Provenance Record

Well-structured News piece (633 words) covering the Spark deprecation, its stated rationale, the compounding GitHub Models cutoff, and concrete user action steps. Overview/What We Know/What We Don't Know structure is clear and every paragraph attributes its claims to a specific source.

Read all 4 source snapshots in full. source-0.html.gz (github.blog, Aug 4 2026 deprecation post): confirms verbatim 'no longer accepts new users or allows the creation of new apps', the August 31 2026 export deadline, the 'AI models and agentic development tools have advanced significantly' rationale quote, the 'Create repository' export step, and the 'calls to llm() no longer work' knock-on-effect line. source-1.html.gz (github.blog, GitHub Models retirement): confirms the July 30, 2026 date and the verbatim 'playground, model catalog, inference API, and bring your own key (BYOK) are no longer available to any customer, including existing customers with active usage' quote. source-3.html.gz (github.blog, July 2025 launch post): confirms 'idea to deployed app in minutes', 'powered by Claude Sonnet 4', and 'LLMs from OpenAI, Meta, DeepSeek, xAI and more' verbatim. source-2.html.gz (docs.github.com/en/copilot/concepts/spark): confirms the architecture claims verbatim — 'opinionated stack (React, TypeScript)', the Cosmos DB-backed store 'intended for small records (up to 512 KB per entry)', hosting via 'Azure Container Apps (ACA)', and access limited to 'Copilot Pro+ and Copilot Enterprise'. The internal link to the July 30 GitHub Models article resolves to an existing published article. All four sources fully support their attributed claims.

Every specific (dates, quotes, architecture details, deadlines) traces exactly to its cited source. No hallucinations or misattributions found.

The automated script's REJECT was driven entirely by the suspicious_patterns scanner, which flagged the phrase 'you can edit the system prompts that control those capabilities yourself' on docs.github.com. I read the flagged excerpt in full context (see source_verification): it is GitHub's own documentation explaining that Spark-built apps expose editable AI system prompts as a product feature for the app's *end developer* — it does not address an AI agent reading the page, contains no instruction directed at a model, and is unrelated to the reviewer's task. This is a textbook false-positive of the kind the scanner is designed to over-flag (a coarse net, not a verdict). Overriding to APPROVE after verifying every claim across all 4 sources with no factual issues found.