# AIRichardMoon Corpus Privacy Policy ## Changelog - **2026-07-25** โ€” Initial publication. Full 10-section policy adopted per ticket #003. Contact address set to `kenneth@quantumaikido.com`. 14-day removal SLA confirmed. DVC history purging kept as "squash or purge when feasible." ยง10 participant notification softened to best-effort. --- ## 1. What the corpus is The AIRichardMoon corpus is a collection of text documents โ€” books, articles, coaching transcripts, recorded dialogues, and text-chat exports โ€” that grounds an AI coaching chatbot. The chatbot is available to registered members of the Quantum Aikido Sangha and is not published openly. The corpus is stored as Markdown files under version control (DVC) and indexed into a SQLite FTS5 database that the chatbot searches at query time. ## 2. Whose words are in the corpus The corpus contains the words of **Richard Moon** (the teacher whose work the system is built around) and, in dialogue-format documents, the words of other participants who conversed with him in group chats, coaching threads, seminars, and Q&A sessions. ## 3. Consent and the excluded-writers list - **No person's words are added to the corpus without their consent.** Consent may be withdrawn at any time. - A writer who has not consented (or has withdrawn consent) is placed on the **excluded-writers list** maintained by the corpus operator. Their messages are redacted from any document before it enters the corpus, and removed from existing corpus documents when the list is updated. - Redaction is **in place by default**: the message slot is preserved as `[REDACTED WRITER]: [REDACTED]` so the surrounding dialogue stays intelligible, but neither the writer's name nor their words are retrievable from the corpus file. Full deletion is available as an option when structure does not need to be preserved. - The excluded-writers list itself is **not stored in the corpus** and is not version-controlled. It lives only in the operator's private records to avoid creating a public roster of people who asked to be removed. ## 4. Attribution - During the current build phase, Richard Moon is attributed by name in corpus documents (e.g. sender label `Moonsensei`, byline `RM`). Other consenting participants may be attributed by name with their explicit permission. - **Before the chatbot goes live to members, attribution to Richard Moon will be removed** from corpus documents, and the chatbot will be rebranded under a name that does not identify him. At that point the corpus becomes an anonymous body of teachings, and the redaction policy applies to Richard's name the same way it applies to any other writer: his words stay, his name does not. - No participant is ever attributed by name in the live chatbot's responses. The chatbot answers as itself; it does not quote "Richard said..." or "Scott said...". ## 5. Personal information beyond message content The following are **always redacted** from corpus documents, regardless of consent status, before ingestion: - Phone numbers, email addresses, physical addresses - Account handles, usernames, and profile URLs tied to a real person - Dates of birth, medical details, financial details - Names of third parties (non-participants) mentioned in passing - Metadata that could re-identify a redacted writer (e.g. a unique phone number in a sender label, even if the name is redacted) Redactions are marked `[REDACTED]` and noted in an HTML comment in the source file for auditability without exposing the original value. ## 6. Access - The corpus source files and FTS5 index are stored in a private DVC remote and on Google Cloud Run infrastructure accessible only to the system operator. - The chatbot is reachable only by registered members who have logged in with an invitation code. There is no public endpoint that returns corpus text. - Corpus documents are never exported, shared, or published outside the system without a separate per-document review. ## 7. Retention and removal - A writer may request removal of their contributions at any time by contacting the operator at **kenneth@quantumaikido.com**. Removal means: re-preparing every affected document with the writer added to the excluded-writers list, rebuilding the FTS5 index, and pushing the updated corpus. The operator aims to complete removal within 14 days. - Because the corpus is version-controlled, prior DVC revisions technically retain old content. The operator will squash or purge affected history when feasible; until then, old revisions remain in the private DVC remote and are not exposed through the chatbot, which always reads the latest index. - The original raw exports (before preparation) are kept only in the operator's private working directory, not in the corpus or the DVC remote, and are deleted once the prepared version is confirmed correct. ## 8. What the chatbot does with corpus text At query time, the chatbot retrieves relevant passages from the FTS5 index and passes them to a Gemini LLM as grounding context. The LLM generates a response; it does not echo corpus passages verbatim unless a passage is directly relevant, and it never attributes a passage to a named person. The LLM provider (Google) receives the retrieved text as part of the prompt; this is disclosed here so members know their queries and the grounding context transit Google's API. ## 9. Member queries - Member chat queries are stored in Firestore for conversation continuity and for the staging/production replay comparison used in deployment validation. - Queries are not shared with third parties other than the LLM provider (Google Gemini) needed to generate a response. - A member may request deletion of their conversation history at any time by contacting **kenneth@quantumaikido.com**. ## 10. Changes to this policy Any change to this policy is logged with a date and a summary in a changelog at the top of this document. Material changes that affect how a participant's words are handled are communicated to known participants by email on a best-effort basis before taking effect, where feasible.