The Relationship Engine

Built for everything after the introduction.

Matching can decide who should meet. The Relationship Engine helps a Circle move from first interaction towards independent contact, deciding when to guide, when to do nothing, and when to step away.

Explore the nine stages Explore a pilot

Rule-based today · Explainable · Versioned · Learned models only after evidence

Illustrative Circle · real versioned rulesStage: Observe

signals-v1.2

participation: 4 of 6
reciprocity: low
independence: 0.18

decision

Observe

Reactions are recorded but do not count as participation.

Guided
Bridge
Tapering
Handoff

The loop

Nine stages. One closed loop.

Understand → Form → Activate → Observe → Interpret → Intervene → Handoff → Measure → Learn. Step through one illustrative Circle.

01 / 09

Understand

Who are these people, and what should change for them?

Takes the context, objective and each person's profile: intent, interests, availability. Consent is captured separately for taking part, research and follow-ups.

Objective set for this population
Engine versions pinned to the cohort

Illustrative Circle with generic initials. The versioned rules and thresholds shown are the real rules the system runs today; they are hypotheses being tested, not proven findings.

Architecture

Five systems, each versioned and frozen per cohort.

A cohort pins the exact versions it runs on. Changing the logic means registering a new version, never editing one in use.

01formation-v1
Formation
Forms small Circles from context and constraints. Keep-apart rules are hard limits, enforced twice.
- Context fields
- Mix and common ground
- Keep-apart constraints
02signals-v1.2
Signals and graph
Turns Circle activity into explainable signals per member and pair. Reactions count as weak acknowledgement only.
- Participation
- Reciprocity and breadth
- Independence
- Fading vs peripheral
03policy-v1.4
Intervention
Ordered rules choose one action, or deliberately none. Do nothing is a first-class decision. Every decision logs inputs, reason and alternatives.
- Bridge
- Do nothing
- Suggest handoff
- Facilitator review
04handoff-v1.1
Handoff
Readiness must hold on observed behaviour, then an observation period. An open safety case always blocks.
- 72h in Tapering
- 48h in Ready
- Human confirmation
05frozen per cohort
Outcomes and learning
Follow-up at 7, 30 and 90 days links back to the exact versions each Circle ran on.
- Tri-state answers
- Aggregate reporting
- Versioned comparisons

Thresholds and weights are documented hypotheses pending real-human evidence.

The software is the mechanism. The long-term asset we are building is the learning about relationship formation.

Every decision is linked to the version that made it and to what happened after handoff. That structure is what makes future learning possible.

Today the engine uses transparent, deterministic rules. No predictive model is in use, and none will be claimed until real outcome data supports it.

Why it is different

Matching decides who should meet. The engine is built for everything after.

How The Relationship Engine compares with matching and community tools
Capability	Matching	Community / networking	The Relationship Engine
Decides who should meet	Central	Sometimes	Designed for it
Guides what happens next	Not usually	Sometimes	Designed for it
Observes relationship signals	Not usually	Sometimes	Designed for it
Adapts what happens next	Not usually	Sometimes	Designed for it
Knows when to step away	Not usually	Not usually	Designed for it
Measures what continues afterwards	Not usually	Not usually	Designed for it

Decides who should meet
Matching
Central
Community / networking
Sometimes
The Relationship Engine
Designed for it
Guides what happens next
Matching
Not usually
Community / networking
Sometimes
The Relationship Engine
Designed for it
Observes relationship signals
Matching
Not usually
Community / networking
Sometimes
The Relationship Engine
Designed for it
Adapts what happens next
Matching
Not usually
Community / networking
Sometimes
The Relationship Engine
Designed for it
Knows when to step away
Matching
Not usually
Community / networking
Not usually
The Relationship Engine
Designed for it
Measures what continues afterwards
Matching
Not usually
Community / networking
Not usually
The Relationship Engine
Designed for it

General descriptions of tool categories, not of any named product. Many tools do parts of this well.

Fits around what you run

Available today, and what is planned.

Roster import by CSV with preview
Single use, expiring invitation links
Separate workspaces per organisation with role based access
Aggregate reporting with small-group suppression
Audited pseudonymous research export

Planned API and webhooks for running Circles inside existing platforms. The example shows the intended shape; it is not available yet.

POST /v1/cohorts/{id}/formation-runs
{
  "strategy": "balanced_diversity",
  "model_version": "formation-v1.4",
  "target_size": 6,
  "hard": { "max_same_group": 2 }
}

→ 201  { "circles": 8, "explanations": [...] }

webhook  circle.handoff_ready
webhook  outcome.day_30_recorded

Start with one population

See the engine run on a population you already have.

No rip-and-replace. No new community to maintain. One defined population, one measurable relationship journey.

Explore a pilot Measurement

We’re currently working with a small number of early partners.

1Choose an upcoming event, intake or cohort
2Invite an opted-in participant group
3Kindred forms and guides small Circles
4The engine hands off when criteria are met
5You receive 7/30/90-day aggregate outcome reporting