Provenance Record

Clean, well-structured News piece (553 words, within the 400-1200 News range). Clear Overview / What We Know / What We Don't Know / Analysis flow. Every substantive claim carries an inline attribution to a specific outlet, and the 'What We Don't Know' section is appropriately candid about the undisclosed partner count and approval criteria.

All three cited sources fetched at 200 and read from disk as gzipped snapshots. source-0.html.gz (TechCrunch, 'OpenAI limits GPT-5.6 rollout after government request'): CONFIRMED — verbatim quote 'We don't believe this kind of government access process should become the long-term default'; 'small group of trusted partners'; partners 'whose participation has been shared with the government'; Sol/Terra/Luna lineup with Sol as flagship; pricing Sol $5/$30, Terra 'half that' ($2.50/$15), Luna $1/$6; Sol slightly better at coding than Anthropic's Claude Mythos 5 and using a third of the output tokens; EO review up to 30 days before release; Trump administration ordered Anthropic to remove foreign-national access to Fable 5. source-1.html.gz (Engadget, 'OpenAI Launches A Limited Preview Of GPT-5.6'): CONFIRMED — Sol 'strongest model yet', Terra everyday use with similar performance to GPT-5.5 despite being twice as cheap, Luna 'the company's lowest cost model', broad release 'in the coming weeks'. source-2.html.gz (Engadget, 'OpenAI Will Initially Only Release ChatGPT 5.6 To Government-Approved Customers'): CONFIRMED — federal leaders 'approving access customer by customer during this preview period' with a broader release a 'couple of weeks later', citing a Sam Altman staff memo; Altman's paraphrased 'not our preferred long term model / more sustainable approach' statement; Office of the National Cyber Director, Office of Science and Technology Policy, and Commerce Secretary Howard Lutnick involvement; Trump executive order asking for voluntary federal review before public release.

All model names (GPT-5.6 Sol/Terra/Luna), the 'limited to trusted partners at the US government's request' framing, pricing tiers, quotes, and the agency/EO details trace to the two live sources (TechCrunch + Engadget). No load-bearing claim rests on a bot-blocked outlet. The Journalist's deliberately dropped blocked-source specifics were verified absent from the body: no '~20 organizations' figure, no executive-order number or date, and no August-process detail appear. The one minor imprecision: TechCrunch attributes the 'uses a third of the output tokens' efficiency to Sol being competitive with 'Mythos preview,' and the article attaches it to the 'slightly better at coding than Claude Mythos 5' sentence — both underlying facts are true and sourced to TechCrunch; the compression is a defensible paraphrase, not a misstatement, so no correction is warranted.

APPROVE. A clean, neutral, fully-sourced News piece. All model names, the government-request framing, pricing, quotes, agency involvement, and the EO context verify verbatim against the TechCrunch and Engadget snapshots read from disk. No load-bearing claim depends on a bot-blocked source, and every specific the Journalist deliberately dropped is confirmed absent from the body. No corrections required.