Skip to main content

Methodology / Internal Audit

Research Quality Audit

We audited all 20 papers on this page against the evidence standards of Anthropic, Stripe, McKinsey, and a16z. 45% of our papers had zero clickable source links. None used inline citations. None disclosed methodology. Here is everything we found and our prioritized plan to fix it.

Polylogic AI Research|Polylogic AI|March 2026

We measured our research page against four benchmarks: Anthropic's research blog (inline hyperlinks, methodology disclosure), Stripe's guides (structured navigation, code examples), McKinsey's publications (survey methodology, raw data release), and a16z's thought leadership (named authors, education-first framing). The audit found systemic gaps in source documentation, citation practice, and evidence presentation. The existing content is solid. The evidence standards need to match the quality of the thinking.

The Benchmark

Across Anthropic, Stripe, McKinsey, and a16z, the minimum research quality bar includes: every factual claim links to its source (inline, not just endnotes), methodology is disclosed (how data was collected, sample size, limitations), authors or institutional attribution is visible, data is presented visually where appropriate, vendor-sourced data is disclosed as such, and limitations are acknowledged before conclusions.

Anthropic uses inline hyperlinks to prior work, figures with captions, and categorizes posts as Features, Workflows, and Field Notes. Stripe uses table-of-contents sidebars, card callouts, and embeds documentation standards in engineering career ladders. McKinsey discloses survey sample sizes and released 206 pages of raw data when their ACA survey methodology was questioned. a16z publishes with named authors and positions content to educate rather than sell.

Current State: The Numbers

MetricResult
Papers with clickable source links11/20 (55%)
Papers with ZERO clickable links9/20 (45%)
Papers using inline citations0/20 (0%)
Papers with methodology disclosure0/20 (0%)
Author attribution0/20 (0%)
Data visualization (charts/figures)0/20 (0%)
Limitations section16/20 (80%)
Summary callout box20/20 (100%)
Total sources cited across all papers~238 (avg 11.9/paper)

Evidence Quality Issues Found

Recycled unattributed statistics. The claim “42% of companies scrapped AI initiatives in 2024” appears in at least 3 papers without a traceable source in any of them.

Vendor-sourced data presented as objective. Multiple papers cite vendor marketing data (Oscar Chat, Keepme, Stammer AI) as findings rather than vendor assertions, without disclosing the source's financial interest.

Dated research cited as current. A 2011 Harvard Business Review statistic about lead response appears in multiple papers without acknowledging the study is 15 years old.

Conflicting data within papers. The White-Label AI paper cites two market projections that conflict ($82.46 billion by 2034 vs. $41.39 billion by 2030) without reconciliation.

Paper-by-Paper Assessment

Industry Guides are the strongest category (all have clickable links, consistent structure, limitations sections). Agent Orchestration is the single best paper. Methodology papers are the weakest category (4/6 have zero links).

PaperLinksGradePrimary Issue
Agent Orchestration18A-Analyst predictions not hedged
Lead Capture23B+Vendor-sourced claims need disclosure
Restaurant AI20B+$20.1B methodology unclear
Platform Integration8B+“92% of photographers” unsourced
D1 Sports Videography17B+7-site sample generalized broadly
Conversation Intelligence0C+Zero source links
Cross-Model Consensus0CDescribes system without proving it works
Agent Visual Design0C6 sources, none linked, unsubstantiated framework

The Fix: Prioritized Improvements

Priority 1: Add source links to 9 papers (~4.5 hours). The single most impactful change. A paper without verifiable sources is a blog post, not research. Papers needing links: Conversation Intelligence, Agent Visual Design, Visualizing AI Agents, Infrastructure Trust, Cross-Model Consensus, Structural Role Separation, Cinematic Portfolio Design.

Priority 2: Inline citations (~15 hours). Move from endnote-only sources to inline hyperlinks. When a paper says “42% of companies scrapped AI initiatives,” the number should link to the source.

Priority 3: Vendor disclosure (~3 hours). Add disclosure when citing vendor-sourced data. “Oscar Chat (a chatbot vendor) reports...” vs. the current “chatbots capture 3x more leads.”

Priority 4: Methodology sections (~10 hours). Add methodology disclosure to papers making empirical claims. Sample size, data collection method, scoring criteria.

Priority 5: Data visualization (~15 hours). Add at least one chart or figure to papers presenting numerical data. The booking funnel with dollar values at each stage should be visual, not prose.

Total estimated effort: ~57 hours across all priorities. The highest-impact fix (P1: source links) takes less than 5 hours and immediately moves 9 papers from unverifiable to properly sourced.

Sources

  1. Anthropic Research Page — research blog structure and citation practices
  2. Anthropic Alignment Science Blog — Features, Workflows, Field Notes categorization
  3. Stripe Guides — guide format and design patterns
  4. Mintlify: How Stripe Creates Documentation — writing culture and documentation standards
  5. a16z News and Content — thought leadership organization
  6. EI Exchange: a16z Thought Leadership Analysis — education-first content strategy
  7. BlogSEO: Inline Citations in Content Authority — citation credibility impact