# SimSim Protocol — a new experimental direction **E02 · September 18, 2026. Founder brainstorm, chronological addition.** This follows [E01: private Universes](/documents/universe-evolution/). It does not rewrite the [30-round baseline](/review/master-plan-round30-baseline/) or retroactively review the new ideas. The [previous living edition](/editions/2026-09-18-universe-model/) remains preserved. The transcript supplied for this revision ended during its product-language section; no missing conclusion has been invented. ## The new thesis Astro is the companion. SimSim is the person's permitted Universe of history, goals, constraints, choices and uncertainty. **SimSim Protocol is a proposed permissioned handshake between private human Universes.** Its useful output is a possible shared experience, not a score ranking human worth. The question is no longer only “What might I do?” It becomes “What unrealized possibility exists between us, here, now?” The context can include two people, a place, a time, an activity and their current intentions. They expose small approved projections, not their whole private histories. “ChatGPT knows about you; SimSim models your universe” is positioning language to test, not a verified limit on current or hypothetical future assistants. Memory, voice and personalization alone may be weak differentiation. A consented longitudinal record of decision-time assumptions and real outcomes, coupled with useful intersections, is the proposed stronger distinction. Interoperability and actual usefulness remain unproven. ## Reality is the game **Live → Remember → Simulate → Choose → Live → Learn.** A successful interaction may end with “Go live it,” followed by closing the app. Protect attention rather than optimizing conversation volume or screen time. An eventual counterfactual autobiography preserves what was known and believed when an old decision was made. Revisiting an unchosen branch is an exploration of a saved model, not recovered alternate history. More branches cannot make a mistaken assumption true. ## Why an intersection, rather than a compatibility score? A shared dinner, a short walk, a creative conversation or a learning activity gives people something concrete to evaluate. “Meeting may be worthwhile” is a hypothesis. “94% compatible” would imply evidence this system does not have. Mutual hard constraints filter possibilities before softer preferences rank them. Neither similarity nor an agent's enthusiasm overrides a refusal. Feedback from one person remains their account; it cannot establish the other person's feelings or automatically authorize renewed contact. ## From applications to Rooms A Room supplies context and eligible opportunities. Dalat.app is the intended first external Room provider for a controlled event, subject to actual integration access and permission. Events, communities, conferences, hostels and other platforms are possible later providers, not adopted integrations. The first implementation on this site is a standalone **two-person invitation room**, identified separately from Dalat.app. It does not scan an event roster or search people nearby. Provider-neutral room naming prevents the experiment from treating one application as the whole future protocol. ## V0.001: SimSim Us [Try the bounded pilot](/us/). Create a small structured card, show a private QR or link, and invite one person to create theirs. Compare only approved fields. Each person answers independently. Only a mutual yes reveals the chosen names. A private optional outcome report closes the smallest loop. The implementation uses transparent deterministic rules. It does not call an LLM, infer hidden traits or negotiate between autonomous agents. Text and a phone keyboard's optional dictation are enough to test the interaction; no recorded voice service is required. Unchecked activities are excluded, a shared time window is required, and no forced suggestion is returned if these conditions fail. This deliberately narrower implementation tests one behavioral premise: will two people voluntarily expose small cards and find a suggested experience worth trying? It does not test predictive validity, global discovery, viral growth or a generalized protocol. See [implementation boundaries](/documents/simsim-us-pilot/). ## Decision register | Decision | Reason | Status | |---|---|---| | E02-01: append a dated chapter | Historical reasoning must remain inspectable | Published documentation; earlier editions retained | | E02-02: shared experiences as the social object | More actionable than an opaque compatibility percentage | Hypothesis awaiting human observation | | E02-03: minimal temporary projections | Whole-history disclosure is unnecessary for one introduction | Bounded two-hour pilot sharing window | | E02-04: mutual human approval | Discovery is not permission to reveal identity or contact | Enforced in the pilot API; human comprehension untested | | E02-05: pairwise before a room pool | Smallest end-to-end test; fewer attention and privacy costs | P0 implementation; P1 not built | | E02-06: rules before autonomous negotiation | Establish whether the basic experience works before buying complexity | No LLM or autonomous agent in this pilot | | E02-07: private outcomes, not engagement metrics | The intended value happens away from the application | Optional report, no sender dashboard | | E02-08: preserve provider independence | Dalat.app should be a Room, not the permanent protocol | Conceptual boundary; external integration pending | ## Later branches, retained rather than promised P1 may compare a small opt-in room pool and surface very few possibilities. P2 may test explicitly authorized Astro-to-Astro negotiation only after simpler comparison earns its burden. Agent dating and “Astro, go socialize for me” require scoped representation, platform permission, meaningful mutual constraints and independent human introductions. Bots may assist discovery; humans conduct relationships. Selected ancestor memories could become interactive family history, without claiming the representation is literally the ancestor or has continuing authority. Visual embodiment may evolve from voice/light to presence, face, spatial world and eventually VR. These are separate research branches, not steps required for V0.001. No autonomous dating-app operation, message scraping, universal social graph, global nearby discovery, cryptographic protocol framework, large simulation farm, 3D world or realistic avatar is included now. ## What would change our minds? In a separately arranged, consenting small event, observe onboarding burden, card comprehension, mistaken expectations, unwanted pressure, mutual suggestions and voluntarily reported real experiences. Include no-overlap, decline, expired invite and failed-meeting cases. “We probably would not have met otherwise” is valuable reported experience, not a causal estimate. If simple activity overlap is useful, improve that experience before adding agents. If people disclose more than they intended, feel scored, or receive irrelevant suggestions, revise the interface and constraints. If the process consumes more attention than ordinary introductions, simplify or stop. No human pilot results, Dalat.app integration, validated network effect or paid demand are established by this release.