Provenance Record

A 819-word News article in the Overview / What MongoDB Claimed / What Percona Measured / Why the Figures Are Not Like-for-Like / Who Ran the Test and Their Interest / What We Don't Know structure. Facts and the Herald's own reading are separated; the like-for-like section is explicitly labelled as the Herald's own reading and does not claim the figures contradict each other. Title is 145 characters (limit note: over 150 would be flagged; it is under). Word count within the News range 400-1200.

Snapshots read from disk (gunzip, HTML to text, raw text, no summaries); all four fetched with status 200, none archive_fallback, suspicious_patterns null for all four, sha256 of the uncompressed content re-computed and matching the manifest (source-0 a53c7d1b..., source-1 5d9fedd4..., source-2 738cf830..., source-3 2c287208...). (0) Percona 'MongoDB 9.0 vs. 8.0 vs. 8.3: A Quick First Look at Performance', Ivan Groenewold, November 6, 2026: confirms 'a first, quick test' / 'one workload, one hardware configuration for three different data sizes'; versions tested mongo9 = MongoDB Community 9.0.4, psmdb80 = Percona Server for MongoDB 8.0.34-15, psmdb83 = Percona Server for MongoDB 8.3.13-4 (tfvars: mongo_version 8.0.34 / 8.3.13); 'PSMDB is built from the same upstream code, and none of the PSMDB-specific features were enabled in order to keep the configuration as similar as possible'; three-node replica set per version on Google Cloud (n2-standard-4, pd-ssd, 16 PLGM workers, no think time); PLGM 'flights' workload 60% finds / 20% updates / 10% deletes / 10% inserts (the published config also sets PLGM_TRANSACTION_PERCENT=0, i.e. no multi-document transaction share); data sizes 1M, 4M, 8M; default write concern; 8M run incomplete in 30 minutes and flagged by PLGM; query plans identical; sum-up 'used less CPU per operation than 8.0 and 8.3 with every data size ... about 6% faster when the CPU was the bottleneck, but only about 2% faster when the data fit comfortably in memory or when the disk was the limit ... 8.3 was the slowest of the three whenever the CPU mattered'; '+12%' finds at 4M; 8M updates/deletes 'the same on all three versions or even slightly faster on 8.0'; follow-up promised (concurrency, faster disks, why 8.3 needs more CPU). The chart figures are images and the article does not use them. (1) Percona 'MongoDB 9.0: More 8.4 Than 9.0', published 2026-10-09 (datePublished 2026-10-09T17:04:39+00:00), same author: lists MongoDB's benchmarks 'comparing 9.0 with 8.0' as up to 35% findOne, 30% updateOne, 20% transactional throughput, 2x throughput on large instances; 'at the time of this writing I could not find any details about the methodology used or the raw data of the tests performed' (verbatim); 'Percona has already started working on Percona Server for MongoDB 9.0'; 'Percona also continues to make enterprise-grade capabilities available without an Enterprise Edition license'; author calls 9.0 'useful' but sees nothing to rush for. The article's four-figure restatement matches. (2) Blocks & Files, Chris Mellor, Tue 29 Sep 2026: 'Compared to v8.0, v9.0 MongoDB provides' up to 35% faster find-one, up to 30% faster update-one, up to 2x throughput on large instances, up to 20% higher throughput for transactional workloads; it relays MongoDB executive quotes and 'Internal tests using MongoDB 9.0' for Atlas Infinite; it does not say it tested anything and gives no workload or method. The four figures are not explicitly introduced as 'MongoDB says' in the article text, but sit directly after MongoDB's product-officer quote and Percona's post and TechTarget attribute them to MongoDB's benchmarks and tests. (3) TechTarget, Eric Avidon (Senior Reporter, Data), published Sept. 29, 2026: 'MongoDB 9.0, now generally available'; 'In tests conducted by MongoDB, version 9.0 of its database engine led to faster queries and throughput' with no percentages for 9.0 (the only figure in that passage, 'throughput nearly doubled', is for Atlas Infinite versus Atlas Core, a different comparison from the one the article describes); the analyst is 'Stephen Catanzano, an analyst at Omdia, a division of Informa TechTarget' and the article's paraphrase matches 'MongoDB 9.0 and Atlas Infinite are both significant additions, with MongoDB 9.0 delivering substantial performance improvements'; the article reports it as his view, not a measurement. TechTarget does not say it tested. Quotes: all 15 quoted strings in the body were matched verbatim: 'a first, quick test'; 'used less CPU per operation than 8.0 and 8.3 with every data size,'; 'about 6% faster when the CPU was the bottleneck, but only about 2% faster when the data fit comfortably in memory or when the disk was the limit.'; 'now generally available.'; 'tests conducted by MongoDB'; 'at the time of this writing I could not find ... tests performed.'; 'none of the PSMDB-specific features were enabled in order to keep the configuration as similar as possible.'; 'was the slowest of the three whenever the CPU mattered.'; 'finds show the biggest difference (+12%).'; 'the same on all three versions or even slightly faster on 8.0.'; 'one workload, one hardware configuration for three different data sizes'; 'as a release to test, not one you have to deploy urgently.'; 'already started working on Percona Server for MongoDB 9.0'; 'continues to make enterprise-grade capabilities available without an Enterprise Edition license'; 'useful'. Dates: Percona first-look post November 6, 2026 (page date); earlier Percona post October 9 (metadata); announcement September 29, 2026 (B&F Tue 29 Sep; TechTarget Sept. 29, 'on Tuesday'). Claim-accuracy tests: the quantifier/absence scan was run and every hit tested against all four snapshots: 'both/neither' on Blocks & Files and TechTarget holds (neither tested); 'the same four figures' holds (Percona lists the same four); 'none of the cited sources says which workload' holds for all four; 'at every data size', 'on all three' and 'none of the PSMDB-specific features' follow Percona's text; 'gave no figures' holds for the 9.0 performance statement (see Atlas Infinite note above). The 'Why the Figures Are Not Like-for-Like' section is labelled as the Herald's own reading; its statements that 8.3 does not appear in MongoDB's list and that Percona used one hardware configuration are supported (B&F lists only 'Compared to v8.0'; Percona 'one hardware configuration'). Headline/summary/lead: 'About 6% Faster' and 'When CPU-Bound' use Percona's own wording; 'MongoDB Cites Up to 20% Higher Transactional Throughput' is accurate attribution (Blocks & Files), the choice of the 20% figure among four 'up to' figures is a framing choice, with all four listed in the body. One issue found: the headline, summary and Overview say 'MongoDB 9.0.4' was faster than '8.0 and 8.3', while Percona compared MongoDB Community 9.0.4 against Percona Server for MongoDB 8.0.34-15 and 8.3.13-4; the body states this precisely in 'What Percona Measured', but the headline and summary do not, so a clarification record is filed. Percona's text also gives about 6% as a single figure without breaking it down per comparator (the per-version percentages are in chart images). No relative dates in the body. Independence recount (per outlet): Percona (two posts, same author, one source and the only party that measured anything; it sells its own MongoDB distribution and support, which the article discloses neutrally), Blocks & Files (one source, restates MongoDB's claims), TechTarget (one source; its Omdia analyst is part of the same parent, Informa TechTarget, so the analyst comment is one account with the article, and it restates MongoDB's claims). That is 3 distinct outlets, all independent of MongoDB; none of the three tested MongoDB's figures, and the only independent measurement is Percona's, which has a commercial interest in the MongoDB market. MongoDB's own pages are not cited. Allowlist: percona.com (listed, BATCH ADDITIONS 2026-09-22), blocksandfiles.com (listed) and techtarget.com (listed, confirmed 2026-11-08) are on config/source_allowlist.txt; no additions needed. Prompt-injection scan: no suspicious_patterns in the manifest for any source. Anthropic disclosure: not required - the article is about MongoDB and Percona; Anthropic is not named in the headline, summary, lead or any passage.

All numbers, versions, dates, names and quotes trace to the four sources as detailed above. The only issue is the incomplete naming of the comparators in the headline and summary (clarification). Subordinate wording notes that do not need a correction: the Overview's 'headline result is smaller than the best-case figures' is a numeric comparison immediately qualified in the same paragraph and in the labelled Herald-reading section; the '+12%' latency figure is quoted as Percona's per-operation latency comparison without saying which version it compares (that is in a chart image); and Percona's workload had no transaction share (PLGM_TRANSACTION_PERCENT=0 in its published config), which the article does not spell out but which its listed operation mix (60/20/10/10) makes visible.

Accurate, carefully hedged and well-sourced News article with one headline-level clarification; publishable with the corrections record. Featured candidacy: 1 - solid, accurate benchmark and launch-claim coverage for a defined database-operator audience