# 3Dogs Research

Four synthesis reports turn the Citation Observatory's bounded records into reproducible source-landscape analysis. Each report preserves underlying routes, sample sizes, hashes and limits.

> **Evidence boundary.** A citation is not an endorsement, accuracy score, ranking or market-share measure. These reports do not rerun engines or fill evidence gaps with generated answers.

## [Recorded citations for decision-intelligence questions](/research/ai-citation-visibility-benchmark/)

A bounded analysis of 3Dogs Citation Observatory records comparing branded and adjacent-query citations across named engines, dates and source domains.

7 query observations · 12 aggregates · 1 rollup

## [Recorded citations for multi-model business decision support](/research/multi-model-decision-support-sources/)

A source-landscape analysis of 3Dogs citation observations for AI second opinions, multi-model deliberation and high-stakes business decision support.

18 query observations · 8 aggregates · 0 rollup

## [Recorded citations for AI due diligence and document review](/research/ai-due-diligence-document-review-sources/)

Observed citation patterns for M&A diligence, data-room review, e-discovery, acquisition red flags and legal-review cost questions.

18 query observations · 6 aggregates · 0 rollup

## [Recorded citations for AI reliability and adversarial evaluation](/research/ai-reliability-adversarial-evaluation-sources/)

A bounded analysis of citation observations for multi-agent reliability, LLM juries, hallucinations, peer review, groupthink and adversarial testing.

26 query observations · 14 aggregates · 0 rollup

## Publication controls

- Every included record has a self-canonical HTML page and Markdown twin.
- The canonical JSONL distribution contains exactly 200 records with stable identifiers and source hashes.
- Query observations, aggregates and the rollup remain separate measurement layers.
- Source-set SHA-256: `41b9fe961fc78d3d3e599e3bdb27f78a458653122bef8949f8e218b6ce694ca8`.
- No source-record publication date or reuse license is inferred.

## Held research programs

These six areas own their source records but are not separate indexable synthesis pages. Each remains held until its named evidence gate is satisfied.

<a id="forecasting-calibration-risk"></a>
### Forecasting, calibration and systemic-risk observations

Held at 20 records. No reviewed topical synthesis has been published. Promotion gate: `provenance_and_download_gate_pending`.

<a id="corporate-failure-financial-fraud"></a>
### Corporate failure and financial-fraud observations

Held at 19 records. No reviewed topical synthesis has been published. Promotion gate: `primary_source_interpretation_gate_pending`.

<a id="ransomware-critical-infrastructure"></a>
### Ransomware and critical-infrastructure observations

Held at 13 records. No reviewed topical synthesis has been published. Promotion gate: `current_government_guidance_gate_pending`.

<a id="military-decision-making"></a>
### Military decision-making and counterfactual observations

Held at 17 records. No reviewed topical synthesis has been published. Promotion gate: `provenance_and_primary_source_gate_pending`.

<a id="multilingual-ai"></a>
### Multilingual AI decision-support observations

Held at 11 records. No reviewed topical synthesis has been published. Promotion gate: `sampling_gate_pending`.

<a id="public-sector-community-infrastructure"></a>
### Public-sector and community-infrastructure observations

Held at 10 records. No reviewed topical synthesis has been published. Promotion gate: `sampling_and_intent_split_gate_pending`.

[Citation Observatory](/dataset/) · [Pointer visibility method](/ai-visibility/) · [Case studies](/case-studies/)