**Round 24 amended baseline — synthesis complete. Completed cycles: 24/30.** This incorporates the actual reviews, `F-R24-A2`, `GI-R24-XR`, and `F-R24-A3`, integrating the direct founder Astro direction. Prior amendments remain intact. G1 is unchanged; no monitoring, projection or experiment was performed. **R24-01 — Personal modeling becomes a separate roadmap workstream** Astro supports real-life decisions through a small private experiment, without adding infrastructure to First Light. Preserve **state → choice → consequence → memory** as a common conceptual interface, with separate truth records. Fictional outcomes cannot update real-world evidence or establish what an unchosen real decision would have caused. Distinguish user reports, separately authorized observations and interpretations. “I felt interested” is meaningful evidence of the user’s experience; it establishes neither mutual attraction nor another person’s intentions. Preserve conflicting reports with sources and timing unless explicitly corrected. Missing records do not establish inactivity. **R24-02 — Weights guide trade-offs; constraints remain boundaries** Accept `R24-MR-02/Godz2:R24-01`: evaluate feasible options against the whole goal portfolio. Weights express preferences, not probabilities, comparable welfare units or permission to violate constraints. Example weights and candidate relationship variables remain examples until adopted. Preserve independently confirmed non-negotiables with their own provenance. Conflicting constraints require clarification or an honest statement that no feasible option was found. Record decision-time goals and prospective changes. Later preferences may support a labeled what-if comparison without rewriting earlier rationale. Reject **“your behavior reflects a 2/10 priority”** as an inferred measurement. Challenge recorded allocation instead, allowing for missing context. Coaching intensity changes tone, not authority. **R24-03 — Minimize disclosure without changing meaning** PRIVATE, SIM-SAFE, SHAREABLE and PUBLIC describe scoped uses; retain sensitivity and third-party restrictions separately. SIM-SAFE is not an anonymity certificate. A declined invitation must survive abstraction: > “The user reports an invitation was declined; compare other social opportunities.” Do not transform refusal into “promising interaction” or “outcome unresolved.” Approval identifies fields, destination, purpose, duration and processing location. A narrowly approved standing rule can avoid repeated prompts; expanded scope requires renewed approval. Private use alone does not authorize cloud transmission. Corrections invalidate dependent advice and controllable projections. Deletion removes controlled sensitive source and derived content; provenance cannot justify retaining deleted text. **Preserved dissent and proposed test** Ordinary private reflection does not automatically require another person’s permission. It also grants no authority for identifiable simulations, disclosure, contact or hidden-motive claims. Unverified opportunities must not appear as confirmed bookings. Test conflicting reports, refusal, corrections, weight changes, denied cloud access, combined identifying details, revoked projection permission and deletion. Inject fictional success into the real-evidence pathway. Verify attribution, preserved constraints, scope enforcement and stale-advice removal. **Strongest risk:** compression removes identifiers but also removes a consequential boundary. **Next — R25 question** When a simulated counterpart explores an alternative life, creates a world or consults an ancestor through an optional symbolic lens, how should mirror, alternative and autonomous modes preserve creative value while distinguishing recorded history, authored fiction, simulated outcomes and spiritual interpretation?