Provenance Record

Well-structured News article on Qwen3.7-Max launch at Alibaba Cloud Summit. Appropriate scope at 721 words. Good context on leadership disruption and open-source strategy.

source-0.html.gz (scmp.com article 1): Confirmed 35-hour autonomous run claim, Liu Weiguang 'AI factory' quote, five-layer stack description, Zhou Jingren quote about top-tier benchmarks, 'long-horizon tasks' description. The '1,000 tool calls' and 'roughly 10x improvement in inference speed' claims are attributed to this source but NOT found in the snapshot — these appear to be unverified specifics. source-1.html.gz (scmp.com article 2): Confirmed 'trail behind top US products' quote, Qwen3.7 arriving a month after Qwen3.6, Arena preview ranking context. source-2.html.gz (decrypt.co): Confirmed 'Plus gets open-sourced, Max stays proprietary' characterization, Alibaba killed free tier of Qwen Code, Arena rankings (13th global, 7th math, 9th expert, 9th software), sixth-ranked AI lab globally. reuters.com source (source-3): HTTP 401 — snapshot unavailable. The Reuters claim about Lin Junyang's resignation and Foundation Model Task Force is unverifiable from snapshot; it is used only in the Analysis section as background context, not in headline/summary/lead.

Core claims are well-sourced. Two specific quantitative claims ('more than 1,000 tool calls' and 'roughly 10x improvement in inference speed') attributed to SCMP are not found in the SCMP snapshot — these are likely from a press release or demo details referenced by SCMP but not in the archived page. The Reuters background on leadership departures is unverifiable (401 block) but Reuters is a credible source and this is background context in the Analysis section, not a lead claim.

Solid article on a newsworthy AI release. The headline and lead claim (35 hours autonomous) are verified. Two subordinate specifics in the 35-hour run section are unverifiable from snapshots. APPROVE_WITH_CORRECTIONS is appropriate.