Provenance Record

News, 939 words (range 400-1200), five sources (News guideline 3-5). Clear structure: lead, announcement, what the performance figures rest on, analyst commentary, the Streamhouse label, a closing 'What Is Not Settled' list. Neutral tone; vendor figures are labelled as the vendor's and the source disagreement over data freshness is reported without being reconciled. Title is 147 characters: long, but under the 150-character limit; noted for the record, no action. No AI self-reference, no promotional language.

Integrity: submission_version 3, contributor_model 'Claude Sonnet 5.5', payload hash and Ed25519 signature valid (chief:review checklist all true). Submission extracted with git show 8c830558701e1edbb4fc948aa044dbe7d144907d:<path> from the PR head and the filename matches the PR's only file. All five snapshots were fetched live by chief:review (HTTP 200, no archive fallback, no error, suspicious_patterns null for all five, so no scanner finding to override) and were read in full as raw decompressed text from sources/2026-11/<slug>/ (sha256 of uncompressed content, first 16 hex: source-0 a01f3a52d3003981; source-1 c35770f93e303381; source-2 26086430d46f7b36; source-3 106866b08dde490f; source-4 d3087aae4b924dad; full hashes in manifest.json). Additionally the IBM release was re-fetched with curl and a browser User-Agent (HTTP 200) and the figures sentence is identical to the snapshot. No source needed an alternative route; none was a bot-check page or JS shell. source-0 (IBM Newsroom release, Nov 4, 2026, San Francisco dateline): CONFIRMED. 'Confluent, an IBM (NYSE: IBM) company, today announced ...'; 'With Lightning Tables now generally available on AWS, Confluent completes its five-pillar Data Streaming Platform ... Stream, Connect, Process, Govern and now Query'; 'Generally available on AWS, Lightning Tables on Confluent Cloud is a read-only serving layer that durably stores and materializes Apache Kafka data into tables, keeping them current as changes arise'; SQL and direct API; 'without separate ETL pipelines, databases, or caches'; Query lists Tableflow ('makes streaming data available to lakehouses and warehouses as open table formats'), Real-Time Context Engine (MCP) and Lightning Tables; Streamhouse 'a new type of data architecture'; DSP 'provides a complete, unified foundation for building a Streamhouse'. The figures, verbatim: 'In Confluent internal testing, approximately 100 milliseconds from new data creation to serving, with query latency of 10-100 milliseconds and throughput of 1,000+ queries per second.' So the headline's 'Latency Figures From Its Own Internal Testing' is stated by the IBM release (the Confluent blog does not say it; the article says exactly that). The release does not say how the tests were run. The Stream pillar figure in the same release ('sub-millisecond latency in Confluent internal testing; actual latency varies by configuration and workload') is also labelled internal testing, so 'the performance figures in the announcement' is accurate. source-1 (Confluent blog 'What's New in Q4'26 Confluent Data Streaming Platform', Nov 4, 2026): CONFIRMED. 'Query joins Stream, Connect, Process, and Govern as the fifth DSP pillar'; 'Six innovations lead this release': Lightning Tables, Managed Model Inference, Schema Registry Updates, Multi-tenancy for Confluent Private Cloud, Confluent Gateway, Tableflow updates (the article's list of six matches in order); Lightning Tables 'a fully managed, read-only serving layer' (the 'read-only ... serving layer' wording is Confluent's) and 'a managed, read-only serving layer that durably materializes Kafka data into queryable tables ... and is generally available on AWS'; 'Data freshness at scale: Changes are available for serving within hundreds of milliseconds; queries return in 10-100 ms at 1,000+ queries per second'; 'Lightning Tables supports BYOK, backward-compatible schema evolution, usage and performance metrics'; Real-Time Context Engine 'serves fresh context to AI agents through MCP as an access pattern for Lightning Tables'; Query section names Snapshot Queries and Streaming Queries. The words 'internal' and 'test' occur nowhere in the page text, so the article's statement that the blog does not use the phrase internal testing is true. The blog quotes 'within hundreds of milliseconds' (article quote exact). No price, maximum table size or test configuration for Lightning Tables appears (the Streamhouse Accelerator discount programme names Lightning Tables among products but gives no Lightning Tables price). No other cloud and no date for other clouds is given for Lightning Tables. source-2 (TechTarget, Eric Avidon, Senior Reporter, Data, 'Confluent advances evolution toward becoming AI streamhouse', published Nov. 4, 2026): CONFIRMED. 'Lightning Tables, which is now generally available on AWS'; Bond called the new features 'moderately' significant additions and 'Lightning Tables is the most valuable for Confluent customers' (reporter's clause, which the article correctly attributes to TechTarget rather than quoting Bond); 'after the longtime tech giant acquired Confluent in December 2025'; quotes 'Lightning Tables stand out because they change the read path' and 'But speed and ease of use alone will not advance the real-time imperative' (article drops the leading 'But' outside the quote marks, fragment verbatim); IDC survey 'earlier this year' showing 'data trust and quality rather than latency are the top technical barrier to real-time data initiatives'; gaps list (open semantic standards and ontologies, agent governance including lineage, more widespread hybrid connections); 'the vendor's streaming data capabilities are the most complete across brokering, stream processing, data serving and context and governance with Microsoft perhaps its closest rival'; 'Microsoft is closest on event-to-semantics-to-action with Fabric Real-Time Intelligence and Fabric IQ'; 'What [Confluent] lacks in coverage, it makes up for in scale ... via the IBM acquisition'; Falconer (Confluent's vice president of product for AI products and strategy): 'We'll keep pushing on how much data Lightning Tables can handle and on overall query throughput'; 'as Confluent builds up its streamhouse, it will focus on expanding capabilities to all public clouds' (the article quotes this as it stands and adds that no cited page dates other clouds; the sentence does not name Lightning Tables specifically); 'Bond, meanwhile, suggested that Confluent provide evidence of the efficacy of Lighting Tables [sic] and other new features'; 'It would be nice to see Confluent publish benchmarks and reference architectures for Lightning Tables,' Bond said. The article does not carry the Lightning Tables 100 ms figures, so it contains no third-party measurement. Attribution: all Bond statements are from this article (Nov 4); Farmer and Petrie are from source-3 (Sept 22), correctly separated. source-3 (TechTarget, Eric Avidon, 'Streamhouse: A new way to serve AI with real-time context', published Sept. 22, 2026): CONFIRMED. Farmer, 'founder and principal of TreeHive Strategy': 'The streamhouse definition provides a vendor-neutral blueprint.' Petrie, 'an analyst at BARC U.S.': 'the Streamhouse website documentation is pretty sparse so far and would benefit from a more detailed reference architecture based on customer implementations' (article's quoted fragments exact). Bond appears only in source-2, Farmer and Petrie only in source-3. source-4 (Confluent blog, Streamhouse Working Group, Sep 15, 2026, joint dateline Helsinki/Mountain View/San Francisco/Munich): CONFIRMED. Aiven, Confluent, Redpanda, StreamNative and Ververica announced the Working Group and an open vendor-neutral definition; three attributes 'Real-time', 'Production-native', 'Decentralized'; 'The Streamhouse name was coined by Ververica'; 'Organizations can build a Streamhouse with open-source technologies, commercial products, or both'; 'no single project, component, or vendor portfolio constitutes a Streamhouse on its own'. 'The term predates this launch' is therefore correct (Sept 15 versus Nov 4). Claim-accuracy tests (policy 'Claim accuracy'): the quantifier/absence grep returned lines 2, 5, 17, 19, 31, 37, 39 of title+summary+body. 'Neither Confluent page explains the difference' (IBM release vs blog): true; the IBM-hosted release is loosely called a 'Confluent page' (it is Confluent's announcement on IBM's newsroom), a wording nit not worth a record. 'none gives a Lightning Tables price, a maximum table size, or the data set, cluster configuration and workload behind the figures': checked against all five snapshots, true. 'none of the cited pages gives dates for Lightning Tables on other clouds': true. 'None of the cited pages reports a Lightning Tables benchmark by a party other than Confluent': true; TechTarget (the only independent outlet) reports no measurement. 'all public clouds' is TechTarget's wording about Confluent's streamhouse. No 'both/every/each/all' claim about items fails item by item. Qualifiers: the headline/summary keep 'generally available on AWS' (AWS-only scope), 'Its release attributes' (attributed to the vendor's release) and 'internal testing'. Dates: November 4, 2026 on the IBM release, the Confluent blog and TechTarget ('Wednesday' = Nov 4, 2026); Sept 15 and Sept 22 correct; no time-zone issue. One relative date in the Herald's own voice: 'from earlier this year' (inherited from TechTarget; see correction 3). Derived figures: none computed by the Herald. Disagreement: the freshness discrepancy ('approximately 100 milliseconds' in the IBM release versus 'within hundreds of milliseconds' in the Confluent blog) is reported with attribution and not reconciled, which is the policy's required handling (neither source explains it). Acquisition dates: the Herald's earlier article (2026-03/30-ibm-closes-11-billion-confluent-acquisition-...) states completion on March 17, 2026 and the internal link resolves to that article. I re-fetched its cited sources: PR Newswire (March 17, 2026) 'IBM (NYSE: IBM) today completed its acquisition of Confluent, Inc.', and TechCrunch (December 8, 2025) 'IBM to acquire Confluent for $11B'. So IBM announced the agreement on December 8, 2025 and closed on March 17, 2026; TechTarget's 'acquired ... in December 2025' most plausibly refers to the December announcement, but TechTarget does not say. The new article reports the disagreement honestly ('does not say which event it means ... does not reconcile') but a source the Herald already holds explains it; clarification filed (correction 2), not a defect in the headline. Title fairness: 'Generally Available on AWS' is stated by the IBM release, the Confluent blog and TechTarget (AWS only, no other cloud dated). 'Asks for Benchmarks': TechTarget's own verb is 'suggested', Bond's words are 'It would be nice to see'. 'Asks' is a mild compression of a polite wish; the lead and summary quote his words faithfully and the body states them in full, so readers are not misled; clarification filed (correction 1) rather than a REJECT, since a headline verb that a reasonable reader takes as a request is not a wrong statement. Independent-source recount: five URLs, four outlets/parties. Independent of the subject: ONE outlet, Informa TechTarget (two articles by the same staff reporter count as one source). It adds reporting beyond the press release: named-analyst commentary (Bond at IDC, Farmer at TreeHive Strategy, Petrie at BARC U.S.), a Falconer interview and a critique of gaps, though the product facts and the Lightning Tables availability it reports come from Confluent. First-party: IBM Newsroom (Confluent's parent), two confluent.io blog posts (Confluent is IBM's subsidiary; the Working Group post is a joint announcement by five vendors, hosted on Confluent's blog, and the Working Group's members include Confluent, so it is not independent either). Not an Analysis, so the 2-independent minimum does not apply; the News guideline (3-5 sources) is met. Allowlist: newsroom.ibm.com (line 1224) and techtarget.com (confirmed 2026-11-08, re-checked here: named staff byline Eric Avidon, Senior Reporter, Data; footer 'Editorial Team' link; no sponsored label) were already allowlisted. confluent.io was not: the file contained only the path form 'confluent.io/blog', which the exact-host matcher does not apply, hence chief:review's warning. Confirmation: IBM's own newsroom release ends 'To learn more, visit www.confluent.io' and links the two cited confluent.io posts; the GitHub organisation github.com/confluentinc ('Confluent Inc.', created 2014-10-28) lists https://confluent.io as its website. Host 'confluent.io' added with a justification comment in this commit; no version bump. Anthropic disclosure: not required - the article does not name Anthropic or Claude anywhere (confirmed by search of title, summary and body); the cited Confluent blog mentions Claude only in a list of AI assistants, which the article does not repeat. Rule 15 is not engaged. Originality: no published or pending article covers Lightning Tables or Confluent's Streamhouse launch; the only related Herald piece is the March 30, 2026 IBM-Confluent acquisition article (linked from the new article and recorded in related_articles).

Every number, date, name, title, firm and quotation traces to a cited source as detailed in source_verification: GA on AWS on November 4, 2026; five-pillar platform; approximately 100 ms / 10-100 ms / 1,000+ queries per second labelled as Confluent internal testing (IBM release); 'within hundreds of milliseconds' (Confluent blog); the Bond, Farmer, Petrie and Falconer statements attributed to the right TechTarget article and date; Working Group facts of Sept 15, 2026. No fabricated quote or specific found. Three subordinate points need clarifications (headline verb 'Asks', the unexplained acquisition-date difference, one inherited relative date), none changes a fact.

Accurate, well sourced and fairly framed vendor-announcement coverage: the headline's 'internal testing' claim is backed by the IBM release (not by the Confluent blog, as the article itself says), the AWS-only scope is stated by all three relevant sources, all quotes are verbatim and correctly attributed, and the data-freshness discrepancy is disclosed. Verdict APPROVE_WITH_CORRECTIONS with three clarification records (headline verb, acquisition dates, relative date); confluent.io added to the allowlist after confirmation. Featured candidacy: 1 - accurate, carefully hedged release coverage with independent analyst reaction for the Kafka and data-streaming developer audience