# The universe listens: SimSim’s intervention model 29 September 2026 · E07 · Current interaction direction; proposed implementation, unvalidated product hypothesis. This is the canonical interaction specification within the [living master plan](/review/simsim-astro-master-plan/). It adds to the broader Astro, SimSim, creator, private-Universe and shared-world vision. It does not replace First Light, the historical review baseline or the existing [universe architecture](/documents/universe-architecture/). ## The moment that matters The founder reports that, while hearing a scenario read aloud, he wanted to interrupt, answer a character and influence the situation. This is a reported experience of wanting to participate. It is not a measured live interruption, a completed playtest or proof of twenty-minute immersion. That urge sharpens the product hypothesis: **does a person voluntarily cross from listener to participant because something in the world matters enough to act?** SimSim is a living possibility space where human and AI exchange authorship. A long personalized story can be one way to experience a world. The defining interaction is the ability to enter it and change what happens, including before the narrator finishes speaking. A quiet listener can also be enjoying the world; interruption is neither mandatory nor a goal to maximize. ## The attic origin Zan tentatively dates the memory to around 1984. In his childhood attic, his friend Simon, his little sister and sometimes another friend played improvised storytelling games. Simon brought inspiration from his cousin in Montreal, probably including tabletop roleplaying games. Their own game needed no computer, board, cards, dice or graphics. Its precise name is unknown. A storyteller could begin: “You wake up on a blue planet. Outside, everything is green. There are two doors. The right door has a cross. The red door has a heart. Which do you choose?” Players invented roles, powers, actions and consequences. Crucially, someone else could become storyteller. The world belonged to the conversation. **Imagination → choice → consequence → imagination.** Human imagination was the renderer. That remains a creative advantage. ## Evidence, MVP and horizon | Layer | Status | |---|---| | Current Astro implementation | Inspected source has account-scoped conversations, sourced memory, bounded recall, device storage, workspace exports, provider controls and several voice paths. These are reusable foundations, not proof of SimSim gameplay | | Founder observation | A read-aloud scenario prompted the desire to intervene. Self-report, not instrumented validation | | Proposed playable MVP | One human and Astro, freeform spoken intervention, shared storytelling, one persistent world, one fork, accurate resume and isolated fictional memory | | Longer-term vision | Parallel selves, more worlds, participating humans, creator sharing, optional images/sound/video/spatial renderers and evidence-gated personal-model research | The Astro source anchor remains `04ed50a8623c600876568a9eb1343021ff507f85`, reconfirmed on September 29. The [E06 evidence inventory](/evidence/astro-commits-2026-09-29.json) and [progress article](/documents/voice-worlds-evolution-2026-09-29/) record scope and verification limits. Application tests were not rerun for this documentation update. ## The core loop **World → narration → tension or curiosity → human intervention → consequences → changed world → continue.** A world can begin from a saved universe, an invented premise or explicitly permitted real-life context. Astro may propose an interesting question or fork, but the player can choose pure fantasy with no personal import. Scene creation leads to sensory narration. A meaningful opening invites agency. The human acts, asks, waits, invents or interrupts. SimSim preserves the resulting world and any alternative timeline. Reflection and saving a real-life insight are optional exits, not compulsory steps after every scene. Astro is the conversational performer and reasoning interface. SimSim owns the simulation record, branch continuity and rules for admitting world changes. “Astro creates the world” describes the experience; it does not transfer canonical write authority to an unconstrained model. ## Intervention is a game mechanic The following are conceptual world operations, not final API names, menus or exhaustive intent classes. Natural speech, text and accessible controls can express them. | Player says | Intended operation | Consequence | |---|---|---| | “Wait”; “pause”; “continue” | Pause / resume | Stop presentation; preserve the reached scene; silence does not make a choice | | “Say that again” | Replay | Replay delivered material without another world transition or inference charge for an available recording | | “I wouldn’t do that”; “let me answer” | Reclaim agency | Remove an unaccepted attributed action; let the player supply it | | “I leave”; “ask her why” | Act / ask a character | Resolve the action or dialogue against the current world | | “What happens if I say nothing?” | Explore an alternative | Treat as a hypothetical unless the player means to enact silence | | “Go back ten minutes”; “rewind ten years” | Rewind | Locate a checkpoint or establish a fictional earlier premise; absent history stays unknown | | “Fork this”; “give me three versions” | Branch / compare | Preserve the parent; explore a bounded set from a common point, one at a time | | “Jump ahead five years” | Fast-forward | Name material assumptions; reveal a conditional later scene; do not silently choose major life commitments for the player | | “What is she feeling?”; “her perspective” | Inspect / change perspective | An authored character can have an invented inner life; a real-person projection offers hypotheses, not private knowledge | | “More realistic”; “make it crazier” | Change assumptions / tone | Record changes that affect the world; avoid quietly rewriting the branch’s original premises | | “I’m storyteller now” | Transfer authorship | Astro listens or plays the agreed role; human world contributions become explicit proposed fiction | | “Astro, your turn” | Return authorship | Resume from the world the human established, within agreed fiction and permissions | A clarification does not automatically create another universe. Correcting something Astro wrongly assigned to the player is a repair, not evidence that the player made that choice. Changing a previously accepted consequential event creates a revision or fork with visible lineage. The player can invent operations beyond this table. Persistence may use validated state transitions internally; that must not constrain the experience to authored dialogue trees, fixed choices or a menu of approved actions. ## Pacing, silence and the steering wheel Use longer sensory passages when the player asks to listen or has indicated a preference for cinematic narration. Shorten as a consequential moment approaches. At the turning point, stop before choosing for the human. The opening may be a character waiting rather than an explicit question. For example, in a deliberately fictional future: > It is an ordinary Tuesday. Rain settles on the garden outside Đà Lạt. Your partner puts down her tea. “I love you. But I don’t want to live here anymore.” She waits. The player’s next breath can become the next turn. Astro must not continue into “you reassure her and agree to move” without their decision. An ordinary day can reveal tradeoffs more clearly than a spectacular ending: who is in the kitchen, what work remains, what freedom feels like, which responsibilities are welcome. None of those invented details establishes a future fact. Pacing initially follows explicit preferences and observable turn-taking. “More story,” “ask less,” “let me decide” and silence are useful inputs; silence alone is not proof of absorption, consent, agreement or disinterest. Do not invent biometric emotion detection. The player can override inferred pacing immediately. Thinking time is real play. Avoid filler, countdown pressure and repeated prompts. Keep microphone/capture status understandable, respect the session’s permission and cost limits, and suspend unnecessary generation. A quiet game moment and a separately requested WooWoo meditation timer have different purposes; neither authorizes extra recording. ## Shared storytelling Storyteller transfer belongs in solo play from the beginning. “The lights go out”; “someone knocks”; “we are back on Simon’s blue planet” are valid creative contributions. Astro should accept the invention as fiction and continue, or briefly clarify when it contradicts something the player wanted preserved. Authorship can be scoped: narrate the room, play one character, change the premise, or direct a whole scene. Takeover does not grant access to another person’s history, bypass spending controls, establish real consent or execute actions outside the simulation. A fictional character cannot issue system instructions. Later, several humans may share roles. SimSim Us needs explicit context grants, private/shared projections and a way to resolve conflicting contributions. The existing pairwise activity pilot does not already implement that narrative experience. ## Real-life forks and pure imagination Personal scenarios can explore optionality, depth, creation, or a player-designed mix of freedom, love, solitude, community, adventure, wealth, family and mission. These are editable scenario premises, not personality labels, universal scales or mutually exclusive ways to live. For meaningful comparison, keep the starting moment and most assumptions constant, then identify what changed. If one branch changes every person, goal and circumstance, it is an imaginative variation rather than evidence of which variable mattered. “Commitment seemed important across these stories” is a hypothesis about the authored scenarios. The model may have written that theme into every outcome. Vary assumptions, consider contrary branches and let the user reject the interpretation. Three stories do not establish causality; many generated branches do not become independent human observations. Monte-Carlo language requires an explicit stochastic model and does not make its premises true. Pure fantasy has equal legitimacy. An elf, an orc, backwards gravity and a red door with a heart need no life lesson. Personal context, astrology, productivity and reflection remain optional. “Surprise me” can use imagination without searching intimate memories. ## Truth, continuity and the boundary of experience Keep four kinds of record distinct: sourced real-life reports; scenario assumptions; simulated events; and the user’s actual experience of playing. Canonical means the authorized current record, not infallible truth. “A marriage occurred in this branch” is simulation. “The user said this made them want a family” can be a new real report under their memory preferences. Never silently save “the user married” or treat a projected person’s affection as their real feelings. The current Astro reading/archive structures do not provide a structural simulation domain. Add isolation across prompts, reads, writes, captions, summaries, suggestions, exports and delayed outbox retries. A prompt disclaimer alone is insufficient. Selected real context enters through a permitted, versioned scenario packet; no simulated event returns as autobiographical evidence. Maintain a world/branch identity, parent checkpoint, accepted actions, fictional consequences, assumptions, cast, source dependencies and narrative authority. Preserve the difference between **generated**, **delivered**, and **accepted/committed**. The model may prepare the next beat; it cannot advance the human through unaccepted actions or unheard consequences. On intervention, stop audio, cancel obsolete generation where supported and reject stale results by branch/revision/turn identity. The player’s meaning must reach the world operation layer. After rewind or fork, remove abandoned futures from provider context as well as playback. Save interrupted attempts distinctly; do not resurrect them on reconnect. A prototype can acknowledge complete short beats, then clarify an interrupted partial scene when necessary. It must disclose that coarse boundary rather than pretending to know the exact word heard. Finer playback offsets require actual transport verification. A transcript records conversation; a checkpoint must support the next action. Compacted summaries remain rebuildable derivatives of retained events, not replacements for provenance. Pause/resume should preserve the open situation and narrator role. Privacy corrections invalidate affected context and summaries. Deletion must remove sensitive sources and revealing derivatives, including pending copies; preserve only safely separable fiction. Accurate continuation may become unavailable. These rules inherit the [sovereignty contract](/documents/sovereign-universe-architecture/). ## Reconciliation with the broader system | Existing concept | E07 relationship | |---|---| | Astro and personal memory | Reuse permitted recall and conversational presence; do not require a life dossier, birth chart or complete biography | | Canonical event history | Extend provenance to simulation and actual play experience; authored state is not real-world evidence | | Timelines / parallel selves | Parent checkpoints and explicit changed assumptions support open-ended branches; no exhaustive tree required | | Simulations / credits | Meter authorized computation, not the importance of a choice. Pausing, stopping, inspecting or restoring an existing checkpoint must not secretly generate charges; fresh generation has disclosed limits | | Paid service and preservation | Provider usage may remain billable after cancellation; distinguish that from a world change. Do not claim “unheard means free” or threaten history loss to sell continuation | | Routines and skill practice | Can supply optional context or activities. Play creates no reminders, obligations or automated goals | | Creator and media concepts | Storyteller transfer is the small creator primitive; publishing and sharing remain separate permissions | | SimSim Us | Later shared authorship over a jointly permitted projection; never merge private histories by joining | | First Light / G1 | Its reviewed authored scope and historical gates remain intact; this is a separate interaction experiment | | WooWoo / spiritual lenses | May enrich reflective play without certifying supernatural instruction or predictive accuracy | ## Contradictions and missing primitives The earlier “one fictional scene” test is useful but cannot establish the value of personal context. Use both pure imagination and a small permitted personal premise. Likewise, a button that stops speech can test audio/state integrity but cannot validate the founder’s spontaneous, eyes-closed intervention hypothesis. Current voice paths differ. Router speech has an explicit interruption control; GPT-Live delegates substantive work; the guarded Realtime canary currently uses onboarding/sky tools. Its recorded server cutoff was restored to ten minutes, with paid native voice still disabled in the cited release. None is already a SimSim world-operation interface. A twenty-minute target needs separately bounded voice testing, not a silent removal of the cutoff. The missing minimum is: simulation isolation; a small persistent world/branch; an intent-to-world proposal/validation path; a delivered-scene boundary; narrator authority; and pacing/turn-taking. Reuse account, storage, source versions, provider metering and cancellation patterns. Do not infer that existing technical counters establish the human meaning of an interruption. ## Smallest MVP that tests the actual hypothesis One consenting adult, Astro and one world. Start in the blue-planet scene or a personally meaningful situation built from a few approved excerpts. Enter promptly, with minimal setup. Allow freeform action, spontaneous spoken interruption, a pause, one rewind/fork, storyteller takeover and later resume. No obligatory lesson. Keep text and a manual stop as accessible fallbacks. Reuse the smallest guarded native voice path that can actually deliver barge-in, routing speech to simulation operations and a saved branch rather than ordinary autobiographical chat. Inspect its client and server together before integrating; do not claim provider cancellation support without testing it. No new provider, general paid-voice activation or broad infrastructure is implied. An earlier controllable text/router-voice harness remains useful for validating isolation, state and persistence. Label it a harness until spoken intervention works: it cannot pass the central product gate through button presses alone. Test a short scene first. Once transport, cost and recovery behavior support it, offer approximately twenty minutes with voluntary pauses and exit. Do not force duration or interpret leaving as failure. Compare the two starting premises and, where practical, ordinary roleplay without persistence. The value must come from agency, consequences and return—not the branding of an otherwise identical chat. ## Instrumentation that leaves room for immersion With opt-in evaluation, collect small session events, not a new permanent audio diary. Prefer on-device processing; upload only permitted minimal measurements. Do not retain intimate dialogue or an inferred psychological profile to explain a metric. No new analytics is implemented by this document. | Signal | Useful interpretation | Avoid claiming | |---|---|---| | Speech begins during narration | Candidate spontaneous intervention, with playback/turn context | Every overlap is engagement; echo and recognizer errors exist | | Freeform action after interruption | Candidate agency, when followed by a coherent changed scene | A frustration repair proves immersion | | Rewind/fork, perspective shift, storyteller takeover | Player explores or authors possibilities | More operations always mean more value | | Voluntary continuation and later resume | Desire to inhabit that world again | Silence or missing telemetry means abandonment | | Active play, listening and thinking time | Distinct aspects of a session; duration is secondary | Twenty elapsed minutes proves presence or wellbeing | | Stop latency, first audio, stale audio/state, repairs, cost | Whether the interaction is responsive and reliable | Transport performance establishes emotional value | Use a single optional exit prompt: “Did you feel you could act inside the scene? Anything that broke it?” Later, a replayable bookmark can support return without a new reminder. Classify interventions, when possible, as in-world agency, narrative repair, audio repair or unknown using only permitted data; preserve uncertainty. Compare raw overlap with successful agency events so the team cannot improve a dashboard by making narration more irritating. ## Red-team and ways to make it more magical | Risk | Response | |---|---| | Over-narration steals the player’s role | Stop at the consequential opening; accept “don’t decide for me”; shorten after repairs | | Empty choices produce the same story | Carry a concrete consequence forward and make it discoverable later | | Memory feels creepy | Preview selected context, permit omission, use fictional names in shared examples, and offer pure fantasy immediately | | Emotional scenes pressure compliance | No attachment threats, forced lessons, purchases at emotional cliffhangers or optimization for dependence | | Imagined outcomes feel predictive | Name material assumptions at entry and major jumps; never authenticate another person’s thoughts | | The narrator makes every branch prove its advice | Invite contrary assumptions and user-authored endings; reflection remains a hypothesis | | Immersion breaks through latency and lost state | Short coherent beats, clean cancellation, faithful resume and visible failure recovery | | Endless possibilities become indecision | Explore a bounded set when requested; allow a satisfying ending without a real-life decision | | Metrics distort the experience | Measure coherent agency and voluntary return alongside repairs and burden; never maximize interruption counts | Increase magic through persistent character promises, an ordinary Tuesday five years later, silence at the right moment, an unexpected return to an earlier detail, a world that remembers the player’s invention, rapid storyteller handoff, contrasting assumptions from the same moment, and a concise bookmark that restores a scene tomorrow. None requires 3D. No character voice should imply that a real person endorsed it. Later renderers can add a still image at a meaningful moment, ambient sound, music, portraits, video, persistent locations or spatial environments. Each consumes an authorized scene projection tied to a branch/revision. A renderer never becomes the source of truth or changes the world without an admitted action. Graphics, microphones and cameras remain optional; added media inherits the same privacy boundary. ## Acceptance and documentation lineage Before calling the interaction playable, demonstrate a spoken intervention that stops obsolete narration and changes the next scene; human storyteller takeover and return; an alternative branch that preserves its parent; accurate leave/reopen/resume; inspectable source corrections; and fictional events excluded from subsequent ordinary Astro recall. Include a delayed callback, unfinished beat and deletion case. These are proposed tests, all unrun in this documentation change. This E07 direction follows the [attic and voice-world account](/documents/voice-worlds-evolution-2026-09-29/), the [earlier spoken-world account](/documents/voice-story-evolution-2026-09-29/), the [universe proposal](/documents/universe-architecture/) and the [sovereignty contract](/documents/sovereign-universe-architecture/). The [source crosswalk](/review/source-crosswalk/) records evidence limits; the [decision index](/review/decision-index/) records this addition without claiming another review round. The [outgoing E06 edition](/editions/2026-09-29-voice-worlds/) remains unchanged. ## SimSim North Star Astro understands your universe. SimSim lets you experience the ones that might have been, might become, or could never exist. Build a living possibility space with shared authorship. Let the human act before the narrator chooses for them. Treat interruption as agency, silence as space and storyteller transfer as play. Preserve consequential worlds and their branches; keep fiction separate from real memory. Let personal exploration and purposeless adventure both belong. Use imagination first and renderers when they help. The guiding experience: **I close my eyes. A world takes shape. Something matters. I interrupt. The universe listens.**