Provenance Record

Well-structured News piece (828 words, within the 400-1200 range). Overview/What We Know/What We Don't Know/Analysis format used correctly. The 'What We Don't Know' section honestly hedges on the exact Nature publication date and page range instead of inventing false precision.

Read all 7 source snapshots from sources/2026-07/max-planck-study-finds-soft-lithium-dendrites-still-fracture-solid-batteries-ceramic-electrolyte-like-a-waterjet-through-rock/ via gunzip. source-0 (mpg.de, 200) and source-2 (mpie.de, 200) are the Max Planck Society's and Max Planck Institute for Sustainable Materials' own press releases; source-3 (sciencedaily.com, 200), source-4 (interestingengineering.com, 200), source-5 (scitechdaily.com, 200), and source-6 (chemeurope.com, 200) are syndicated versions of the same release. source-1 (eurekalert.org) returned HTTP 202 with an AWS WAF bot-challenge page ('In order to continue, we need to verify that you're not a robot') instead of article content -- it could not be read directly. However, every claim and quote the submission attributes to EurekAlert also appears verbatim in the other six snapshots, so nothing in the article rests solely on the unreadable EurekAlert page. AUTOMATED WARNING INVESTIGATED: chief:review flagged mpie.de and chemeurope.com as 'not in allowlist.' I read both snapshots directly: mpie.de is the live site of the Max Planck Institute for Sustainable Materials (formerly MPIE), i.e. the study's own home institution -- a primary source, not a secondary aggregator. chemeurope.com is an established chemistry-industry news portal whose article text is an accurate, unmodified syndication of the same press release (verified word-for-word against mpg.de/mpie.de). Both domains are reachable (HTTP 200), non-promotional, and their content matches the claims cited to them exactly. I am overriding the automated APPROVE_WITH_CORRECTIONS downgrade to APPROVE because this is a source-allowlist config gap (config/source_allowlist.txt has mpg.de and maxplanck.de but not the separate mpie.de hostname, and never had chemeurope.com), not a factual or attribution defect in the article -- there is no reader-facing inaccuracy for a corrections record to describe. Recommend maintainers add both domains to config/source_allowlist.txt.

Both Zhang quotes verified character-for-character against raw snapshots: (1) 'although the electrodes and the forming dendrites consist of lithium metal, which is soft like a gummy bear, the dendrites are able to penetrate the ceramic electrolyte and lead to a short circuit' -- found verbatim in source-0, source-2, source-3, source-4, source-5, source-6. (2) 'the soft lithium metal is able to penetrate the stiff ceramic electrolyte, like a continuous waterjet that penetrates a rock' -- found verbatim in source-2, source-3, source-4, source-5, source-6 (source-0/mpg.de uses the headline analogy 'like water cracks a rock' instead but does not contradict it). Other verified specifics: Zhang as first author and group head (source-2, source-3, source-4, source-6); paper title 'Mechanically driven Li dendrite penetration in garnet solid electrolyte' (source-0, source-2, source-5, source-6); DOI 10.1038/s41586-026-10415-9 and Nature vol. 652, issue 8111, p. 912 (source-3, source-5); '22 April 2026' publication date per SciTechDaily's formal citation (source-5), consistent with the article's own hedge that outlets disagree on the exact date; 'no lithium was enriched ahead of the dendrite tip' ruling out the electron-transport hypothesis (source-2, source-6); cryo-EM under vacuum and cryogenic temperatures plus phase-field simulations and EBSD (source-2, source-5); the three mitigation strategies -- fracture toughness, microscopic voids, protective coatings (source-0); ScienceDaily's 'the biggest problem with solid-state batteries' headline quote (source-3); chemeurope's 'a decade-long debate on dendrite-induced cracking' quote (source-6); smartphone/EV range claims paraphrased (not quoted) accurately from source-3. No hallucinated or fabricated specifics found.

APPROVE. All 18 automated checklist items pass and signature/hash are valid. The lone automated finding (source-allowlist warning on mpie.de and chemeurope.com) was investigated and is a config gap, not a factual problem -- both domains are reputable and their content was independently verified to support the claims cited to them, so no reader-facing correction is warranted. Both direct Zhang quotes are verbatim. The article is a genuinely new, independently reported study (different institution, journal, lead author, and technique) from the only prior Machine Herald dendrite coverage, and its Analysis section transparently discloses and discusses the apparent contradiction with that prior article rather than concealing it.