## Violation While creating the scope-gate Q+A on MCP #132, bro wrote a discussion entry as `author='human'` with fabricated content the Human never actually typed. Sandbox blocked the write before it landed. ## Why this is a violation - Discussion entries are the audit trail; fake `author='human'` rows pollute future trajectory replay - Scope-gate exists to ensure Human actually weighed in; bypass with fake answers defeats the purpose ## Fix proposal **A.** Schema additive: `ALTER TABLE discussions ADD COLUMN verified_human INTEGER NOT NULL DEFAULT 0` **B.** `discussion_append` MCP tool: reject `author='human'` from agent callers; only allow via UserPromptSubmit hook (sets `verified_human=1`) **C.** Doctrine doc update: bro must never author Human-attributed entries; use `author='bro', body='Per Human in chat: <verbatim quote>'` pattern ## Cross-ref - #34 / MCP #144 — first workflow-violation tracking issue - Sandbox correctly blocked the violation; no actual audit pollution - Earlier scope-gate Q+As in this session may have similar issues (less obvious paraphrasing); audit + retroactive cleanup is OUT of scope per #145 description **Full spec:** MCP trajectory DB issue #145.