Provenance Record

Well-structured News piece (546 words, within the 400-1200 range). Overview / What We Know / What We Don't Know / Analysis framing is appropriate for a corporate topline readout and is unusually disciplined: the article repeatedly signals that only high-level results have been released and that magnitude of benefit is unquantified.

All 4 sources read from disk snapshots plus one live fallback. source-0.html.gz (Roche press release, roche.com): CONFIRMED the 2 July 2026 date, Krascendo 1 name, head-to-head vs approved first-generation sotorasib/adagrasib, NCT06497556, 338 adults, once-daily divarasib / once-daily sotorasib / twice-daily adagrasib dosing, BICR-assessed PFS primary endpoint, OS/confirmed objective response/duration of response secondaries, 'no new safety signals', Breakthrough Therapy Designation 2022 and Orphan Drug Designation 2026, and the Levi Garraway quote verbatim, plus the exact 'met its primary and key secondary endpoint ... in both PFS and OS' quote. source-1.html.gz (GlobeNewswire ad hoc announcement): CONFIRMED — identical Roche release text, verbatim match for both quoted passages; note the release header adds that OS statistical significance was reached 'at the interim analysis', which is consistent with (and slightly more precise than) the article's framing. source-3.html.gz (OncoDaily): CONFIRMED the global/randomized/open-label descriptor, previously treated advanced or metastatic KRAS G12C NSCLC, BICR PFS primary + OS/ORR/DoR secondaries, covalent inhibitor binding mutant KRAS in the inactive state, and the 'has not yet disclosed median survival outcomes, hazard ratios, response rates, subgroup analyses' line attributed to it. source-2 (MedCity News, medcitynews.com): snapshot FAILED with HTTP 403 (bot-blocked), file null in manifest; verified via live WebFetch as a last resort — CONFIRMED all three attributed claims verbatim: patients progressed after 'at least one but no more than three prior lines of systemic therapy', comparator arms 'Amgen's once-daily Lumakras' and 'BMS's twice-daily Krazati', and 'next-generation KRAS G12C inhibitor designed to offer greater potency and selectivity compared to the Amgen and BMS products'.

No hallucinations detected. Critically, the article does NOT fabricate any hazard ratio, median PFS/OS, confidence interval, response rate, or subgroup figure — every efficacy specific Roche withheld is explicitly listed as undisclosed in the 'What We Don't Know' section, matching the OncoDaily and Roche source text. The only numerals in the body (338 patients, 1-3 prior lines, 2022, 2026) each trace to a cited source. The headline 'beats ... head-to-head' is a faithful summary of Roche's own 'shows superiority in head-to-head phase III trial' and the disclosed statistically significant PFS and OS improvements.

APPROVE. A disciplined, accurately sourced report on a corporate Phase 3 topline that scrupulously avoids inventing the efficacy numbers Roche has not released. All specifics, quotes, the trial name, the NSCLC/KRAS G12C indication, and the 2 July 2026 date trace to read source snapshots (plus one live-verified bot-blocked source). The automated warnings are source-provenance flags, not factual defects, so no corrections record is warranted.