Eve library

OSHUN Unified Consumer, Admin, and Core Systems PRD

PRD

Revision 1

On this page

Status: Draft v1.0
Date: March 22, 2026
Owners: OSHUN Product, OSHUN Engineering, OSHUN Design, OSHUN Safety, OSHUN Platform
Requested Filename: OSHUN_WEB_APP_ADMIN_V1_PRD.md
Scope Note: The filename is retained per request, but this document defines the end-state product requirements for the full OSHUN ecosystem: customer-facing web and mobile apps, admin-facing web and mobile apps, and the critical agentic, media, research, trust, compliance, and operations subsystems that power them.
Related:

  • docs/releases/v1/reports/unified-product-gap-analysis-and-nisaba-integration-report-2026-03-22.md
  • docs/releases/v1/reports/subsystem-ownership-matrix.md
  • docs/releases/v1/specs/product-brief.md
  • docs/adr/ADR-0029-oshun-studio-audit-and-compliance-surfaces.md
  • docs/adr/ADR-0035-oshun-studio-review-and-approval-workflows.md
  • docs/reference/data-ownership-matrix.md (legacy pre-V1 cross-domain snapshot)

Ownership note: when this PRD conflicts with older ownership docs, this PRD and docs/releases/v1/specs/program-charter.md, together with docs/releases/v1/reports/subsystem-ownership-matrix.md, define the current OSHUN V1 platform model. docs/reference/data-ownership-matrix.md remains historical context only.


Table of Contents#

  1. Executive Summary
  2. Product Vision
  3. Product Thesis
  4. Scope
  5. Product Principles
  6. Users and Stakeholders
  7. Product Surface Map
  8. Experience Narrative
  9. Domain Model and Product Roles
  10. Cross-Domain Shared Objects
  11. Customer-Facing Product Requirements
  12. Admin-Facing Product Requirements
  13. AI Assistant, Teacher, Guru, and Expert System Requirements
  14. Research, Authoring, and Agentic System Requirements
  15. Image, Audio, Video, Avatar, and Model System Requirements
  16. Platform Architecture and Canonical Ownership
  17. Review, Safety, Compliance, and Governance Requirements
  18. Data, Memory, and Personalization Requirements
  19. Search, Discovery, Recommendation, and Knowledge Graph Requirements
  20. Monetization, Entitlements, and Commerce Requirements
  21. Analytics, Experimentation, and Quality Requirements
  22. Non-Functional Requirements
  23. Operational Readiness Requirements
  24. Success Metrics
  25. Phased Delivery Model
  26. Anti-Patterns to Avoid
  27. Acceptance Criteria for the End-State V1

1. Executive Summary#

OSHUN V1 is the fully unified contemplative intelligence ecosystem.

It is not a loose bundle of apps. It is one product system with:

  • one account
  • one identity
  • one shared memory and preference model
  • one cross-domain activity graph
  • one grounded knowledge substrate
  • one controlled generation substrate
  • one rights and provenance substrate
  • one review and governance substrate
  • one consumer shell across web and mobile
  • one admin shell across web and mobile
  • one cohesive product philosophy centered on Tara, meditation, awakening, discipline, truth, wonder, and textual depth

The product combines six conceptual layers:

  • Tara: contemplative center and experiential spine
  • Arete: disciplined embodiment and life practice
  • Veritas: epistemic clarity and grounded claims
  • Nyx: wonder, sky literacy, and cosmological perspective
  • Nisaba: textual depth, lineage, scholarship, and comparative wisdom
  • Sophia: grounding, citations, evidence, retrieval, knowledge graph, and research integrity behind the scenes

The system is operated through a parallel but deeply connected operations layer:

  • OSHUN Admin Web: the canonical operator cockpit
  • OSHUN Admin Mobile: urgent review, escalation, and on-call companion

The product is powered by platform systems with explicit ownership:

  • Iris: persistent memory, assistant identity, memory controls, memory privacy
  • Psyche: real-time multimodal text/voice/avatar session orchestration
  • Lilith: contemplative teacher logic, persona safety, spiritual-domain tone policy, voice quality, and domain-specific review rules
  • Sophia: grounding, citations, source graph, evidence packs, notebooks, and research workflows
  • Isis: canonical generation control plane, model registry, workflow registry, provenance, release gates, and execution policy
  • Civitai: external model-intelligence and asset-source layer, not the runtime platform of record
  • ComfyUI: workflow intermediate representation and execution adapter layer
  • RunPod: scalable execution substrate behind controlled platform workflows
  • ElevenLabs and equivalent providers: voice providers behind Oshun-owned safety, disclosure, watermarking, and policy layers

The goal of this PRD is to define the full end-state system once the major gaps identified in the March 22, 2026 gap-analysis report have been addressed.


2. Product Vision#

OSHUN should become the most compelling contemplative intelligence product in the market: a product that helps people calm the mind, cultivate discipline, understand reality, experience wonder, and study wisdom traditions with unusual depth and unusual trustworthiness.

The customer should feel that:

  • the system understands their long-term path
  • the product is coherent rather than fragmented
  • the assistants feel alive, attentive, and helpful without becoming deceptive
  • every deep claim can be traced to evidence when appropriate
  • every generated artifact has provenance
  • every sensitive workflow is governed
  • every domain meaningfully reinforces the others

The operator should feel that:

  • every important action is reviewable
  • evidence is easy to find
  • moderation and appeals are deterministic
  • rights and licensing posture is explicit
  • models and personas can be approved safely
  • incidents, privacy requests, and release gates are handled from one place

3. Product Thesis#

The product thesis is:

  1. People do not actually want six disconnected tools. They want one trusted, elegant system that helps them live with more clarity, depth, and intentionality.
  2. Tara should be the organizing emotional center. OSHUN should feel like a contemplative product first, even when it is helping the user research, verify, study, or explore.
  3. The unification layer must be deeper than navigation. The product must share objects, continuity, personalization, assistants, provenance, and review.
  4. Nisaba is essential because it adds lineage, meaning, source depth, and comparative wisdom; without Nisaba, the product lacks its deepest moat.
  5. Sophia is essential because the product cannot scale trustworthy answers, explanations, or research without a canonical grounding substrate.
  6. Premium assistants only work when memory, embodiment, grounding, disclosure, and safety are designed together instead of bolted together.
  7. Civitai, ComfyUI, RunPod, ElevenLabs, and similar tools are important, but they must be normalized under Oshun-owned contracts and governance before they become production-critical parts of the system.

4. Scope#

This PRD defines the end-state V1 requirements for:

  • OSHUN customer-facing web app
  • OSHUN customer-facing mobile app
  • OSHUN admin-facing web app
  • OSHUN admin-facing mobile app
  • shared APIs, orchestration layers, and domain adapters required to support those apps
  • grounding, research, citation, and knowledge graph systems
  • assistants, personas, teacher systems, memory systems, and avatar systems
  • image, audio, video, and media-generation systems
  • model and asset intake systems, including Civitai-connected workflows
  • moderation, compliance, review, licensing, provenance, watermarking, and audit systems
  • analytics, experimentation, observability, incident management, and release governance systems

This PRD does not define:

  • standalone marketing-site copy
  • raw implementation tickets
  • team staffing plans
  • pricing tables at the SKU level
  • legal language that should ultimately live in policy or terms documents

5. Product Principles#

5.1 Philosophy First#

OSHUN must feel like one philosophy expressed through multiple experiences, not several unrelated apps gathered under one roof.

5.2 Tara as the Center#

Tara is the emotional and behavioral center of gravity. The system should orient the user toward presence, reflection, calm transitions, intentional practice, and awakening-oriented growth.

5.3 Cross-Domain Depth#

Every domain should matter on its own, but also become better in the presence of the others.

5.4 Trust as a Feature#

Grounding, citation, provenance, auditability, moderation, and review are not back-office concerns. They are user-value features.

5.5 Human-AI Clarity#

Premium assistants are allowed to feel fluid, warm, expressive, and highly capable. They are not allowed to impersonate undisclosed humans or hide memory behavior.

5.6 Cohesion Over Parity#

Web and mobile should feel like one product, but they do not need identical feature sets. Each platform should be strongest in the contexts it serves best.

5.7 Governed Creativity#

Media generation and synthetic personality systems must be constrained by rights, review, policy, and provenance from the start.

5.8 Operator-Grade Design#

Admin tools must be as intentionally designed as consumer tools. Internal users should get a premium, coherent, safe, and efficient product, not an afterthought.


6. Users and Stakeholders#

6.1 Customer Personas#

The Seeker#

Wants meditation, rituals, meaning, and a calmer relationship with technology.

The Builder of Character#

Wants structure, habits, reflection, journaling, discipline, and coach-like guidance.

The Evidence-Oriented User#

Wants grounded facts, source trails, nuanced explanations, and clear confidence signals.

The Wonder-Oriented Explorer#

Wants astronomy, cycles, events, sky awareness, symbolic depth, and awe.

The Deep Reader and Student#

Wants texts, translations, annotations, manuscripts, comparative reading, and conceptual lineage.

The Power User / Researcher#

Wants notebooks, evidence packs, comparative analysis, citation trails, and expert-level reading/research workflows.

6.2 Admin Personas#

Trust and Safety Reviewer#

Handles moderation, risk scoring, crisis patterns, abuse reports, and appeals.

Editorial and Content Operator#

Reviews articles, meditations, lessons, rituals, prompts, curricula, and premium content updates.

Persona and Avatar Operator#

Approves teachers, guides, stylized avatars, voices, disclosure states, and persona policy packs.

Model Ops and Media Operator#

Approves models, staging recipes, workflows, provider connections, and production promotions.

Research Integrity Reviewer#

Audits citations, evidence packs, source provenance, summaries, comparative claims, and AI-generated reports.

Support and Member Success Operator#

Handles account issues, billing, subscriptions, entitlements, memory questions, and continuity across devices or surfaces.

Privacy and Compliance Operator#

Handles consent reviews, DSAR workflows, retention requests, deletions, exports, and sensitive investigation flows.

On-Call and Incident Operator#

Handles service degradations, abuse spikes, provider outages, workflow failures, and urgent approval escalations.

Product and Executive Analyst#

Uses dashboards, cohort analysis, quality reports, and release-readiness views.

6.3 Platform Stakeholders#

  • Product
  • Design
  • Engineering
  • Safety
  • Legal and compliance
  • Support
  • Research and knowledge operations
  • Editorial
  • Growth and lifecycle
  • Partnerships and licensing

7. Product Surface Map#

7.1 Customer Surfaces#

OSHUN Mobile#

The primary habit and daily relationship surface.

Key roles:

  • rituals
  • meditation
  • habits
  • reminders
  • voice conversations
  • quick answers
  • nightly sky moments
  • saved passages
  • daily continuity
  • notifications

OSHUN Web#

The primary breadth, depth, dashboard, and workspace surface.

Key roles:

  • immersive home/dashboard
  • cross-domain discovery
  • long-form reading
  • research and evidence exploration
  • deeper journaling and planning
  • study and annotation
  • profile and account management
  • creator and premium experiences

7.2 Admin Surfaces#

OSHUN Admin Web#

The canonical operational cockpit for all review, governance, support, policy, release, analytics, and platform operations.

OSHUN Admin Mobile#

The urgent actions companion for queues, approvals, escalations, incidents, and executive/on-call awareness.

7.3 Platform Surfaces#

Sophia Workbench#

Operator and pro-facing research/grounding workbench that powers evidence-heavy flows but is not the default consumer shell.

Isis Operator Surfaces#

Model, workflow, execution, release, and provenance control plane surfaces.

Persona and Avatar Studio#

Shared internal surfaces for persona registry, avatar packs, voice packs, disclosure assets, and rehearsal/review workflows.


8. Experience Narrative#

8.1 What the Product Should Feel Like#

OSHUN should feel serene, capable, intelligent, and deeply integrated.

A user should be able to:

  • wake up to a Tara-led centering ritual
  • continue a discipline streak in Arete
  • get a grounded update from Veritas
  • see what is happening in the sky through Nyx
  • open a Nisaba passage connected to their current theme
  • talk to a teacher or guide by text or voice
  • save thoughts, passages, questions, and sources into one continuity system

The user should never feel like they are starting over between domains.

8.2 Daily Flow Model#

Morning#

  • Tara: wake-up ritual, breathing, intention
  • Arete: first deliberate action, plan, or discipline prompt
  • Veritas: one grounded briefing if relevant
  • Nisaba: one passage, note, or concept thread

Midday#

  • assistant: quick grounded answer or coaching exchange
  • Arete: recovery of plan after drift
  • Veritas: explain the significance of a developing event
  • Nisaba: source or quote lookup for a current question

Evening#

  • Nyx: sky events, cycles, wonder, observability
  • Tara: downshift, reflection, sleep transition
  • Nisaba: annotation, reflection, or comparative reading
  • journal/notebook: one shared closure space

8.3 Cross-Domain Moments That Must Exist#

  • A Tara session on impermanence should be able to recommend Nisaba passages, Arete reflection prompts, and Veritas context on a related modern issue.
  • A Veritas article should be able to show primary sources, lineage, related texts, and concept graphs through Nisaba and Sophia.
  • A Nyx event should be able to connect to ancient cosmologies, calendars, and comparative texts through Nisaba.
  • An Arete program should be able to tie disciplines to Tara rituals, Nisaba source passages, and reflective journaling.
  • Any assistant response that depends on evidence should be traceable back through Sophia.

9. Domain Model and Product Roles#

9.1 Tara#

Tara is the contemplative center.

Required product role:

  • meditation
  • transitions
  • rituals
  • emotional regulation
  • presence and awareness
  • sleep and recovery
  • meaning-oriented reflection

9.2 Arete#

Arete is disciplined embodiment.

Required product role:

  • habits
  • goals
  • routines
  • self-mastery
  • planning and review
  • reflection into action

9.3 Veritas#

Veritas is epistemic clarity.

Required product role:

  • verified reporting
  • claims and counterclaims
  • confidence and source signals
  • explanations with traceability
  • current events interpreted with care

9.4 Nyx#

Nyx is wonder and cosmological perspective.

Required product role:

  • astronomy
  • sky events
  • cosmological education
  • cycles and calendars
  • awe and perspective

9.5 Nisaba#

Nisaba is the textual depth layer.

Required product role:

  • reading
  • comparison
  • annotation
  • manuscripts
  • translations
  • morphology and lexicon
  • concept lineage
  • source trails across traditions

9.6 Sophia#

Sophia is the grounding substrate.

Required product role:

  • ingestion
  • retrieval
  • semantic search
  • source graph
  • evidence packs
  • citations
  • notebooks
  • grounded report generation
  • evaluation and research integrity

9.7 Relationship Model#

The product relationship model is:

  • Tara gives the product its center
  • Arete turns insight into disciplined life
  • Veritas prevents confusion and distortion
  • Nyx restores perspective and wonder
  • Nisaba adds depth, source lineage, and serious study
  • Sophia makes the whole system trustworthy when claims or research are involved

10. Cross-Domain Shared Objects#

The shell must unify the product through shared objects, not just shared chrome.

10.1 Canonical Shared Objects#

User#

One identity across customer and admin surfaces, with role-aware capabilities.

Profile#

Preferences, interests, accessibility needs, tone settings, disclosure preferences, notification settings, and spiritual/educational preferences where appropriate.

Continuity State#

The system’s answer to "where did the user leave off and what should happen next?"

Ritual#

A reusable, schedulable, context-aware flow spanning Tara and other domains.

Practice#

A deliberate repetition structure, often used in Arete and Tara.

Concept#

A shared semantic node that can connect claims, passages, sky phenomena, exercises, lessons, and assistant explanations.

Passage#

A textual unit with provenance, translation, commentary, and annotation support.

Claim#

A verifiable statement that can link to evidence, sources, status, and debate.

Source#

A canonical source object with provenance, license, versioning, citation fields, and related concepts.

Notebook#

A cross-domain workspace that can contain notes, passages, claims, sources, ritual reflections, tasks, audio snippets, and media references.

Collection#

A saved bundle of any meaningful object type.

Session#

Any active user interaction span, including meditation sessions, assistant sessions, study sessions, research sessions, review sessions, or generated media sessions.

Persona#

A durable AI identity with policy, memory rules, style, disclosure, and modality contracts.

Voice Profile#

A controlled voice asset with provider metadata, consent status, watermarking status, and allowed-use scope.

Avatar Pack#

A controlled visual identity pack for stylized or embodied AI characters, including still image assets, animation metadata, disclosure overlays, and allowed-use scopes.

Evidence Pack#

A structured bundle of sources, excerpts, rationale, and trace metadata for claims, reports, or grounded assistant turns.

Review Package#

The canonical unit for approval, rejection, appeal, escalation, or release gating.

Provenance Bundle#

A canonical lineage object for generated artifacts, model use, workflow use, asset source, and watermark signals.


11. Customer-Facing Product Requirements#

11.1 Shared Customer Shell Requirements#

The customer-facing shell across web and mobile must provide:

  • one authentication and session model
  • one home/dashboard model
  • one universal search and discovery model
  • one library and save model
  • one activity and continuity model
  • one notification model
  • one profile, privacy, and accessibility model
  • one assistant entry layer
  • one consistent design language
  • one domain switcher with Tara, Arete, Veritas, Nyx, and Nisaba
  • one cross-domain recommendation layer
  • one shared premium and entitlement model

Home Requirements#

The home/dashboard must support:

  • time-of-day awareness
  • continuation of the user’s active journeys
  • Tara-led centering
  • cross-domain recommendations with reason labels
  • upcoming rituals, sessions, and reminders
  • concise evidence-backed news or context when relevant
  • streaks and discipline summaries
  • saved reading and unfinished study
  • nightly sky moments and calendar-sensitive content

Search Requirements#

Search must support:

  • cross-domain blended results
  • domain filters
  • concept search
  • passage search
  • claim search
  • source search
  • assistant-assisted search refinement
  • saved and recent searches
  • web and mobile continuity of search context

Library Requirements#

Users must be able to save:

  • meditations
  • rituals
  • habits
  • claims
  • sources
  • passages
  • notebooks
  • sky events
  • assistant conversations or excerpts
  • generated media outputs where allowed

Activity Requirements#

The activity timeline must show:

  • sessions completed
  • notes created
  • passages saved
  • streak events
  • assistant highlights
  • research or reading continuation points
  • approvals or support case updates relevant to the user

11.2 OSHUN Mobile Requirements#

OSHUN Mobile is the primary daily habit surface and the primary real-time assistant surface.

Primary Mobile Jobs#

  • open and begin value in under 30 seconds
  • recover the user after drift
  • carry the user through rituals and habits
  • support voice/text interaction with assistants anywhere in the app
  • keep saved texts, notes, and highlights close at hand
  • support notifications that feel helpful, not noisy

Mobile Navigation#

The mobile app must support:

  • home
  • explore
  • activity
  • library
  • assistant
  • profile
  • deep-linked domain entry points
  • contextual modal surfaces for quick actions

Mobile Tara Requirements#

  • guided meditations
  • breathing sessions
  • sleep/downshift routines
  • quick reset flows
  • adaptive rituals based on time, mood, and continuity state
  • ambient audio support
  • journaling/reflection capture
  • teacher-led voice and text interactions

Mobile Arete Requirements#

  • habits, routines, and goals
  • day planning and review
  • streaks with humane recovery design
  • friction logging and self-discipline coaching
  • post-session reflection
  • assistant-supported planning and accountability

Mobile Veritas Requirements#

  • briefings and digest cards
  • claims and evidence summaries
  • read-later and follow-up workflows
  • assistant explainers with citations on demand
  • push-alert-aware story prioritization

Mobile Nyx Requirements#

  • nightly sky cards
  • event alerts
  • moon and planet visibility
  • cosmic perspective content
  • quick telescope/observation companion features where applicable

Mobile Nisaba Requirements#

Nisaba on mobile is a companion-first but still real product surface.

Required features:

  • reading queue
  • daily passage
  • highlights and annotation
  • quick compare view
  • concept thread follow
  • glossary and morphology lookup
  • saved manuscripts or editions
  • study reminders
  • assistant-assisted explanation with citations

Mobile Assistant Requirements#

  • universal entry from anywhere
  • voice-first and text-first modes
  • real-time response streaming
  • interruptibility
  • context carry-over from current screen
  • explicit disclosure
  • visible memory status
  • visible microphone/camera state when applicable
  • switchable personas where allowed
  • handoff from teacher mode to research mode to coach mode without losing continuity

Mobile Offline and Low-Connectivity Requirements#

  • cached routines and meditations
  • cached recent readings and notes
  • cached saved passages
  • graceful degradation for assistant and evidence-heavy flows
  • sync recovery with clear user feedback

11.3 OSHUN Web Requirements#

OSHUN Web is the primary breadth, depth, dashboard, and workspace environment.

Primary Web Jobs#

  • present the user’s broader life/research/practice picture
  • support long-form reading and evidence exploration
  • support notebook, comparison, and annotation workflows
  • provide deep configuration and account management
  • support creator or pro-level surfaces where relevant

Web Shell Requirements#

  • dashboard with multiple panels and continuity modules
  • universal search with advanced filters
  • split-view and multi-panel reading/research workflows
  • side-by-side comparisons for passages, translations, sources, or claims
  • assistant dock with context carry-over
  • deep library and collection management
  • stronger analytics and progress views for the user

Web Tara Requirements#

  • immersive ritual and meditation pages
  • long-form collections and programs
  • teacher pages
  • audio controls and transcript support
  • room-like contemplative spaces where appropriate

Web Arete Requirements#

  • program builder
  • weekly and monthly review
  • journals and reflection workspaces
  • plan boards and progress maps
  • accountability history and pattern analysis

Web Veritas Requirements#

  • story pages with source trails
  • timeline and counterclaim views
  • topic hubs
  • explainers with source sidebars
  • grounded research views

Web Nyx Requirements#

  • richer event pages
  • maps, sky charts, and educational modules
  • historical or symbolic overlays where appropriate
  • event calendars and observation planning

Web Nisaba Requirements#

Web is Nisaba’s deepest early surface.

Required features:

  • Nisaba landing/dashboard
  • reading workspace
  • side-by-side translation and edition comparison
  • manuscript view
  • annotation and highlights
  • concept graph exploration
  • lexicon and morphology panels
  • timeline and influence-network views
  • note-taking and notebooks
  • source export and citation workflows
  • personal collections and study programs

Consumer Sophia-Powered Web Requirements#

Sophia should usually be embedded rather than exposed as a raw shell tab.

Required web capabilities:

  • evidence sidebars
  • citation drill-down
  • notebooks
  • source sets
  • grounded report views
  • research assistant flows
  • explanation pages with citation trails

12. Admin-Facing Product Requirements#

12.1 Admin Product Positioning#

OSHUN Admin is a first-class product, not a collection of internal tools.

It exists to operate:

  • content quality
  • trust and safety
  • support and member operations
  • privacy and compliance
  • model and media systems
  • persona and avatar systems
  • research integrity
  • rights and provenance
  • release governance
  • incidents and operational reliability

12.2 Admin Shared Requirements#

The admin product must provide:

  • one admin identity and RBAC model
  • one queue and inbox model
  • one case model
  • one review and approval model
  • one escalation model
  • one audit and evidence model
  • one investigation model
  • one notification and on-call model
  • one policy and rules configuration layer
  • one analytics and quality dashboard layer

12.3 OSHUN Admin Web Requirements#

OSHUN Admin Web is the canonical operator cockpit.

Core Admin Navigation#

  • dashboard
  • inbox/queues
  • trust and safety
  • content operations
  • research integrity
  • support and customer operations
  • privacy and compliance
  • personas and avatars
  • model and workflow operations
  • rights and provenance
  • incidents and reliability
  • analytics and executive reporting
  • policy and configuration

Global Admin Capabilities#

  • universal search across users, artifacts, cases, sessions, personas, models, workflows, and audit events
  • saved views
  • role-based dashboards
  • keyboard-driven queue triage
  • side-by-side evidence review
  • bulk actions where safe
  • collaboration comments and annotations
  • deterministic approval states
  • exportable investigation bundles

12.4 Admin Mobile Requirements#

Admin Mobile is the urgent operations companion.

It must support:

  • triaging urgent review items
  • approve/reject/request-changes actions where policy allows
  • escalation and reassignment
  • incident acknowledgement and handoff
  • high-priority support cases
  • creator/reviewer alerts
  • push-based on-call workflows
  • compact investigation/evidence views
  • secure step-up auth for sensitive actions

12.5 Review and Approval Center#

This workspace operationalizes the accepted Studio review/approval ADRs as the platform-wide standard.

Required capabilities:

  • canonical review package envelope
  • deterministic approval stage graph
  • actor attribution
  • dependency modeling
  • rationale capture
  • request-changes flows
  • delegation and escalation with policy checks
  • immutable stage history
  • SLA clocks
  • release-validity checks
  • review templates by artifact type

Reviewable artifact types must include:

  • content assets
  • meditations
  • rituals
  • premium curricula
  • articles and explainers
  • claims bundles
  • source packs
  • grounded reports
  • personas
  • voice profiles
  • avatar packs
  • models
  • workflow templates
  • media generations
  • policy exceptions

12.6 Trust and Safety Center#

Required capabilities:

  • content moderation queues
  • user-report triage
  • crisis-risk review
  • self-harm escalation workflows
  • child-safety escalation workflows where applicable
  • harassment/abuse handling
  • image moderation review
  • audio abuse and cloned-voice review
  • avatar deception review
  • policy-violation taxonomy
  • appeal workflows
  • repeat offender tracking
  • safety metrics and drift detection

Required integrations:

  • Lilith domain-specific safety logic
  • media moderation services
  • account and session history
  • evidence bundles
  • notification and escalation systems

12.7 Content and Editorial Operations#

Required capabilities:

  • content calendar
  • drafts and publishing states
  • version history
  • content QA checklists
  • teacher/curriculum publishing flows
  • homepage and campaign curation
  • experiment-aware content targeting
  • localization review
  • accessibility review
  • content retirement and archival

12.8 Research Integrity and Evidence Operations#

Required capabilities:

  • source review queues
  • citation validation
  • evidence-pack review
  • claim-to-source linkage inspection
  • hallucination-risk review for generated reports
  • comparative-text review
  • notebook or source-set promotion workflows
  • provenance inspection for summaries and explainers
  • policy gating for unsupported claims

Required platform dependencies:

  • Sophia evidence services
  • shared concept graph
  • review packages
  • audit evidence

12.9 Support and Customer Operations#

Required capabilities:

  • case intake
  • omnichannel conversation history where connected
  • subscription and entitlement inspection
  • account recovery
  • cross-device continuity troubleshooting
  • memory/privacy question handling
  • refund and exception workflow
  • issue tagging and root-cause analysis
  • quality escalation to product or engineering

12.10 Privacy and Compliance Operations#

Required capabilities:

  • consent review
  • memory consent state inspection
  • disclosure audit review
  • DSAR export workflow
  • deletion workflow
  • retention exception workflow
  • region and residency controls
  • legal hold support
  • policy attestation support
  • audit export generation

12.11 Persona and Avatar Operations#

Required capabilities:

  • persona registry
  • persona family taxonomy
  • policy pack management
  • disclosure-state management
  • tone/style calibration review
  • memory-scope management
  • allowed-context management
  • voice profile linkage
  • avatar pack linkage
  • rehearsal and evaluation workflows
  • premium-cast release approvals

Supported persona families must include:

  • Tara meditation teachers
  • Arete discipline coaches
  • Veritas explainers and reporters
  • Nyx cosmic guides
  • Nisaba research guides and literature scholars
  • philosophy experts
  • self-discipline experts
  • literature experts
  • spiritual guides and gurus with explicit scope controls
  • support copilots
  • moderation copilots
  • operator/reviewer copilots

12.12 Model and Workflow Operations#

Required capabilities:

  • model registry
  • model versions and lineage
  • license review
  • content-risk tags
  • staging recipes
  • workflow templates
  • provider endpoints
  • environment promotion
  • health monitoring
  • cost monitoring
  • rollback
  • output spot-checking
  • provider failover policies

12.13 Rights, Provenance, and Watermark Operations#

Required capabilities:

  • provenance bundle inspection
  • source-asset lineage review
  • training-data source evidence where applicable
  • model-license review
  • asset-license review
  • watermark verification
  • synthetic media labeling verification
  • consent record inspection for voice/avatar use
  • exportable rights reports

12.14 Incident and Reliability Operations#

Required capabilities:

  • service health dashboards
  • queue health
  • provider health
  • incident declaration
  • severity assignment
  • runbook launch
  • blast radius inspection
  • customer-impact estimation
  • communication workflow support
  • postmortem attachment and closure

12.15 Policy and Configuration Center#

Operators need a dedicated workspace for platform rules, not scattered settings pages.

Required capabilities:

  • policy bundle management
  • prompt policy management
  • retrieval and grounding policy management
  • disclosure policy management
  • memory-policy configuration
  • moderation threshold configuration
  • appeal-policy configuration
  • voice and avatar policy configuration
  • region and retention policy configuration
  • workflow and provider allowlists
  • entitlement and feature-flag inspection
  • experiment guardrail configuration
  • change review and approval before policy promotion
  • rollback to known-good policy bundles

12.16 Executive, BI, and Readiness Views#

Required capabilities:

  • product health overview
  • trust and safety overview
  • support health overview
  • revenue and entitlement health overview
  • assistant quality overview
  • model and workflow readiness overview
  • release-readiness scorecards
  • launch go/no-go dashboards
  • exportable board/executive summaries

13. AI Assistant, Teacher, Guru, and Expert System Requirements#

13.1 Product Requirement#

OSHUN should offer the most compelling, premium, clearly disclosed AI guides and assistants reasonably possible in the product category.

They must feel:

  • warm
  • responsive
  • context-aware
  • multimodal
  • continuous across sessions
  • stylistically distinct
  • grounded when evidence is required
  • safe in contemplative and high-stakes contexts

They must not feel:

  • deceptive about being AI
  • uncontrolled
  • memory-invasive
  • casually hallucinated in evidence-sensitive flows

13.2 Assistant Families#

The system must support distinct assistant families with shared infrastructure.

Consumer-Facing Families#

  • Tara teachers
  • Tara sleep guides
  • Tara short-form reset companions
  • Arete coaches
  • Veritas explainers/reporters
  • Nyx guides
  • Nisaba scholars
  • philosophy experts
  • literature experts
  • self-discipline experts
  • study companions
  • general Oshun concierge

Admin-Facing Families#

  • moderator copilots
  • reviewer copilots
  • support copilots
  • research integrity copilots
  • model ops copilots
  • compliance copilots
  • executive briefing copilots

13.3 Assistant Experience Requirements#

Required capabilities:

  • text chat
  • voice conversation
  • streaming response
  • low-latency turn-taking
  • context from current task and screen
  • retrieval-augmented responses when appropriate
  • proactive follow-up where user-authorized
  • handoff between personas
  • mixed initiative without excessive intrusion
  • shared memory when consented
  • explicit memory controls
  • transcript access
  • safety fallback behavior
  • escalation to human operator in admin contexts where relevant

13.4 Disclosure Requirements#

Every assistant, teacher, expert, guide, or guru mode must include:

  • clear AI disclosure
  • persona name and type
  • memory state indicator
  • grounding state indicator when evidence is used
  • voice/avatar synthetic indicator where applicable
  • user-accessible explanation of what is remembered and why

13.5 Persona Registry Requirements#

Each persona must have:

  • stable identifier
  • display name
  • family classification
  • use-case scope
  • domain affinities
  • allowed modalities
  • tone/style profile
  • disclosure rules
  • memory rules
  • grounding rules
  • safety rules
  • escalation rules
  • voice profile linkage
  • avatar pack linkage
  • entitlements/premium status
  • language coverage
  • evaluation scores
  • approval status

13.6 Memory Requirements#

Assistant memory must support:

  • explicit opt-in where required
  • scoped memory types
  • durable profile memory
  • session memory
  • notebook-linked memory
  • memory review and deletion by the user
  • policy-based memory suppression for sensitive categories
  • admin inspection only under governed conditions

13.7 Real-Time Embodiment Requirements#

For stylized avatar and real-time assistant experiences, the system must support:

  • animated avatars
  • stylized visual identity
  • low-latency voice conversation
  • turn-taking
  • interruption handling
  • emotional-state modulation within policy
  • screen context awareness where appropriate
  • synchronized disclosure overlays
  • synthetic media watermarking and provenance

13.8 Quality Bar#

The premium cast should feel unusually high-end, but the design standard is:

  • premium stylization, not deceptive realism
  • strong identity, not undisclosed human impersonation
  • coherence across text, voice, animation, and memory behavior
  • safe and reviewable behavior across all channels

14. Research, Authoring, and Agentic System Requirements#

14.1 Grounded Research Requirement#

OSHUN must support serious research, authoring, and evidence-heavy explanation workflows without fragmenting them across disconnected systems.

14.2 Sophia as the Research Backbone#

Sophia-backed capabilities must provide:

  • source ingestion
  • normalization
  • semantic search
  • source graphing
  • citation resolution
  • evidence-pack assembly
  • notebook management
  • comparative source mapping
  • grounded report generation
  • evaluation and scoring
  • trace export

14.3 Consumer Research Experiences#

Consumers must be able to:

  • inspect evidence for claims
  • open source trails from articles or assistant turns
  • compare passages and translations
  • save evidence to notebooks
  • follow concept threads
  • request grounded explainers
  • move from a claim to primary sources

14.4 Operator Research Experiences#

Operators must be able to:

  • review generated reports
  • inspect retrieval sets
  • inspect quote alignment
  • inspect citation correctness
  • compare competing evidence summaries
  • approve or reject high-impact grounded artifacts

14.5 Agentic Authoring Requirements#

Agentic authoring systems must support:

  • draft generation with explicit scope
  • evidence-assisted drafting
  • review checkpoints
  • source requirement policies
  • fact-check loops
  • revision lineage
  • approval gates before publication or broad distribution

14.6 Research Agent Guardrails#

Agentic research systems must:

  • declare when claims are unsupported
  • distinguish retrieval from synthesis
  • include traceable evidence objects
  • support human review
  • log model/provider/workflow lineage
  • avoid hidden in-memory-only state in quality-critical flows

14.7 Deep Study and Scholarship Requirements#

The product must support:

  • passage-level annotations
  • manuscript or edition comparison
  • translation comparison
  • lexicon and morphology support
  • influence-network exploration
  • timeline views
  • concept maps
  • curriculum building
  • study plans
  • exportable citations and notes

15. Image, Audio, Video, Avatar, and Model System Requirements#

15.1 Canonical Principle#

No media-generation capability may be production-critical without:

  • provenance
  • reviewability
  • policy coverage
  • rights awareness
  • release gating

15.2 Isis as the Control Plane#

All image, audio, video, avatar, and workflow execution must ultimately resolve through an Oshun-controlled control plane with:

  • workflow registry
  • model registry
  • provider registry
  • staging recipes
  • release gates
  • provenance bundles
  • cost and performance tracking
  • rollback
  • audit logging

15.3 ComfyUI Requirements#

ComfyUI must be treated as:

  • a workflow representation layer
  • an adapter layer
  • an execution substrate behind controlled templates

Product requirements:

  • versioned workflow templates
  • parameter guardrails
  • policy-aware input validation
  • environment-specific execution profiles
  • workflow portability across providers
  • no raw graph sprawl as the product contract

15.4 RunPod Requirements#

RunPod usage must support:

  • standardized endpoint registration
  • queue and health monitoring
  • region awareness
  • secret handling
  • provider failover strategy
  • concurrency and cost control
  • release gating before production routing

15.5 ElevenLabs and Voice Provider Requirements#

Voice systems must support:

  • provider abstraction
  • voice profile registry
  • consent capture
  • cloned-voice safeguards
  • watermarking
  • quality scoring
  • abuse scoring
  • disclosure and provenance signals
  • premium teacher and guide voices

15.6 Civitai Integration Requirements#

Civitai should be integrated deeply but carefully.

Required posture:

  • Civitai is a model-intelligence and asset-source layer
  • Civitai is not the canonical runtime of record
  • all imported model metadata must be normalized into Isis
  • all preview assets selected for use must be copied into Oshun-controlled storage when licensing permits
  • all imported models require review before becoming eligible for staging or production use
  • recommended parameters, creator metadata, licenses, hashes, preview images, content ratings, and dependency notes must be preserved
  • any model promoted for use must result in an Isis staging recipe for approved ComfyUI/RunPod execution

Required admin features:

  • Civitai model intake queue
  • creator and model card review
  • license and rights review
  • preview-asset approval
  • compatibility notes
  • recommended configuration capture
  • operational recipe generation
  • promotion or rejection decision logging

15.7 Avatar Requirements#

Avatar systems must support:

  • stylized character packs
  • pose and animation assets
  • expression systems
  • lip-sync and turn-taking alignment
  • persona-linked disclosure overlays
  • provenance and watermark metadata
  • quality review workflows
  • brand/style consistency

15.8 Media Review Requirements#

Every generated media artifact that can reach customers or operators must have:

  • source workflow identity
  • model identity and version
  • provider identity
  • source asset references
  • license/risk state
  • reviewer state
  • watermark or labeling state
  • audit trail

16. Platform Architecture and Canonical Ownership#

16.1 Canonical Platform Split#

The platform ownership model must be explicit and enforced.

Iris#

Owns:

  • assistant identity
  • persistent conversation history
  • durable memory
  • memory scopes
  • memory privacy boundary
  • memory consent state
  • user memory controls

Psyche#

Owns:

  • real-time session orchestration
  • live text/voice/avatar runtime
  • conferencing and turn-taking
  • multimodal embodiment
  • low-latency interaction state

Lilith#

Owns:

  • contemplative teacher logic
  • meditation-tone policy
  • spiritual-domain persona policy
  • voice quality and safety policies for relevant classes
  • specialized safety patterns for contemplative contexts

Sophia#

Owns:

  • grounding
  • retrieval
  • citations
  • evidence packs
  • knowledge graph
  • notebooks
  • research evaluation

Isis#

Owns:

  • model registry
  • workflow registry
  • generation execution policy
  • staging and promotion
  • provenance bundles
  • release gates
  • media workflow orchestration

OSHUN Shell#

Owns:

  • consumer UX
  • admin UX
  • routing and shells
  • domain adapters
  • shared object composition
  • entitlement-aware orchestration
  • continuity and recommendation presentation

16.2 Domain Adapter Requirement#

Every domain and subsystem must connect through explicit adapters rather than ad hoc UI-to-service coupling.

Adapters must exist for:

  • Tara
  • Arete
  • Veritas
  • Nyx
  • Nisaba
  • Sophia-backed evidence
  • Iris memory
  • Psyche embodiment
  • Lilith persona policies
  • Isis media operations

16.3 Contract Requirements#

The platform must define canonical contracts for:

  • domain metadata
  • cross-domain navigation
  • shared object schemas
  • review packages
  • evidence packs
  • provenance bundles
  • persona registry entries
  • avatar packs
  • voice profiles
  • memory scopes
  • consent records
  • workflow templates
  • model cards

16.4 Durability Requirement#

Any quality-critical runtime using only in-memory state today must be refactored to durable storage, durable queues, or durable event logs before it is considered production-complete.

This applies especially to:

  • research job state
  • authoring job state
  • generation job state
  • assistant session state that requires replay or audit
  • persona release state
  • review state

16.5 End-to-End Execution Model#

The end-state system should behave predictably across consumer, admin, and media-heavy flows.

Consumer Assistant Flow#

  1. The user invokes the assistant from a customer surface.
  2. The shell passes screen context, user context, and entitlement context.
  3. Iris resolves identity, durable memory, and allowed memory scopes.
  4. Psyche handles live multimodal orchestration when the interaction is real-time, voiced, or embodied.
  5. Lilith applies persona, contemplative tone, and domain-safety policy where relevant.
  6. Sophia is invoked when evidence, citations, or source-aware synthesis are required.
  7. The shell renders the response with disclosure, memory state, and grounding state.
  8. Audit, trace, and quality events are emitted.

Grounded Research Flow#

  1. A claim, question, or draft requests evidence-backed support.
  2. Sophia assembles retrieval, source graph, evidence packs, citations, and notebook context.
  3. The product renders grounded outputs with traceable evidence links.
  4. High-impact outputs enter review workflows before broad publication or promotion.

Media Generation Flow#

  1. A domain or operator requests media generation or model-backed inference.
  2. Isis resolves the approved workflow template, model version, provider endpoint, and policy bundle.
  3. Approved ComfyUI workflows and RunPod or equivalent execution providers are selected according to environment and policy.
  4. Provenance bundles, watermark states, and release metadata are attached.
  5. If the artifact is sensitive or user-facing, it enters the review gate before promotion or publication.

Civitai Intake Flow#

  1. Operators or automated agents discover a Civitai asset or model.
  2. Metadata, hashes, licenses, preview assets, and recommended configurations are normalized into Isis-owned records.
  3. Rights, safety, compatibility, and quality reviews occur.
  4. Approved models are translated into staging recipes for Oshun-controlled execution, not used as unmanaged external runtime dependencies.

17. Review, Safety, Compliance, and Governance Requirements#

17.1 Canonical Review Object Model#

The accepted review/approval and audit/compliance ADRs must become the platform-wide standard, not a studio-only pattern.

Required objects:

  • ReviewPackage
  • ReviewStage
  • Decision
  • Escalation
  • EvidenceSet
  • PolicyBundle
  • Attestation
  • Exception
  • AuditEvent

17.2 Safety Requirements#

The product must support:

  • content policy enforcement
  • crisis-aware flows
  • voice abuse detection
  • image abuse detection
  • deceptive-avatar safeguards
  • teacher-tone safety
  • support for protected or sensitive categories
  • alerting and escalation
  • appeal and remediation workflows

17.3 Compliance Requirements#

The product must support:

  • retention policies
  • region-aware data handling
  • exportability of audit evidence
  • DSAR workflows
  • consent traceability
  • model and asset rights review
  • synthetic media labeling
  • policy attestation

17.4 Rights and Licensing Requirements#

The system must capture:

  • license metadata for models
  • license metadata for assets
  • creator/source metadata where required
  • downstream-use restrictions
  • consent state for voice or likeness-derived systems
  • provenance lineage for outputs

17.5 Release Gate Requirements#

No high-impact artifact may be promoted without the required gates.

Gate classes include:

  • content gates
  • evidence gates
  • persona gates
  • voice/avatar gates
  • model/workflow gates
  • media output gates
  • compliance gates

17.6 Audit Requirements#

All governed actions must produce immutable audit events with:

  • actor
  • action
  • reason
  • artifact
  • outcome
  • trace identifiers
  • policy scope
  • timestamps

17.7 Evaluation Requirements#

The platform must include formal evaluation programs for the highest-risk and highest-value systems.

Evaluation suites must exist for:

  • grounded-answer correctness
  • citation integrity
  • research report hallucination risk
  • contemplative teacher safety and tone
  • support copilot correctness
  • moderation assistant correctness
  • voice quality and abuse resistance
  • avatar coherence and disclosure visibility
  • model/workflow output quality
  • watermark detection and verification

Evaluation requirements:

  • benchmark suites must be versioned
  • production promotions must reference current evaluation results
  • regressions must block release where severity thresholds are exceeded
  • operator review must remain available for ambiguous or novel cases

17.8 Human-in-the-Loop Requirements#

The platform must clearly distinguish between:

  • AI-assisted review
  • AI-suggested action
  • human-approved action
  • fully automated policy enforcement

Human approval must remain mandatory for:

  • high-impact persona releases
  • production model promotions
  • policy exceptions
  • legal or privacy-sensitive deletions and exports where policy requires it
  • major content launches
  • ambiguous trust and safety escalations

18. Data, Memory, and Personalization Requirements#

18.1 One Profile, Many Contexts#

Users must have one unified profile with context-sensitive expressions across domains and assistants.

18.2 Personalization Inputs#

Personalization may incorporate:

  • stated goals
  • preferred teachers or guides
  • schedule and rhythm
  • reading and practice history
  • saved sources and passages
  • current focus themes
  • notification preferences
  • accessibility preferences

18.3 Memory Governance#

The product must expose:

  • what memory is active
  • what categories are remembered
  • when memory was last updated
  • how to clear or edit memory
  • how memory affects responses and recommendations

18.4 Sensitive Memory Controls#

Sensitive memory categories must support:

  • explicit opt-in
  • clear separation from casual memory
  • stricter admin access controls
  • audit trails for access
  • policy-driven retention

18.5 Shared Continuity Requirements#

Continuity must span:

  • web to mobile
  • customer to customer assistant
  • customer to admin support context where appropriate and allowed
  • domain to domain transitions

18.6 Data Portability and Recovery Requirements#

The system must support:

  • export of user notes, highlights, and notebooks
  • export of saved sources and passages where rights permit
  • export of conversation history according to policy
  • recovery of accidental loss through governed restore flows
  • admin tooling for user-visible data correction when policy allows

19. Search, Discovery, Recommendation, and Knowledge Graph Requirements#

Search must unify:

  • content
  • passages
  • claims
  • sources
  • concepts
  • notebooks
  • assistants
  • personas where user-facing
  • sky events
  • programs and rituals

19.2 Recommendation#

Recommendations must support:

  • cross-domain continuation
  • time-aware prompting
  • ritual suggestions
  • study suggestions
  • grounded story/context suggestions
  • sky-event suggestions
  • persona suggestions where appropriate

Every recommendation should have a reason model available internally and a human-readable reason label where user value warrants it.

19.3 Shared Concept Graph#

The product must maintain a shared concept graph connecting:

  • passages
  • sources
  • claims
  • topics
  • events
  • habits and programs
  • journals and notebooks
  • sky phenomena
  • assistant turns

The concept graph is crucial for:

  • Nisaba depth
  • Veritas provenance
  • Tara and Arete meaning-aware recommendations
  • Nyx comparative cosmology and symbolism
  • assistant continuity

20. Monetization, Entitlements, and Commerce Requirements#

20.1 Entitlement Principles#

Premium should unlock more depth, continuity, guidance, and creation power without undermining trust or core usability.

20.2 Customer Entitlement Areas#

Potential entitlement classes:

  • premium meditations and programs
  • advanced coaching and planning
  • advanced grounded explainers
  • advanced Nisaba workspaces
  • premium voice and avatar assistants
  • premium study plans
  • premium sky tools
  • advanced notebooks and exports
  • media-generation features where applicable

20.3 Admin Requirements#

Operators must be able to:

  • inspect subscriptions
  • inspect entitlements
  • grant or revoke exceptions
  • handle refunds or credits
  • inspect experiment and campaign effects
  • trace entitlement-linked issues

21. Analytics, Experimentation, and Quality Requirements#

21.1 Analytics Requirements#

The platform must measure:

  • activation
  • retention
  • cross-domain usage
  • ritual completion
  • study depth
  • evidence engagement
  • assistant usage by mode
  • memory usage and opt-in rates
  • queue SLAs
  • moderation outcomes
  • support outcomes
  • release gate outcomes
  • media workflow quality

21.2 Experimentation Requirements#

The platform must support:

  • feature flags
  • experiment cohorting
  • policy-safe experimentation
  • outcome dashboards
  • rollback support

21.3 Quality Requirements#

Quality must be measured through:

  • product performance
  • grounded-answer correctness
  • citation integrity
  • moderation precision and recall
  • voice and audio quality
  • avatar quality
  • user trust signals
  • support satisfaction
  • review consistency

22. Non-Functional Requirements#

22.1 Performance#

Consumer surfaces must feel fast enough to preserve calm and flow.

Required standards:

  • fast cold starts on mobile
  • responsive shell transitions
  • low-latency assistant streaming
  • p95 evidence fetch latency within acceptable reading UX thresholds
  • queue responsiveness for admin workflows

22.2 Reliability#

  • partial domain outages must degrade gracefully
  • provider failures must not collapse the shell
  • generation-provider failures must fail with policy-aware fallback
  • review and audit events must not be silently dropped

22.3 Security#

  • RBAC and least privilege
  • step-up auth for sensitive actions
  • strong secrets handling
  • auditability for privileged actions
  • tenant and environment separation

22.4 Accessibility#

  • WCAG AA minimum for web
  • screen-reader coverage
  • keyboard access
  • dynamic type support
  • captioning and transcript support
  • motion sensitivity controls

22.5 Internationalization and Localization#

  • domain content localization support
  • persona language coverage tracking
  • translation workflows
  • locale-aware calendars and timings where relevant

22.6 Privacy#

  • explicit memory controls
  • clear consent surfaces
  • export and deletion support
  • policy-aware data minimization

23. Operational Readiness Requirements#

The product is not complete until it is operable at scale.

Required operational readiness elements:

  • dashboards for customer and admin surfaces
  • subsystem health dashboards
  • queue health dashboards
  • runbooks
  • escalation policy
  • release checklists
  • rollback pathways
  • provenance inspection tools
  • audit export tools
  • on-call workflows
  • provider failover policies
  • cost monitoring

23.1 Release Readiness Requirements#

No significant launch or promotion should occur without:

  • product QA signoff
  • safety signoff
  • evidence/citation signoff where applicable
  • rights and provenance signoff where applicable
  • rollback validation
  • monitoring and alert coverage
  • operator training and runbook readiness
  • support readiness

24. Success Metrics#

The product must be evaluated as one integrated system.

24.1 Consumer Metrics#

  • D1, D7, D30 retention
  • weekly active use across more than one domain
  • percentage of users with meaningful Tara continuity
  • cross-domain attachment rate
  • Nisaba adoption and repeat depth
  • evidence engagement rate in Veritas and assistant flows
  • assistant satisfaction
  • memory opt-in with sustained retention benefit

24.2 Trust and Quality Metrics#

  • share of evidence-sensitive responses backed by Sophia
  • citation integrity pass rate
  • moderation SLA attainment
  • appeal overturn rate
  • synthetic media provenance coverage
  • watermark verification coverage
  • percentage of premium personas with complete approval bundles

24.3 Platform Metrics#

  • share of assistant sessions routed through canonical Iris/Psyche contracts
  • share of grounded outputs produced through canonical Sophia contracts
  • share of generation workflows routed through canonical Isis contracts
  • Civitai import success rate into Isis model registry
  • percentage of externally sourced models with completed rights review
  • provider failover success rate

24.4 Admin Metrics#

  • review queue SLA
  • support first-response and resolution SLAs
  • incident acknowledgement and mitigation time
  • audit export success rate
  • policy exception closure time

25. Phased Delivery Model#

The system should be delivered in phases, but the end-state architecture must be designed from the start.

Phase 0: Alignment#

  • finalize product role model
  • confirm Tara-centered philosophy
  • confirm Nisaba and Sophia roles
  • confirm canonical subsystem ownership
  • confirm review and provenance contract model

Phase 1: Shared Contract Expansion#

  • add Nisaba across shell contracts
  • define shared object schemas
  • define persona registry contracts
  • define provenance and review contracts

Phase 2: Consumer Surface Expansion#

  • launch Nisaba web and mobile companion surfaces
  • add grounded evidence views across Veritas and assistant experiences
  • improve Tara-centered home flows

Phase 3: Admin Product Unification#

  • ship unified admin web shell
  • ship admin mobile urgent workflows
  • unify queues, cases, and evidence bundles

Phase 4: Assistant and Persona Platform#

  • wire Iris memory
  • wire Psyche real-time sessions
  • wire Lilith persona policy layer
  • launch premium assistant families with explicit disclosure

Phase 5: Media and Model Hardening#

  • centralize model registry in Isis
  • normalize Civitai intake
  • centralize workflow execution via approved ComfyUI/RunPod templates
  • harden watermark, provenance, and rights workflows

Phase 6: Deep Integration and Optimization#

  • deepen shared concept graph
  • improve recommendation quality
  • refine admin copilot workflows
  • improve lifecycle, experimentation, and monetization

26. Anti-Patterns to Avoid#

  • adding Nisaba as just another icon
  • treating Sophia as optional
  • building premium assistants without explicit memory and disclosure controls
  • letting persona logic sprawl across multiple systems without a registry
  • letting Civitai become an uncontrolled runtime dependency
  • exposing raw ComfyUI graphs as the product contract
  • relying on in-memory state in quality-critical services
  • shipping admin tools without unified review and audit models
  • optimizing for superficial realism over trustworthy stylized identity
  • allowing unsupported claims to appear grounded

27. Acceptance Criteria for the End-State V1#

OSHUN V1 is complete only when all of the following are true.

27.1 Product Cohesion#

  • the customer experiences one product across web and mobile
  • Tara is visibly the product’s thematic center
  • Arete, Veritas, Nyx, and Nisaba are all meaningfully integrated
  • Sophia powers grounded experiences where evidence is required

27.2 Consumer Feature Completeness#

  • web and mobile each have strong, context-appropriate domain depth
  • Nisaba is fully exposed and cross-wired
  • library, activity, search, recommendations, and assistant systems are shared
  • notebooks, passages, claims, and sources connect across domains

27.3 Admin Completeness#

  • admin web is the canonical operator cockpit
  • admin mobile supports urgent approval and escalation workflows
  • review, moderation, support, privacy, persona ops, model ops, and incidents are unified under one shell and one shared evidence model

27.4 Assistant Completeness#

  • assistants operate through canonical persona, memory, grounding, and embodiment contracts
  • explicit disclosure, consent, and memory controls are visible
  • voice and avatar systems carry provenance and safety coverage

27.5 Platform Completeness#

  • Sophia is the canonical grounding substrate
  • Iris is the canonical memory substrate
  • Psyche is the canonical real-time embodiment substrate
  • Lilith is the canonical contemplative persona policy layer
  • Isis is the canonical generation control plane
  • Civitai is normalized into Isis rather than treated as runtime of record

27.6 Governance Completeness#

  • every high-impact artifact has a review path
  • every governed action has immutable audit evidence
  • every promoted media or persona asset has provenance and rights review
  • every privacy-sensitive workflow has consent and retention controls

27.7 Operational Completeness#

  • dashboards, alerts, runbooks, incident flows, and queue SLAs are in place
  • failure modes degrade gracefully
  • product and admin teams can operate the system responsibly at scale

Closing Position#

The fully built OSHUN product should not be understood as "a meditation app with extra tabs" or "an admin dashboard plus some consumer apps." It is one unified contemplative intelligence platform with customer, operator, and subsystem layers designed as one whole.

Its defining characteristics are:

  • Tara-centered coherence
  • Nisaba-enabled depth
  • Sophia-enabled trust
  • Iris/Psyche/Lilith-enabled premium assistants
  • Isis-governed media and model operations
  • operator-grade review, provenance, and compliance

That is the standard this PRD sets for the final integrated V1.