Enflame S60 fabric - VERIFIED NEGATIVE ON FIRST-PARTY ACCESS: both of Enflame's own domains, including its hardware documentation centre, return a WAF 403, so no Enflame scale-up figure can be read from the vendor

NULL, AND IT MEANS 'UNREADABLE AT THE VENDOR', NOT 'NOT APPLICABLE' AND NOT 'NO FABRIC'. THE THREE-WAY DISTINCTION IS THE POINT AND IT IS WHY THIS IS A RECORD RATHER THAN A PARAGRAPH SOMEWHERE ELSE. (1) The vendor publishes a figure and it is readable - most records in this directory. (2) The vendor publishes nothing, or nothing about this - a verified negative of the 'published nothing' kind. (3) THE VENDOR MAY PUBLISH A FIGURE AND IT CANNOT BE RETRIEVED FROM HERE - this record, which is a third state that a null bandwidth on an ordinary record silently collapses into case (2). WHAT WAS TESTED on 2026-10-05, and every one of these is a first-party Enflame URL: https://www.enflame-tech.com/product-technologies/s60 (the S60 product page) and, newly, the vendor's DOCUMENTATION CENTRE on a different hostname, https://support.enflame-tech.com/onlinedoc_hw/5-s6x/S60/index.html and https://support.enflame-tech.com/onlinedoc_hw/5-s6x/index.html. ALL THREE RETURNED AN IDENTICAL HTTP 403 INTERSTITIAL of 837 characters reading 'img 403 / 您的请求已中断 / 您提交的请求可能对网站造成威胁，无法访问 / 可能原因 / Web应用防护服务检测您当前访问存在Web安全风险或访问不合规', attributed to '腾讯云WAF / 访问拦截' (Tencent Cloud WAF), each fetch carrying a DISTINCT request UUID (c41ee021397ef17e20235a33f6fa7862-f14e77c98299493bdfbfe601b55d1b0d843eaf and 5cb94611e238d2feb01ca55a36d1998a-d835b10389ebcfe601b55d1b0d843eaf). Distinct UUIDs prove a per-request firewall verdict rather than a cached error page. NO BANDWIDTH, PORT COUNT, LINK COUNT, TOPOLOGY, DIRECTION OR SWITCH WAS OBTAINED, AND NONE WAS ESTIMATED, BACK-COMPUTED OR CARRIED OVER FROM ANOTHER ENFLAME GENERATION. THE SECURE READING OF THIS RECORD IS 'Enflame's fabric cannot be read from here', NOT 'the S60 has no scale-up fabric' - the S60 is a multi-card training part and almost certainly has one; the vendor simply does not let it be read.

NONE OBTAINED. Not because Enflame stated there is none, but because both of its domains block retrieval. No topology, port, link or direction could be read from any first-party source on 2026-10-05.

NOT OBTAINED. Enflame names no switch part in any source readable from here, and absence of a readable source is not evidence of absence of the part.

FILED AS A VERIFIED NEGATIVE WITH THE URLS NAMED, AT DRAFT STATUS DELIBERATELY. The negative itself is solid - multiple first-party URLs, multiple client types, distinct WAF request UUIDs - so the absence of readable first-party interconnect data is well established. But status is draft rather than verified because what is established is a NEGATIVE ABOUT ACCESS, not a fact about the S60's fabric, and this repo's convention (AGENTS.md: 'A wrong confident record is worse than an open question') says an access-blocked vendor should not carry a verified record that reads like a specification finding. Confidence 0.7 reflects the same thing: high confidence that the pages are blocked, no confidence at all about the underlying hardware. THE NEW FACT RELATIVE TO THE EXISTING SOURCE RECORD: data/sources/w5a-enflame-s60-vendor-site-unreachable-waf-confirmed had already established that www.enflame-tech.com/product-technologies/s60 is blocked, using four independent client types. This pass adds the DOCUMENTATION CENTRE, support.enflame-tech.com, which was worth testing separately precisely because a subdomain can carry different WAF configuration from its apex domain. It does not - both paths return the same 403 body. That upgrade matters because the documentation centre is where an interconnect specification would live if Enflame published one: the S60 doc tree is titled '燧原S60推理' / '燧原S60 — 燧原硬件文档中心' and search engines index those paths, so content there is probably not absent, only unreadable from here. The consequence is a precise and defensible statement - Enflame's most detailed interconnect material is doubly inaccessible - rather than a vague one. WHAT WAS TRIED AND WHAT THE SECONDARY SOURCES YIELD: the TechInsights teardown (cn-enflame-techinsights) establishes the part's existence, the 'ENFLAME SCORPIO-AO' die markings and the TSMC N6NTO-HPC process, which is why data/accelerators/enflame-s60 has a process_nm of 6 from an independent teardown - but its quantitative content is behind a paywall and no fabric bandwidth is available from it. The aggregator mirrorfrog.com carries an S60 entry and the Chinese-language encyclopaedia baike.baidu.com has an S60 article; neither was used as a bandwidth source, because a scraped aggregator listing is precisely what AGENTS.md tells this repo not to cite, and a number lifted from one would be indistinguishable in this record from a vendor-published one. WHAT WOULD OVERTURN THIS RECORD: any readable Enflame first-party page, a hardware manual PDF hosted on a non-enflame-tech.com domain, or an archived capture of either blocked URL. Any of those would replace this with an ordinary interconnect record, and whoever finds one should delete this rather than amend it - the whole value of this record is that it says 'we looked at the vendor and could not read them'. KIND: filed as 'other' because the schema enum has no entry for an interconnect whose mechanism is unknown because the vendor is unreadable. 2026-10-10: null_bandwidth left unset on purpose. The basis says the first-party Enflame pages return WAF 403s, so the figure is unreadable, not absent. verified-negative means a fetched page that returned 200 and contained no spec section (iluvatar-bi-v150-and-bi-v250-fabric). sourcing means an aggregate was published. definitional means the fabric kind has no single link rate. None of those is 'the page could not be read'. Setting any of them would record a fact the fetch did not establish. The enum has no value for an unreadable page, and this record is the reason one would be needed.