VERIFIED NEGATIVE: MetaX publishes no compute figure for C500, C500X, C550 or N100 - vendor site unreadable AND no figure in the one readable MetaX-citing source

VERIFIED NEGATIVE, 2026-10-05, with the honest caveat that the vendor's own pages could NOT be read - which is a weaker negative than a completed read and is stated as such. WHAT IS BEING NEGATED. data/accelerators/metax-c500-n100 carries fp32 15, fp16 80, int8 160 at basis 'secondary-aggregator', source cn-metax-wikipedia, vendor_claim false - while its own notes and data/accelerators/metax-c550's notes both assert that MetaX publishes no compute figure. Those two statements are in tension and this record resolves it. METHOD 1 - the vendor's own product pages: UNREADABLE. https://www.metax-tech.com/en/goods/prod.html?cid=107&id=21 (C500) and .../id=35 (C550) both fail with 'All scraping engines failed to retrieve content from this URL. Engines tried: [playwright, fetch]'. The site is not merely anti-bot on a scraper: cn-metax-c550's existing note already records a plain probe returning HTTP 403. METHOD 2 - the Wayback Machine: a capture EXISTS and is readable. http://archive.org/wayback/available returned timestamp 20260609083954 for the C500 URL, and the capture renders. It confirms the vendor's own PRODUCT NAVIGATION - which is itself the useful finding, because it enumerates the entire current line-up: MetaX N-Series (N100, N260, N300), C-Series (C500, C550, C500X, C588, C600), G-Series (G100), X-Series (X206, X301, X302), plus matching servers and supernodes. So the vendor has pages for all four parts of this record and none of them is reachable from here. METHOD 3 - what the readable source actually says. The one MetaX document that CAN be read is the Wikipedia article (12,558 characters extracted). A sweep finds '15' 3 times, '80' once and '160' once - so the three stored figures DO appear there, which is why the citation is not fictitious. But the article attributes them to Digitimes reporting (Li 2025-01-22, Liang 2025-04-10), which is press coverage of a company, not a company publication. vendor_claim false and basis 'secondary-aggregator' are therefore CORRECT for these three rows, and the record's contested/draft status is right. THE RESOLUTION: MetaX's absence of a first-party compute figure is very probably true - no MetaX datasheet PDF has surfaced in any search, and the vendor publishes specification tables as HTML on pages that are unreadable from here - but it is NOT established by a completed read of the vendor's pages. What IS established is: (a) the vendor pages exist for C500/C500X/C550/N100 and are unreachable from this environment by three client types; (b) a capture of the C500 page renders and shows navigation only in the extracted text, with no specification table in the extractable body; (c) the three stored figures trace to press coverage via Wikipedia. A SECONDARY COMPILATION, recorded as a lead rather than evidence: a card-by-card compilation carries a much fuller C500 table - FP32 54 TFLOPS '(vector 18 + matrix 36)', TF32 140, FP16 280, BF16 280, INT8 560 TOPS, 64GB HBM2e, 7 MetaXLink ports to 64 cards - explicitly labelling process, TDP and bandwidth as undisclosed/estimated. Those numbers are 3.5x the Wikipedia-sourced ones on the same part, so the two disagree materially. I have NOT propagated them: a compilation that marks its own bandwidth and TDP as estimates is not a citable source for the un-estimated rows, and adopting its compute figures while its memory figures are estimates would import the estimate into a field that does not need one. The disagreement is recorded here so a future pass with access to the vendor page knows there is a 3.5x question to settle, not a gap to fill.