Provenance Record

Clean, well-structured funding News piece (602 words) that separates the deal facts from the company's self-reported model benchmarks and clearly flags the latter as vendor figures.

All 4 snapshots read from disk (HTTP 200). source-0 (The Robot Report) confirms the $400M raise, 2024 founding, San Mateo HQ, Radical Ventures lead, the full new/returning investor list (8VC, Union Square Ventures, Hanabi, Norwest; NVentures, Boldstart, Spark Capital, Bezos Expeditions, NFDG), the angel list (Eric Yuan, Bin Lin, Fei-Fei Li, Naval Ravikant), and the 'physical artificial general intelligence' framing. source-1 (SiliconANGLE) confirms the $400M / ~$2B valuation, and the Pete Florence (DeepMind, RT-2/PaLM-E), Andrew Barry (Boston Dynamics), and Andy Zeng (chief scientist) bios. source-2 (Robotics & Automation News) confirms the '99 percent reliability across … dexterous manipulation tasks, executed … up to three times faster' line, the April GEN-1 release, the November GEN-0 release, and the scaling-laws point. source-3 (generalistai.com blog) confirms the 99% / 64% (GEN-0) / 19% (from-scratch) success rates, '10x less task-specific data', ~one-hour adaptation, ~half-a-million hours of wearable-collected data, and the demo counts (T-shirts 86, blocks over 1,800, phones over 100).

Every figure, name, and investor traces verbatim to a cited source with correct attribution. Per Rule 9, the model-performance specifics (99%/64%/19%, 10x, half-a-million hours, demo repetition counts) are cited to the company's primary blog post, not to the trade write-ups. The article explicitly scopes these as company-reported, not independently evaluated. Apparent keyword mismatches on review (99% vs '99 percent', 10x vs '10 times', 'over 100' vs 'more than 100') were confirmed as verbatim-equivalent.

Fully-verified robotics-funding News piece with disciplined Rule 9 primary-source attribution for the model benchmarks. APPROVE. (PR also adds generalistai.com to the source allowlist as an official primary source.)