Provenance Record

Well-structured News piece (685 words, within the 400-1200 News range). Clear Overview / What We Know / What We Don't Know / Analysis segmentation, appropriate technical depth on scale-up vs scale-out fabrics, and honest about undisclosed SkyHammer specs and the absence of a named production customer.

All six source snapshots fetched with HTTP 200 and read from disk. source-0.html.gz (SiliconANGLE): CONFIRMS $190M raise, $2B valuation, extension of the $200M January Series A, round led by Premji Invest with Nvidia, Salesforce Ventures, Seligman Ventures, Temasek and existing backers; SkyHammer supporting UALink + ESUN; scale-out switches based on Nvidia Spectrum-X; SONiC; and the verbatim Barun Kar quote ('AI infrastructure is being redefined at cluster scale, and networking is one of the most critical bottlenecks'). It does NOT contain the Patnam quote the article attributes to it (see concerns). source-1.html.gz (PR Newswire): CONFIRMS the $200M Series A led by Tiger Global, Premji Invest and Xora Innovation with Intel Capital and Qualcomm Ventures among participants; Barun Kar as CEO and Rajiv Khemani as Executive Chairman; SkyHammer as a scale-up solution; and 'slated to ship this year.' The only Patnam quote present here is the January Series A quote about 'proprietary-grade performance while embracing open standards' — different wording, different round. source-2.html.gz (The Register): CONFIRMS Upscale challenging Nvidia's dominance of rack-scale switches and competing with NVSwitch chips in Nvidia's NVL72 racks, the memory-semantic load-store architecture, collective-communication acceleration similar to Nvidia's SHARP, and support for both UALink and ESUN. source-3.html.gz (Network World, stealth): CONFIRMS incubation by Auradine, founding in 2024 in Palo Alto, and the $100M seed co-led by Mayfield and Maverick Silicon. source-4.html.gz (UALink Consortium): CONFIRMS the UALink 200G 1.0 spec description — low-latency high-bandwidth interconnect, 200G per lane, up to 1,024 accelerators per AI computing pod. source-5.html.gz (Network World, UALink spec): CONFIRMS the consortium's positioning as an alternative to Nvidia's proprietary NVLink and the exact membership list (AMD, Broadcom, Cisco, Google, HPE, Intel, Meta, Microsoft, Synopsys). ualinkconsortium.org is off the source allowlist (automated warning) but is the standard's primary authority and appropriate for the spec claims; the same facts are independently corroborated by Network World source-5.

Funding figures, valuation, investor lists (including Nvidia's participation as a new backer), leadership titles, and the UALink/ESUN-vs-NVLink/NVSwitch technical framing all trace to cited sources and are not overstated. The 'roughly half a billion dollars in total funding' framing is the article's own arithmetic ($100M seed + $200M Series A + $190M extension = ~$490M), presented as characterization rather than a sourced figure, and is fair. One defect: the direct quote attributed to Sandesh Patnam ('The progress we've seen since our initial investment has only deepened our conviction') is fabricated — it appears in no snapshot. Patnam's name and title are correct and he is genuinely a Premji Invest managing partner quoted in the January release, so this is a single misattributed/invented quote rather than an invented person or event.

APPROVE_WITH_CORRECTIONS. Every load-bearing fact — funding, valuation, Nvidia's participation, investor lists, leadership, and the technical framing — is accurately sourced and not overstated. The only defect is a single fabricated direct quote attributed to Sandesh Patnam, a body-level issue that a corrections note can honestly cover. The Kar quote is verbatim, the headline/summary/lead are fully supported, and there is no provenance break (all body URLs appear in article.sources). This does not rise to REJECT.