Field Notes
What AI believes. What is actually true. The gap between them.
Weekly reports from the measurement layer. Named brands, live data, dated cuts — never screenshots.
Cross-model coverage gaps
January 7, 2026
Western AI Blind Spots: 7 Chinese FPGA Competitors GPT-4 Cannot See
GPT-4, Gemini, and Groq agree Xilinx dominates FPGAs at 95.8. DeepSeek finds 7 Chinese competitors the Western models miss entirely.
Read the field note →
AI memory persistence
January 7, 2026
Ghost Brands: AI Still Recommends Companies That No Longer Exist
$117B in semiconductor acquisitions. Xilinx, Linear Technology, Freescale — acquired years ago, still scoring 70+ across 11 AI models.
Read the field note →
Inter-model divergence
December 30, 2025
Google Rates Bing 98. OpenAI Rates Bing 80.
The competitor is generous. The partner underrates. An 18-point gap across 11 AI models that should not exist.
Read the field note →
New notes ship as the weekly LLM run produces material worth framing. Each note is reproducible from the public data referenced in it — see methodology for the collection cadence and the data contracts.