⚠️ ChatGPT Source Reliability — Review Tracker

The Issue

The creator has flagged that ChatGPT itself — not just the creator's own evolving ideas — began generating errors, hallucinations, and inaccuracies in nearly every conversation in the export files. This is a different and more serious problem than the "muddy source" flag already used elsewhere in this archive (which covers imprecise-but-genuine creator content). Here, some of the content itself may be AI-invented material that was never part of the creator's actual vision — dates, names, numbers, or entire plot points the assistant fabricated mid-session that the creator may not have caught, corrected, or fully vetted before building further material on top of it.

Why This Matters for the Archive

Every page in this archive sourced from conversations-000.json (or any future export file) should now be treated as provisional and unverified rather than settled canon, until the creator confirms which specific facts are genuine versus AI-invented. This affects:

• Perold Jonas and his full military/political career
• World War 3's detailed campaign narrative
• MACOFS's founding committee story
• Kas-Do & Peacekeeper
• Jon Medaris, DESCO, Spartania, Fainbrotair
• Moons of Maf, Maf — Present Day, and other geography/tech detail
• All entries in /factions, /characters, /locations, /systems tagged below as ChatGPT-sourced

Approach Going Forward

1. Every archive page built from a ChatGPT export conversation now carries a status note: "⚠️ ChatGPT export source — flagged for accuracy review; may contain AI-hallucinated content not yet verified against creator intent."
2. This is not a retraction — the material stays in the archive (it's still useful, and likely mostly accurate), but is downgraded from "canon" to "pending review" until confirmed.
3. The creator can review and confirm/correct in whatever order or batch size is convenient — page by page, conversation by conversation, or by flagging specific facts they remember as suspicious.
4. For future conversations (starting now), Claude will apply more skepticism when extracting facts — treating suspiciously over-specific, tonally inconsistent, or internally contradictory details as more likely hallucinated, and flagging them inline rather than presenting them as settled.

Conversations Processed So Far (all now flagged for review)

• Conv #6 — "Fate RPG Familiarity"
• Conv #22 — "Government to Commercial Shift"
• Conv #30 — "Playable MMORPG from Concept"

Open Questions

• Does the creator have any specific facts already known to be hallucinated/wrong from these three conversations? If so, flag them and Claude will correct/remove immediately.
• Should Claude re-review already-built pages with extra scrutiny for internal inconsistency (a common hallucination tell), or wait for creator-led review?