How we score
Which notices. New Form D notices filed with the SEC that day. Amendments are skipped. Real estate, oil and gas, restaurants, hedge funds and real estate funds are skipped before any model sees them. Investment vehicles count only when their name points at one company that has its own website.
Is it AI? The evidence is the filing itself plus up to three public web pages whose web address or title carries the company's name; pages about other companies with similar names are dropped. A small model (Jev, on our free lane) answers the question twice, with the answer options in opposite order. It counts only when both runs agree. Then one more check that no model makes: a sentence on the company's own site has to show the AI itself (the word AI, machine learning, LLM, agentic, computer vision, autonomous, robotics and similar), and the company's own name does not count. Gated or preview profiles such as PitchBook, Crunchbase, LinkedIn or ZoomInfo are never used. Each card quotes that sentence. Anything else is unclear.
The company's own pages. For every company judged AI or unclear we read its site further: the pricing, integrations or partners, customers or case studies, docs or developer, and product pages it links from its homepage or lists in its sitemap (we never guess addresses), plus a GitHub organization it links to. Each card lists the pages read and the ones the site does not link. Unclear companies are asked again with those pages.
The moat score. The published Mode VC Distribution Moat rubric, the same code as the get_distribution_moat_score tool: start at 2. Add 1.5 if the evidence says the product runs inside tools customers already use, ships bundled or as a default, spreads through sharing, or gets more valuable as more users join. Add 1 if use generates proprietary data. Add 0.5 for three or more of those levers. Capped at 5. Each lever is asked twice the same way, and counts only when both runs say the evidence states it. Every lever then shows one of four states: found; evidence says no; runs split (one run found it, so it is not counted); or not in public evidence, which means unknown, not absent. Pricing model is recorded but the published rubric does not score it.
Read 2.0 carefully. A 2.0 means no lever was found in public evidence yet. Young companies rarely publish how they will win distribution, so 2.0 is a starting point for a conversation, not a judgment.
People and fund size. We join the people named in each filing to the partners in the Mode VC Investor Atlas. A confirmed match needs a second signal tied to the firm (its name in the filing, or the same address as its Form ADV). A possible match shares only the name and is marked not verified. Fund size is the private fund assets the firm reports on SEC Form ADV, shown only when that adviser match is confirmed.
Limits. A Form D is a notice the company files itself. It does not confirm that a round closed, who led it, or the valuation. Amounts are as reported.
For AI agents
Call get_daily_ai_raises on the Mode VC MCP server at https://modevc.com/api/mcp with an optional date (YYYY-MM-DD). It returns the same raises, scores, reasoning, evidence links and partners, with as_of and source URLs on every record. Recent days: 2026-10-20 (1 AI) · 2026-10-19 (6 AI).