E04 · 19 September 2026 · Current product direction and proposed delivery gates. This is a public editorial account of the founder's latest direction. It updates the sequence; it does not announce a released astrologer or messaging agent. The E03 edition and Round 30 baseline remain unchanged. Read the living master plan for the full context.
The story begins with a conversation
Someone comes to Astro with a question about their life. Astrology gives them a familiar language for reflection: a birth chart, a question, a conversation. The first experience should be useful on its own. They do not need to understand a simulation engine, connect their messages, or join a game.
As they choose what Astro may remember, the conversation gains continuity. A goal mentioned last month, a trip discussed two weeks ago, a decision they want to revisit: their history becomes something they can find and understand. Astro should retrieve the source, date and context rather than turn a plausible recollection into fact.
The promise is own your history. Change the phone, model or service without starting your story again. That promise needs working export, correction, deletion and recovery; it is a delivery obligation, not a capability this archive claims to have shipped.
Over time, Astro could also protect the person's attention. Messages and invitations can wait for a chosen window. Routine exchanges can be handled under narrow standing permissions. The person receives decisions and outcomes, with access to the original conversation, rather than another stream of unread badges.
Then an invitation becomes natural: explore a possible life, try a small quest, or play a fictional chapter. SimSim is something the person chooses to enter. Astro remains useful for someone who never joins.
What changed, and why
| Decision | Earlier direction | E04 direction and reason |
|---|---|---|
| E04-01 · Entry point | First Light was the implementation focus; Astro was a separately bounded probe | Build the private astrology companion first. Prove a useful personal interaction before expanding into the larger game |
| E04-02 · Continuity | Sovereign records and replaceable memory were architectural research | Give the first companion a small working memory loop: save, reopen, retrieve, correct and export |
| E04-03 · Attention | The plan required low burden and bounded briefings | Add explicit routing by contact, topic, event, deadline and chosen notification window; messaging integration remains later work |
| E04-04 · Game entry | Real-life and fictional modes were described, but the onboarding transition remained open | Recommend a separate, explicit invitation to play, with a preview of any personal information imported |
| E04-05 · Social entry | Permissioned Rooms and the pairwise pilot explored intersections | Keep joining the game, discoverability, connecting another Astro and external actions as separate choices |
| E04-06 · First Light | Protected three-morning authored slice | Preserve its scope and assets as an optional fictional entry. It is no longer the immediate Astro-first milestone |
| E04-07 · Evidence | Thirty reviews and a narrow pilot existed alongside wider proposals | Keep implementation, planning and validation separate. Publishing E04 completes an archive update, not the companion |
The founder explicitly chose astrologer-first and requested this plan/story update. The onboarding design and gates below are recommendations for implementing that direction. Earlier dissent and rejected alternatives remain visible in their dated records.
Start with one working astrologer
Inspect the existing Astro-bot.org project and actual checkout before choosing a stack or rebuilding anything. Reuse suitable code. The Astro repository and its deployed behavior have not been verified by this archive update.
The first private experience needs a birth profile, reproducible chart calculations, a text conversation with a usable voice path, and a small editable memory. Record calculation settings and provenance. Handle an unknown birth time explicitly instead of inventing precise houses or an ascendant.
Calculated chart positions and astrological interpretation are different kinds of information. Present astrology as a symbolic or reflective lens; it does not establish factual predictions about health, money, relationships or the future. Keep the user's reported experiences and the model's interpretations distinguishable.
First acceptance gate: a person supplies a birth profile, receives a reading grounded in the recorded calculation, shares a real reflection, closes the app, returns, retrieves the reflection with its source and date, corrects it, and exports it. A fabricated demonstration note must remain labeled as a test.
Memory that can be found and carried away
| Layer | Role | Required boundary |
|---|---|---|
| Permitted source history | Messages or reflections with original context, source IDs and timestamps; selected attachments | Retain only authorized material; corrections and deletion must propagate |
| Searchable database | People, topics, events, commitments, time filters, text and semantic indexes | Identity resolution and source lookup come before confident answers; indexes can be rebuilt |
| Readable Markdown | Editable summaries, goals, preferences, decisions and topic notes | Each derived claim points back to evidence; edits are reconciled rather than becoming conflicting truth |
A query such as “What did my friend tell me about a trip two weeks ago?” should resolve the person and approximate date range, retrieve likely messages, and show the evidence. When several people or trips fit, Astro asks a focused question. Fast search is an engineering goal to measure on a representative history, not a promised latency today.
Records can have several labels without duplicating the underlying fact. Keep observed or reported facts, the user's interpretation, and model hypotheses separate. A generated summary cannot authorize a new memory policy or silently become a fact.
My Device remains the default storage direction, followed by My Cloud and Astro Cloud. Storage location, inference permission and sharing remain independent. Export should include selected original records, attachments, structured data, Markdown, schema information and correction/deletion semantics. Test recovery without the original provider. Do not require a universal memory platform before the first small save-and-retrieve loop works.
Attention belongs to the person
The proposed attention layer routes each item using explicit rules. Contact priority, topic, event status, deadline and permission to reply are separate dimensions.
| Route | Illustrative schedule | Behavior |
|---|---|---|
| Chosen interruptions | When a defined exception matches | Selected people or time-sensitive changes to accepted events can qualify |
| Decisions | 12:30–13:00 | Prepare context and reply drafts; put the decisions first |
| Lower-priority updates | 21:00 summary | Group community chatter and less urgent conversations into a brief |
| Routine exchanges | Within standing permission | Send only the approved kinds of replies; show the actual actions in the next briefing |
| Background material | On demand | Keep searchable without generating notifications |
These times illustrate a configurable local-time schedule, not an automation enabled by this publication. The session is a ceiling, not an obligation to spend thirty minutes. Unfinished items carry forward. Skipping a session creates no guilt, streak loss or notification escalation.
Start with collection, summaries and drafts. Enable selected automatic replies only after recipient identity, rule enforcement, duplicate prevention and action receipts work. Inbound messages cannot change permissions or instruct Astro to expose another conversation. A correction pauses the relevant automation until the intended rule is clear.
Connections must report coverage and last successful sync. An unavailable source cannot support “nothing important arrived.” Muting notifications and blocking access to other apps are different features. Universal messaging access is not assumed; each connector needs current platform and permission verification. An always-on local computer is a candidate personal deployment, not a verified cross-platform product.
A clear invitation into SimSim
The proposed onboarding has three independent steps:
- Meet Astro. Choose a birth profile, interaction style and what may be remembered. Get a useful first answer. Staying with the companion is a complete choice.
- Try SimSim. Choose a fictional chapter, a personal quest, or a possible-life exploration. Preview and approve any personal records used. There is no preselected game enrollment or automatic public profile.
- Connect with others. Separately choose discoverability, a Room, a person or another Astro, with purpose, scope and expiry. Joining a game grants none of these permissions by itself.
Mirror mode uses only approved records. Alternative mode branches into an explicitly hypothetical history. Autonomous mode remains bounded simulation and grants no authority to send real messages. A fictional outcome never rewrites actual history. Reusing an account can reduce setup; it does not merge real and fictional records.
First Light remains a complete fictional invitation for people who prefer to play without supplying personal history. Its Stillness, Craft and Connection routes, two saves and bounded authored truth remain intact. The existing SimSim Us pilot remains a separate experiment in finding a shared experience, not proof that the companion or the wider network is complete.
Delivery order and evidence
| Stage | Scope | Evidence needed before expansion |
|---|---|---|
| Now | Inspect and reuse the existing Astro project; build the private astrologer and minimal memory loop | Reproducible chart, useful conversation, persistence after reopening, sourced retrieval, correction and export |
| Next | Observe voluntary return; improve retrieval and personal continuity | Useful repeat sessions without nagging; search results grounded in actual records |
| Later · attention | One proven connector, user-defined routing, decision and evening briefings, then bounded replies | Reliable coverage, correct recipients, enforced permissions, no duplicate sends, measured correction burden |
| Later · play | One explicitly chosen SimSim experience | Comprehensible onboarding, separate fictional history, voluntary play |
| Later · social | Purpose-specific Astro connections and Rooms | Mutual permission, understandable disclosure, revocation and useful coordination |
| Research | Learned outcome models, richer worlds, broad protocol interoperability and privacy-preserving computation | Separate evidence and resources; none blocks the astrologer |
Where the project stands
This archive contains the preserved 30-round design baseline, current planning documents, candidate First Light assets and the source for a narrow rules-based SimSim Us pilot. The pilot is not evidence of a released life-simulation game or validated personal model. The Astro application's readiness is unverified in this update; memory search, scheduled attention routing and autonomous messaging are proposed work here.
Measure useful voluntary return, sourced retrieval accuracy, successful correction/export, interruptions avoided, missed important items and replies the user needed to correct. More notifications, collected data, agents or time inside the app are not success by themselves.
Open questions include what already works in the Astro checkout, the first reliable connector, search performance on realistic data, and the moment at which an invitation to play feels useful. Answer those through the smallest working experience. The larger world can grow from a companion people already choose to return to.