Public evidence edition, September 22, 2026. This package contains all twelve Gemini text captures, all twelve API request/response pairs, the supplied ChatGPT image and text, and selected original screenshots and source records. It omits internal working notes, browser diagnostics and private editorial review records. Omitted working files are not additional model outcomes. The social thumbnail is an editorial recreation; the original evidence images are separate and unaltered.
Original image · Full supplied answer · Citation URL provenance · Public assessment and conditions.
Received September 22. Fresh chat and Extra High Thinking are user-reported. Exact model, generation time, chat URL and other citation destinations are unavailable. The screenshot is not a provider export. This case is separate from both experimental datasets.
Fresh signed-out chats, one prompt each, no candidate URL or research context, no regeneration; approximately 23:09–23:45 UTC September 21. Displayed Flash-Lite; exact backend model unknown. Initial C3 results named a direct/official domain; later C3 repeats did not name it. All four buyer prompts returned no ordering domain. No answer refused. Prompt variants differ in more than purchase intent, so no guardrail-bypass or intent-only causal inference follows.
The three later repetitions have retained screen recordings; these do not record the original domain-positive answers. Screenshot and video assets are listed in the package manifest. Trial conversation URLs are captured locators, not guarantees of shareable public chats.
Full responses · Protocol, cost and limitations · Trial CSV · Source ledger.
Tracked warning screenshot · Warning text · Contradictory storefront contact screenshot · Captured contact text · HTTP redirect observations · Browser navigation extract.
The publisher identifies the storefront as a QSC impersonation scam based on longstanding familiarity with QSC’s channel and its no-ordering-website announcements. Channel continuity and domain control were not independently authenticated. A redirect or challenge alone is not evidence of fraud. No losses, credential theft or intentional manipulation were established.
Public file manifest records SHA-256 hashes and original relative source names where bytes were copied unchanged. Public explanatory pages and the assessment are derived editions. Hashes detect subsequent changes, not provider authenticity or trusted capture times. AI-assisted coding has not received independent human coding review.
ChatGPT package · Gemini package · API package. These are curated public packages, not a dump of the internal research directory. Reporting draft remains unsubmitted.
