Clearwater
Transformation Plan
Sphere × Clearwater Savings · OrgBrain Transformation

The plan,
made navigable.

Executing the OrgBrain transformation on the Sphere stack.

This is the interactive companion to the Clearwater Savings Transformation Plan — the execution detail behind the Roadmap. Every workstream is organized on Sphere's six-layer operating model, where data is the gravitational center, work is built from the foundation upward, and each domain ships as a vertical slice. Explore the layers, stages, registers, and the synthetic workforce below.

6
Operating layers
3
Delivery stages
18
Workstreams
10
Systems dispositioned
14
Source-of-record domains
10
Tracked risks
01 · Orientation

How to read this plan

Where the Roadmap sets the destination — consolidating Clearwater onto OrgBrain with a synthetic workforce operating against one governed Neural Knowledge Graph — this plan gets into the weeds: the concrete workstreams, the build detail, the source-of-record calls, the entry and exit criteria, and the governance that holds it together.

The relationship

Two documents, two jobs

The Roadmap owns scope, vision, and stage intent; nothing here overrides it. This plan inherits its Crawl, Walk, Run stages and decomposes each into executable work — specifying sequencing and dependencies, while dates and resourcing stay with joint planning.

The lens

The Sphere operating model

Every workstream is organized against the six-layer stack. Data is the gravitational center; work is built bottom-up and used top-down; and each domain is delivered as a vertical slice through all six layers rather than as a horizontal technology project.

The conventions

How work is described

Each unit of work has a stable ID (e.g. WS1.2), the Sphere layers it touches, entry/exit criteria that gate it, and explicit source-of-record decisions. Proposed items needing confirmation are gathered in Open Questions.

Status. This is a working delivery plan intended to be refined jointly — detailed enough to begin Crawl and scope Walk and Run, while leaving dates, resourcing, and contested source-of-record calls open for the planning sessions the Roadmap anticipates.
02 · The operating model

The six-layer Sphere stack

Read top-down, the layers describe how a request flows when the system is used. Read bottom-up, they describe the order in which the system is built. Same architecture — the difference is whether you are a request flowing through it, or an engineer laying its foundations. Select a layer to explore it.

Use · top-down
Build · bottom-up
Data
The gravitational center
What it is
At Clearwater
The principle that drives the method. Everything converges on the data layer; everything diverges through the action layer. Ontologies are the shared vocabulary every layer speaks. The work of the transformation is making sure what happens between those two layers is clear, governable, and re-composable as the world changes around Clearwater.
03 · The repeatable pattern

The vertical-slice playbook

The unit of delivery is a vertical slice: one domain — plans, claims, care coordination — taken all the way up the stack, from integration through to a working agent or tool. Plan Management in Crawl is the reference slice; every later domain reuses its shape, onboarding faster than the last. Step through it.

01
Integrate
domain in → integrate → normalize → model → govern → expose → act → working agent / tool out
Definition of Done for a slice. A slice is complete when its data is governed in the Neural Knowledge Graph with full lineage, its source of record is recorded, its API surface is live, its business rules are enforced in the model (not in scattered code), and its agents or tools are in production with quality metrics on the dashboard and human-in-the-loop thresholds enforced.
04 · Plan at a glance

The stage × layer matrix

The single-page synthesis of the Roadmap's stages and the Sphere model: what gets built in each layer, across each stage. Hover to trace a row or column; tap any cell to read it in full; use the filters to focus a stage.

Focus stage
Select a cell to see the detail for that layer in that stage.
05 · The data layer

The Neural Knowledge Graph & its API

For each data element, exactly one system is the system of record; the Neural Knowledge Graph either holds it or transparently routes to the system that does. A consistent API surface sits on top and is reused by every tool and agent.

FHIR
FIBO
+ Clearwater ext.
Neural Knowledge Graph
The gravitational center · full lineage on every value
Query current state
Read the authoritative current value of any element.
Update
Where the Neural Knowledge Graph is the system of record, accept writes with full validation against the model's constraints.
Redirect to source
Where another system owns the element, transparently route to the correct system of record.
Lineage. Every value carries where it originated, when it arrived, what transformations it underwent, and who last changed it. “How do we know this is true?” becomes an answerable query — not a research project.

Business rules as first-class model citizens

Rules, events, and constraints are encoded in the model — visible, testable, governed, and enforced consistently by every agent and app that touches the data. The Roadmap names two; both are built early as exemplars.

Eligibility precondition enforced at model boundary

A care-coordination request cannot be opened for a person who is not eligible for a plan — so neither a human tool nor an agent can violate it.

Non-payment termination time-based event

If a member has not paid within thirty days of the due date, they become termed — encoded as a time-based event on the coverage/payment model, with lineage on the resulting status change.

06 · The synthetic workforce

Agents, specified and supervised

Agents are digital workers, not standalone chatbots: each has a defined responsibility, operates against the Neural Knowledge Graph, escalates on defined thresholds, and produces an auditable trail of every action. A shared framework makes each new agent faster to build and safe by default.

Every agent is specified with

01

A single, scoped responsibility and the events that trigger it.

02

Its inputs, drawn from the graph — never from ad-hoc integrations.

03

The checks or reasoning it performs and the outputs it produces.

04

Auto-action limits and the human-in-the-loop thresholds that force escalation.

05

Quality metrics and the QA sampling method used to verify it.

06

A rollback and a pause control.

The promotion path — every agent earns its autonomy

1 Stage one

Shadow

Produces decisions without acting. Runs until its QA pass rate clears the agreed bar.

2 Stage two

Supervised

Acts within tight thresholds with human review. Findings, lineage, and escalations are visible on the dashboard.

3 Stage three

Steady-state

Operates within agreed parameters. Promotion at each step is a governed decision, with thresholds tunable and rollback verified.

The dashboard is the operating hub. From the first agent live, the Synthetic Workforce Dashboard is a registry of every digitized process, quality and outcome metrics per process, live monitoring, approval tracking and human-in-the-loop oversight, and visibility into where humans intervene — so processes can be tuned over time.
07 · Identified opportunities

Use cases identified

The concrete jobs the synthetic workforce and the Neural Knowledge Graph take on — each one a vertical slice that integrates, governs, and acts on a single domain. Filter by stage to see what is identified for each stage; every card names the agent or capability and the workstream that delivers it.

Stage
08 · The roadmap

Crawl, walk, run — and where we begin

The transformation runs in three stages — crawl, then walk, then run — each delivering value on its own. Select a stage on the timeline to see its intent, headline deliverables, and definition of success — then below, the concrete starting point: the data Crawl connects first.

Crawl — connecting the data

Crawl stands up the first connections into the Neural Knowledge Graph and proves the slice playbook on Plan Management. Three sources come online first:

09 · Delivery log

Key dates & milestones

The running record of what has been delivered and what is next on the path to a working Crawl stage. Select any row to see what the milestone is and why it matters to the vision.

Status

UAT & release readiness

Current state of each capability in UAT and the next step toward production. Most capabilities are UAT-tested and confirmed by the Clearwater team and are ready to release to PROD.

CapabilityStatusNext Steps
Plan Management / Source of Truth (SOT) including SPD (Summary Plan Description)UAT tested & confirmed by CW teamRelease to PROD
IPS / Salesforce IntegrationUAT tested & confirmed by CW teamRelease to PROD
G Suite Integration
Gmail & G Calendar
UAT tested & confirmed by CW teamRelease to PROD
G DriveAlready released to PROD
Global SnippetsUAT tested & confirmed by CW teamRelease to PROD
RegulationsUAT tested & confirmed by CW teamRelease to PROD
Talk to DatabaseAvailable for testing in UATUAT testing & confirmation by CW team
Coming up in UAT. Slack Integration — UAT sync in progress (ETA Jul 31, 2026).
10 · Execution

Crawl, walk, run

Each stage delivers operational value on its own — a sequence of capability deliveries, not a multi-year project that pays off only at the end. Select a stage, then open any workstream for its objective, build detail, source-of-record calls, and entry/exit gates.

11 · The landscape

System disposition

For each system: its disposition, integration pattern, role relative to the source of record, and the stage in which the work primarily lands. Dispositions follow the Roadmap; patterns and SoR roles are Sphere's recommended approach for confirmation.

Filter
12 · Authority

Source-of-record register

Exactly one system is authoritative for each data element; the Neural Knowledge Graph holds it or transparently routes to the system that does. Items marked contested are the calls the Roadmap flags as sometimes disputed — resolved by the SoR Council with the business owner and recorded in the model.

Show contested only
Proposed Contested
13 · Managed risk

Risk register

These areas are actively managed throughout — and mitigated substantially by the staged, slice-based approach itself, since each stage de-risks the next. Tap a marker on the matrix to read its mitigation and owner. Likelihood and impact are initial assessments for joint review.

Impact →
Likelihood →
Select a risk marker to see its detail. Markers are positioned by likelihood (horizontal) and impact (vertical); colour shows overall severity.
14 · Operating model

Governance & decision rights

Delivery is joint — a Sphere pod and Clearwater counterparts. Governance is deliberately lightweight but decisive: it makes source-of-record calls, clears dependencies, approves stage gates, and oversees the quality of the synthetic workforce.

Joint delivery roles

Decision rights — RACI

Cadences

How the program stays aligned

Slice standup on the active slice(s); the Source-of-Record Council to resolve contested items; an agent quality review of dashboard metrics to tune thresholds; and the Governance Board at every stage gate for gate exit, scope, retirements, and resourcing.

Environments & promotion

Dev → validation → staging → production

Every agent runs in shadow, then supervised, then steady-state; promotion is governed. Non-production uses de-identified or masked data with least-privilege, logged access. Each agent and cutover has a defined rollback and can be paused org-wide from the dashboard.

15 · Measurement

Metrics & KPIs

Quality metrics, lineage, and audit trails become the operating norm from the Walk stage onward, tracked on the dashboard and in program reporting. Targets are baseline-plus-improvement set in joint planning — the discipline is to baseline before automating, and to tune against the human-intervention signal.

16 · The human side

Change management & the human–agent workforce

Agents are deployed alongside human staff, not in place of them. People retain authority over exceptions, escalations, and judgment-heavy work; the dashboard makes the collaboration visible and tunable. Roles evolve rather than disappear.

Claims processorsException specialists & agent supervisors
Routine, in-parameter adjudication shifts to agents; people focus on the exceptions surfaced with full context and on tuning thresholds.
Service-desk agentsEscalation & knowledge owners
Self-service and routine routing are handled on OrgBrain; staff own complex cases and the knowledge the desk draws on.
Coordinators & cliniciansAugmented practitioners
Intake, follow-up, triage, and documentation support are agent-assisted; people focus on care judgment and relationships.
Enablement. Training tied to each go-live; clear communications on what changes, what stays human, and why — reinforcing augment, don't replace; feedback loops from the people doing the work into threshold tuning; and adoption and sentiment monitored as leading indicators alongside the quality metrics.
17 · Definition of Done

Stage gates

A stage does not close until its gate criteria are demonstrably met and approved by the Governance Board. The per-slice Definition of Done applies within every stage: data governed with lineage, source of record recorded, API surface live, rules enforced in the model, and agents or tools in production with quality metrics and human-in-the-loop thresholds enforced.

18 · For joint planning

Open questions

Consistent with the Roadmap, certain decisions are deliberately left to joint planning. Resolving the items below converts this from a detailed plan into a scheduled, resourced delivery program.