Content Quality: Well-structured 613-word News piece (range 400-1200). Title is 100 characters, within the 150-character cap. The headline correctly frames the central claim as 'Huawei Says', and the body consistently attributes figures and statements to Eric Xu or Huawei. No new market figure is asserted by the Herald itself.
Source Verification: Read all three snapshots from disk (gunzip -c, tag-stripped). Manifest: all three status 200, no archive fallback, suspicious_patterns null for all three (nothing to override). source-0.html.gz (The Register, Tobias Mann, 30 Sep 2026): CONFIRMED verbatim - Xu 'It's pretty hard to collect data about the market share of Nvidia in China, but based on the data we have collected, Ascend has surpassed Nvidia'; Register wording 'had exceeded Nvidia in Chinese market share' in 'a Q&A-style release this week' (so the measure is market share; no segment, period, shipments or revenue is given); 'We don't have enough capacity to satisfy the demand in China. We don't have plans to expand the international market in a fully-fledged way'; 'Even though our chips may be less advanced, at least their supply is assured'; 256,000-card Atlas 950 SuperCluster and 'as many as one million NPUs'; H200-series approval, shipments 'less than 1 percent of its datacenter revenues' per Nvidia's Q2 call; officials pressuring 'bit barn operators'; 'DeepSeek reportedly released several software tools this week aimed at making Huawei's Ascend accelerators more accessible to other AI labs'. The Register does NOT say the claim is 'Huawei's own estimate rather than independently audited data' (see correction 2) and contains no Nvidia comment on Huawei's claim. source-1.html.gz (geopolitechs.org): a genuine full English translation of the HUAWEI CONNECT 17 Sep 2026 journalist Q&A (Xu and Liao Heng), published 2 Oct 2026 by the Substack author (footer: Peng ZHANG), preceded by the author's five-point summary. It is NOT an official Huawei transcript; it is a third-party translation. Quotes verified verbatim inside the translated Q&A body (not in the author's summary): '...are under testing, and their large-scale supply will start at the end of this year or early next year' (Ascend 950DT, training); 'it's not about how hard we are going to push model providers, but about how much supply capacity we will have' (attributed to Xu, 'I think at the end of the day...'); 'a very severe shortage of supply compared to market demand'; 'will be achieved somewhere around 2029' with mainland China 'probably gonna be even slower'. Event date, venue and participants confirmed. The Q&A also has Xu stating Huawei has 'shipped more than 1,000 sets' of a 384-chip 910C SuperPoD (see correction 3). source-2.html.gz (GitHub deepseek-ai/DeepGEMM-Ascend README): CONFIRMED - 'DeepGEMM Ascend is a port of DeepGEMM to the HUAWEI Ascend platform. It is fully API-compatible with DeepGEMM and supports BF16, FP8, FP4 GEMM, MQA logits, and MegaMoE'; 'News 2026.09.30: Initial release of DeepGEMM Ascend, with support for Ascend 950 devices.' The page contains no mention of Nvidia or market share and the article uses it only for the port, as cited.
Factual Accuracy: Core claims are accurate to the sources and correctly attributed as Huawei's unaudited claim. 'Overtaken' = market share, per Xu and The Register, with no segment, period, or definition given; the article says so in 'What We Don't Know'. Supply-not-demand framing is supported by Xu's own words in both The Register and the translation. 950DT 'under testing' is verbatim. DeepGEMM-Ascend claims match the README. Three issues need a public note: (1) the Geopolitechs source is a non-allowlisted third-party translation and the 950DT timing, 'supply capacity' and 2029 shortage statements rest solely on it; (2) the lead attributes to The Register a characterization ('Huawei's own estimate rather than independently audited data') the Register does not make, and the article does not state whether Nvidia responded; (3) 'No cited source gives Ascend shipment volumes' is imprecise given Xu's 1,000-SuperPoD-set remark in the Geopolitechs translation.
Overall Assessment: Substantively sound, appropriately hedged, and every quote verified verbatim against the snapshots. Issues are minor and recoverable with a short corrections note, none touching the headline's factual basis. Verdict: APPROVE_WITH_CORRECTIONS.