Status: Accepted — human-controlled foundation
Date: 2026-08-22
Decider: Valentyn Yakovliev
Feature: CRM-001 in docs/planning/FEATURES.md
Extends: ADR-008, ADR-010, ADR-012
Roadmap: proposal (13); milestone integration is unblocked by PLAN-002
Acceptance record: Product owner approved on 2026-08-23. v1 is public-data-only, network-scoped, DO_NOT_ENGAGE-first, recommendation-only and human-controlled for identity linking, meaningful relationship stages and outreach. No DMs, cold outreach or private enrichment.
SPA stores public interactions but has no durable model for repeated professional relationships, creator ownership by account/persona, cooldown, collaboration proposals or outcomes. Optimizing only reply count risks repetitive targeting and misses higher-value collaborations.
CreatorProfile stores minimal public professional fields, sources, status and freshness. Clear
handles are personal data and remain protected/purgeable. Cross-network identity links require
explicit public evidence and human confirmation.
Use DISCOVERED, OBSERVED, ENGAGED, RECIPROCAL, COLLABORATION_CANDIDATE,
ACTIVE_COLLABORATOR, DORMANT, DO_NOT_ENGAGE. Automated evidence may propose a transition;
meaningful stages and DO_NOT_ENGAGE changes require human review.
Next actions and collaboration opportunities are suggestions. No automated DMs, cold pitches, follow/unfollow campaigns or external contact enrichment are added.
Do not infer health, fertility, relationship distress, mental state, wealth or other vulnerability. Store shared public topics and interaction evidence only.
A relationship edge is account-specific with optional persona revision. Portfolio Planner respects ownership, cooldown and collaboration reservations to prevent multiple SPA accounts targeting the same creator inconsistently.
Measure repeated substantive exchanges, collaboration acceptance/completion and direct/assisted campaign outcomes. Follower count/touch volume are secondary context.
- Durable relationships produce more strategic value than indiscriminate comments.
- Network-scoped identity avoids unsafe automatic entity resolution.
- Human outreach preserves context and platform expectations.
- Explicit cooldown/DO_NOT_ENGAGE prevents harassment-like repetition.
- Existing Interaction, persona memory, demand clusters and Z4 attribution are reusable.
- Operator sees history, shared topics, ownership and next-action rationale.
- Collaboration workflow and outcomes become measurable.
- Repetitive targeting is constrained.
- Privacy boundary is narrower than a conventional enriched CRM.
- Manual review/outreach limits autonomous scale by design.
- Clear public handles require access control, retention and purge.
- Cross-network identity and relationship stages can be ambiguous.
- Collaboration outcomes are sparse and slow.
Rejected due poor relevance, spam incentives and relationship harm.
Rejected as out of scope and privacy-incompatible.
Rejected because logs do not express ownership, stage, cooldown, collaboration or DO_NOT_ENGAGE.
Rejected. The initial system is recommendation/manual execution only.
Rejected. Require evidence and human confirmation.
- Public professional data and purpose limitation only.
- DO_NOT_ENGAGE precedes all candidate recommendations.
- No raw profile/content archive beyond retention need.
- Clear identity/manual notes excluded from Langfuse/Sentry.
- Stage/action models pass ADR-009.
- Proposal
(13)owns schema, workflow, privacy, analytics and rollout.
- privacy/retention/do-not-engage policy;
- identity ambiguity and unlink tests;
- cooldown/over-targeting gates;
- human-only outreach proof;
- stage/next-action labelled evaluation;
- purge across relationships/evidence/recommendations;
- collaboration attribution and operator workflow review.
docs/roadmap/13-creator-relationship-crm.mddocs/roadmap/08-editorial-personas-conversational-engagement.mddocs/roadmap/10-conversation-intelligence-demand-radar.mddocs/roadmap/11-reply-to-revenue-assisted-attribution.mddocs/roadmap/12-platform-policy-reputation-control-plane.md