Content Quality: Well-structured News piece (Overview / What We Know / What We Don't Know / Analysis) at 988 words, within the 400-1200 News range. Neutral, non-sensational tone throughout; frames the discovery as preliminary per Anthropic's own caveats rather than overstating it.
Source Verification: Both cited sources were read from the local gzipped snapshots (not re-fetched), and uncompressed sha256 hashes were verified against manifest.json (source-0.html.gz: f90a1c4c...55c4 matched; source-1.html.gz: 61ed03e8...149c matched). Source 0 (anthropic.com/news/claude-discovers-novel-enzyme-system, source-0.html.gz): every claim and quote attributed to Anthropic in the article was checked and confirmed verbatim, including the specific figures under extra scrutiny — 'gathered over 200,000 RTs, picked out 3,500 new candidate systems, and narrowed those to the 20 most compelling candidates,' 'can take weeks to months,' 'After 21 hours spent searching this data by roughly 950 agents using 210 million tokens,' the agent's exclamation ('[The DNA next to the RT] is spectacular...'), the ART composition description, the Feng Zhang quote in full ('This is an exciting example...I hope this work encourages more scientists to explore how AI can support their research.'), Feng Zhang's title description, BSL-1/BSL-2 and 'All of the lab work is performed by human scientists,' and the Model Hardware Standard reference. All matched the snapshot text exactly. Source 1 (techcrunch.com/2026/09/23/..., source-1.html.gz): the Amodei/Stanford quote ('that is in some ways similar to the one Claude found') and the Amodei autonomous-lab-equipment quote ('eventually it may even be possible for Claude itself to safely perform the experiments...') both verified verbatim. However, two other quotes the article attributes to TechCrunch do NOT appear in the snapshot: (1) 'all physical experiments are conducted by human scientists' — TechCrunch actually wrote 'The physical experiments were done by humans'; (2) 'the AI processed approximately 210 million tokens using roughly 950 agents to search through biological data' — TechCrunch actually wrote 'Claude searched through data using about 950 agents that burned through 210 million tokens.' Both are logged as findings above. In both cases the underlying facts are correct and independently supported by the Anthropic primary source; only the quotation-mark wording is invented, not the substance.
Factual Accuracy: All specific figures in the article (200,000+ RTs screened, 3,500 candidates, 20 finalists, 21 hours, ~950 agents, 210 million tokens) trace verbatim to Anthropic's own announcement, the primary and authoritative source for this story. No numeric discrepancies found. The two fabricated TechCrunch quotes do not introduce any false information — they restate figures/facts that are independently true — but they misrepresent paraphrase as verbatim quotation, which is a sourcing-integrity issue that merits a public correction rather than silent approval.
Overall Assessment: Substantively strong, well-sourced News piece on a striking scientific claim; the primary Anthropic source was verified verbatim across every quote and every numeric figure under extra scrutiny. The only defect is two secondary TechCrunch quotes that paraphrase rather than reproduce the source exactly — a recoverable issue that does not touch the headline, summary, or lead and is honestly addressable with a corrections note. Verdict: APPROVE_WITH_CORRECTIONS.