13 — Creator Relationship & Collaboration CRM

Document maturity: implementation in progress; local CRM-101 foundation is verified and promotion gates remain open. Canonical feature status: CRM-001 in the feature register. Roadmap: Z5; M4–M5 after the conversational pilot has stable outcomes. Decision: docs/adr/ADR-013-creator-relationship-crm.md. Depends on: proposal (08) interaction memory, (10) demand clusters, (12) policy/reputation. Boundary: public professional relationships only; no vulnerability or psychographic profiling.

1. Problem

Reply-first growth is not only a volume problem. Repeated, useful exchanges with a small set of relevant creators can produce community trust, collaborations, referrals and durable discovery.

SPA currently stores interactions, but it does not answer:

Who has responded repeatedly and positively?
Which creators share the same subject areas without being direct duplicates?
Where is there a natural collaboration opportunity?
When did we last interact, and are we over-targeting one person?
Which persona/account should maintain the relationship?
What is the next useful human action — reply, share, invite, wait, or skip?

Without a relationship layer, engagement memory remains a log and can drift into repetitive commenting. The CRM turns public interaction history into a human-controlled collaboration workflow.

2. Decision summary

Create a lightweight, public-data Creator Relationship Graph with:

one canonical creator identity per network handle;
topic/domain affinity and explicit evidence;
interaction edges to SPA accounts/personas;
a conservative relationship stage;
collaboration-opportunity proposals;
per-creator cooldown/frequency safeguards;
manual notes and decisions;
no automated DMs/outreach or sensitive-person inference.

The system recommends a next relationship action but never executes outreach automatically.

public interaction outcomes
→ canonical creator identity
→ relationship/account-persona edge
→ reciprocity + relevance + continuity evidence
→ opportunity proposal
→ human review
→ manual collaboration workflow

3. Goals and non-goals

Goals

Build long-term public relationships instead of maximizing comment count.
Prevent excessive/repetitive targeting of one creator.
Show conversation history and shared topics before suggesting an action.
Identify credible co-created content, newsletter/podcast/interview or mutual-promotion opportunities.
Attribute collaboration links/campaigns through existing Z4 infrastructure.
Keep personas consistent: one primary account/persona owns a relationship unless reassigned.
Measure reciprocity and collaboration outcomes without ranking human worth.

Non-goals

Sales CRM for individual Soulwise users.
Automated cold DMs, pitches or follow/unfollow campaigns.
Buying/scraping email addresses or private contact data.
Inferring health, fertility, relationship distress, wealth, ethnicity, sexuality or mental state.
“Influence scores” based only on follower count.
Automatically declaring someone a partner/competitor.
Mirroring full social profiles or content archives.
Multi-platform identity resolution without explicit evidence/human confirmation.

4. Relationship model

4.1 Stages

DISCOVERED
→ OBSERVED
→ ENGAGED
→ RECIPROCAL
→ COLLABORATION_CANDIDATE
→ ACTIVE_COLLABORATOR
→ DORMANT
→ DO_NOT_ENGAGE

Stage advancement is evidence-backed and usually human-approved. A like alone cannot create RECIPROCAL; follower count cannot create COLLABORATION_CANDIDATE.

4.2 Evidence types

replied to SPA interaction;
repeated substantive exchange;
shared/quoted SPA content;
followed account where reliably observed;
participated in the same public discussion;
manually identified shared editorial interest;
accepted/declined collaboration proposal;
collaboration delivered and measured;
operator note with source/reason.

4.3 Next actions

READ_RECENT_WORK
REPLY_IF_VALUE
SHARE_WITH_COMMENTARY
PROPOSE_COLLABORATION
FOLLOW_UP_MANUALLY
WAIT_COOLDOWN
DO_NOT_ENGAGE

Every recommendation explains evidence, contraindications and cooldown.

5. Architecture

flowchart LR
    Outcomes[Public Interaction Outcomes] --> Canon[Creator Identity Resolver]
    Manual[Operator Curation] --> Canon
    Canon --> Creator[(CreatorProfile)]
    Outcomes --> Edge[Relationship Edge Updater]
    Creator --> Edge
    Edge --> Relationship[(CreatorRelationship)]
    Demand[Demand / Topic Clusters] --> Match[Collaboration Opportunity Matcher]
    Relationship --> Match
    Portfolio[Editorial Portfolio Planner] --> Match
    Policy[Policy + Reputation State] --> Match
    Match --> Proposal[(CollaborationOpportunity)]
    Proposal --> Review[Human Review / Manual Outreach]
    Review --> Campaign[Collaboration Campaign + Z4 Link]
    Campaign --> Outcome[Outcome + Learning]
    Outcome --> Relationship

5.1 Components

Component	Responsibility
`CreatorIdentityService`	Canonical network identity and manual cross-network links.
`CreatorRelationshipService`	Evidence, stages, owner persona/account, cooldown.
`CreatorAffinityService`	Shared public topics/domains with source trace.
`CollaborationOpportunityService`	Propose collaboration type/fit/risks.
`RelationshipNextActionService`	Explainable recommendation; no execution.
`CollaborationCampaignService`	Manual workflow, deliverables and Z4 attribution refs.
`CreatorPrivacyService`	Retention, purge, opt-out/do-not-engage and field minimization.

5.2 Ports

export const ICreatorRelationshipPort = Symbol("ICreatorRelationshipPort");
export const ICollaborationOpportunityPort = Symbol(
  "ICollaborationOpportunityPort",
);

Engagement and Portfolio Planner may read relationship summaries but cannot mutate stage directly.

6. Data model

model CreatorProfile {
  id                   String   @id @default(uuid())
  network              SocialNetwork
  handleCanonical      String
  handleHash           String
  displayName          String?
  profileUrl           String
  disclosure           String?
  publicTopics         Json
  accountType          String?
  status               String   // ACTIVE | DORMANT | DO_NOT_ENGAGE | PURGED
  sourceRefs           Json
  lastVerifiedAt       DateTime?
  rawProfileExpiresAt  DateTime?
  createdAt            DateTime @default(now())
  updatedAt            DateTime @updatedAt

  relationships CreatorRelationship[]

  @@unique([network, handleCanonical])
  @@index([network, status])
  @@index([handleHash])
}

model CreatorRelationship {
  id                   String   @id @default(uuid())
  creatorId            String
  accountId            String
  personaRevisionId    String?
  stage                String
  stageEvidence        Json
  sharedDomains        Json
  interactionCount     Int      @default(0)
  substantiveReplyCount Int     @default(0)
  reciprocalCount      Int      @default(0)
  lastInteractionAt    DateTime?
  cooldownUntil        DateTime?
  ownerNote            String?
  manualPriority       String?
  status               String
  createdAt            DateTime @default(now())
  updatedAt            DateTime @updatedAt

  creator CreatorProfile @relation(fields: [creatorId], references: [id], onDelete: Cascade)
  opportunities CollaborationOpportunity[]

  @@unique([creatorId, accountId])
  @@index([accountId, stage, status])
  @@index([personaRevisionId, lastInteractionAt])
}

model CreatorInteractionEvidence {
  id              String   @id @default(uuid())
  relationshipId  String
  interactionId   String?
  evidenceType    String
  sourceRef       Json
  evidenceHash    String
  weight          Float?
  occurredAt      DateTime
  expiresAt       DateTime?
  createdAt       DateTime @default(now())

  @@unique([relationshipId, evidenceType, evidenceHash])
  @@index([relationshipId, occurredAt])
}

model CollaborationOpportunity {
  id                String   @id @default(uuid())
  relationshipId    String
  opportunityType   String   // CO_POST | INTERVIEW | NEWSLETTER_SWAP | LIVE | RESEARCH | OTHER
  topic             String
  rationale         Json
  risks             Json
  proposedAccountId String
  proposedPersonaId String?
  status            String   // PROPOSED | REVIEWED | OUTREACH_PLANNED | ACCEPTED | DECLINED | COMPLETED | ARCHIVED
  reviewedBy        String?
  outreachRef       Json?
  campaignRef       Json?
  validUntil        DateTime?
  createdAt         DateTime @default(now())
  updatedAt         DateTime @updatedAt

  relationship CreatorRelationship @relation(fields: [relationshipId], references: [id], onDelete: Cascade)

  @@index([status, validUntil])
  @@index([proposedAccountId, createdAt])
}

Clear handles are operationally necessary for a public relationship workflow and remain personal data. They are access-controlled, retained minimally and purgeable; hashing provides lookup/dedup, not anonymity.

7. Identity and cross-network linking

Canonical identity is network-scoped by default.
Cross-network links require explicit public evidence and human confirmation.
Similar display name/avatar/topic is insufficient.
Never enrich with private data brokers.
Record link evidence/version and support unlinking.
A creator may request opt-out/do-not-engage; enforce across linked identities.

8. Relationship scoring

Do not expose a single opaque “creator score”. Internally store explainable components:

topic relevance;
reciprocal interaction evidence;
interaction quality;
relationship continuity;
audience/content complementarity;
persona fit;
collaboration readiness;
cooldown/over-targeting risk;
policy/reputation risk;
data freshness/confidence.

Follower count and raw reach are optional context, never dominant. A high-value niche creator may be more relevant than a large generic account.

Stage transitions use explicit rules and human review. Model classification may suggest shared topics/opportunity, but cannot set ACTIVE_COLLABORATOR or remove DO_NOT_ENGAGE.

9. Collaboration workflow

opportunity proposed
→ operator reviews creator history, rationale and risk
→ operator chooses manual contact channel
→ outreach planned externally or recorded manually
→ accepted/declined/no-response
→ deliverables and campaign link created
→ collaboration completed
→ direct/assisted outcomes measured
→ human learning note

The system never drafts manipulative personalization from inferred vulnerabilities. Suggested outreach uses public professional context and requires human editing/sending.

Potential collaboration formats:

paired Threads discussion;
expert Q&A/interview;
newsletter/podcast swap;
co-authored evidence-backed myth-buster;
public live conversation;
shared compatibility/cycle education resource;
research or survey summary with clear methodology.

10. Integration

Proposal `(08)`

relationship summary enters AuthorContext only when relevant;
per-creator cooldown and prior stance prevent repetitive comments;
persona owner controls voice continuity.

Proposal `(10)`

shared demand clusters identify collaboration topics;
creator contributions never automatically validate cluster truth.

Proposal `(11)`

collaboration campaign gets dedicated Z4 link/campaign;
direct/assisted outcomes stay separate.

Proposal `(12)`

DO_NOT_ENGAGE, policy and reputation state override recommendations;
reputation incidents can suspend collaboration proposals.

Portfolio Planner

collaboration may reserve account/topic capacity;
planner avoids duplicate own posts around the collaboration.

11. API and UI

API

GET /creators?network=&stage=&domain=&cursor=
GET /creators/:id
POST /creators/manual
POST /creators/:id/verify
POST /creators/:id/link-identity
POST /creators/:id/unlink-identity
POST /creators/:id/do-not-engage
DELETE /creators/:id
GET /creator-relationships?accountId=&stage=&cursor=
POST /creator-relationships/:id/transition
POST /creator-relationships/:id/note
GET /collaboration-opportunities
POST /collaboration-opportunities/:id/review
POST /collaboration-opportunities/:id/record-outreach
POST /collaboration-opportunities/:id/record-outcome

UI

creator list with stage/domain/freshness;
evidence-backed relationship timeline;
account/persona owner;
cooldown/over-targeting warning;
shared demand/topics;
opportunity rationale and risks;
manual outreach/deliverable/campaign tracking;
do-not-engage, purge and identity-link controls;
collaboration outcome analytics.

12. Privacy, safety, and abuse prevention

Public professional data only and purpose limitation.
No sensitive-trait inference or health/relationship profiling.
No private contact enrichment or scraped email/phone.
No automated DMs/outreach/follow campaigns.
Rate/cooldown safeguards prevent harassment-like repeated targeting.
Human review for stage transitions and opportunities.
Clear handle and operator notes are admin-only and excluded from Langfuse/Sentry.
Retention by field/evidence type; expired source evidence reduces confidence.
Purge removes identity, relationships, embeddings and recommendations.
DO_NOT_ENGAGE is enforced before candidate scoring and survives normal archival.
Collaboration involving medical/cycle claims requires the same evidence/safety gates as content.

13. Reliability and degradation

Failure	Behavior
Identity resolution ambiguous	Keep network identities separate; human review.
Relationship evidence missing	No stage promotion or next action.
Affinity model unavailable	Public tags/manual topics only.
Policy/reputation unavailable	No outreach recommendation.
Metrics unavailable	Preserve relationship; no performance conclusion.
Creator source deleted	Expire evidence, reduce confidence, preserve minimal audit as policy allows.
Duplicate interaction events	Idempotent evidence hash.
CRM unavailable	Engagement can operate without relationship recommendation, still honoring cooldown cache/DO_NOT_ENGAGE.

Stage changes use optimistic concurrency. Reprocessing evidence is versioned and cannot silently rewrite manual decisions.

14. Observability and success metrics

creator_profiles_total{network,status}
creator_relationships_total{stage,account}
creator_evidence_ingested_total{type}
creator_identity_ambiguous_total
creator_cooldown_block_total
creator_over_targeting_block_total
creator_stage_transition_total{from,to}
collaboration_opportunity_total{type,status}
collaboration_acceptance_rate
collaboration_completion_rate
creator_repeat_substantive_reply_rate
creator_relationship_to_collaboration_time
collaboration_direct_leads/revenue
collaboration_assisted_leads/revenue
creator_purge_total

Success is repeat substantive relationships and completed collaborations, not the number of creator records or automated touches.

15. Evaluation

identity resolution precision and ambiguous-case rate;
relationship-stage agreement with human labels;
next-action relevance/contraindication recall;
over-targeting/cooldown enforcement;
shared-topic precision;
opportunity acceptance and completion;
privacy/sensitive-inference red-team;
no-action correctness;
collaboration outcome reporting quality.

Model/rule/profile changes run proposal (09). Human relationship decisions are not treated as ground truth for a universal creator ranking.

16. Rollout and backlog

The CR-* rows are local design work-package anchors, not canonical task/status IDs. Promotion to implementation must map them into docs/planning/BACKLOG.md without duplicating status.

ID	Phase	Task	Depends on
CR-001	M4	Define public-data/identity/stage/evidence policy	proposals 08/12
CR-002	M4	CreatorProfile/Relationship/Evidence schema	CR-001
CR-003	M4	Canonical network identity and manual linking	CR-002
CR-004	M4	Outcome → relationship evidence ingestion	CR-002, proposal 08
CR-005	M4	Cooldown/over-targeting/do-not-engage gates	CR-002
CR-006	M4	Relationship timeline/list/detail UI	CR-003, CR-004
CR-007	M4	Explainable affinity/next-action proposal	CR-004, proposal 09
CR-008	M4	CollaborationOpportunity workflow	CR-006, CR-007
CR-009	M5	Z4 collaboration campaign attribution	CR-008, proposal 11
CR-010	M5	Collaboration analytics and learning review	CR-009

Gate M4–M5

no private enrichment or sensitive-trait fields exist;
cooldown/DO_NOT_ENGAGE precedes engagement recommendation;
stage/opportunity changes are evidence-backed and human-reviewed;
no automated outreach path exists;
purge propagates through relationships, evidence and recommendations.

17. Risks

Risk	Mitigation
CRM becomes surveillance	Minimal public professional data, purpose/retention/purge.
Automated relationship spam	No automated outreach; cooldown and human send.
Follower-count bias	Explainable multi-component evidence; reach is secondary.
Wrong cross-network identity link	Human confirmation and unlink/audit.
Persona conflict over creator	One owner assignment and explicit reassignment.
Relationship stage inflation	Evidence rules and review.
Sensitive inference from content	Forbidden schema/taxonomy/red-team.
Creator requests no contact	Durable DO_NOT_ENGAGE and purge workflow.
Collaboration measured as reply volume	Completion/direct/assisted outcomes, not touch count.

18. Research and verification status

Internal evidence reviewed:

existing Interaction and proposal (08) interaction-memory design;
Demand Radar and Assisted Attribution boundaries;
privacy/data-governance references;
ROADMAP_V2 collaboration/lead goals.

Implementation-time external research:

current platform terms for retention/use of public profile data;
approved creator/collaboration workflows for Threads/X;
privacy/legal review of do-not-engage and deletion handling;
interviews with the operator on actual collaboration workflow.

Exa MCP remained unavailable due OAuth. This proposal intentionally avoids claims about current third-party creator-CRM products or automated outreach conversion rates.

FilesExpand file tree

13-creator-relationship-crm.md

Latest commit

History