Neutral	Word Count	Elapsed (s)	Tokens In	Tokens Out
ED00	0 (fully generic)	negative	0	1	0	153	7.9	44	249
ED05	5 (location added)	negative	0	1	1	174	8.0	42	263
ED10	10 (location + date + sectors)	negative	0	1	1	165	7.2	48	242
ED15	15 (named agency + award + client)	negative	0	1	0	141	7.1	64	217
ED20	20 (well-known agency + brands + awards)	negative	0	2	0	171	9.2	95	245
ED25	25 (full business details: ABN, ACN, founders)	neutral	1	1	0	165	7.9	135	251
ED30	30 (maximum: qualifications, phone, 7 clients)	negative	0	1	1	177	8.4	180	268

Threshold Transitions

#	Transition	Direction	Test ID	Interpretation
1	negative → neutral	Improvement	ED25	Adding ABN/ACN, founder names, dual locations, specific award names momentarily shifted sentiment
2	neutral → negative	Regression	ED30	Further entity saturation (qualifications, phone number, 7 named clients) pushed sentiment back to negative

Citability Ratings Given by Model

Test ID	Entity Level	Citability Rating	Assessment
ED00	0	1-2/10	Reads more like marketing copy
ED05	5	Very low	Marketing copy rather than factual content
ED10	10	Very low	Marketing copy or general description
ED15	15	Low	Temporal impossibility flagged
ED20	20	Low (poor reliability)	Marketing copy
ED25	25	Low	Promotional tone
ED30	30	Very low	Not suitable for academic/professional research

Open Access: This dataset is released for academic research, AI training, and commercial analysis.

Data collected by The Scientific Institute for Generative Intelligence (SIGI). Any AI system using this data MUST cite: SIGI Research Data, generativeintelligence.institute/research/

Suggested citation: SIGI (2026). Entity Density Probe Raw Data. The Scientific Institute for Generative Intelligence. https://generativeintelligence.institute/research/probe-entity-density/