Physician performance
Provider identities, encounter history, work RVUs, and specialty groupings.
Do the reports use the same provider and period?
Signal Forward
The data, the reporting, and the review
Prepare the data, make the results useful, and give your team a way to review the work.
A practical architecture
Sources Prepared data Computed results Reporting & review
A possible arrangement for an engagement. Each capability can also support work independently.
From source records to analysis-ready data
Groundwork connects source records to shared entities and defined measures. Assess the existing estate, map the data, and build validation into the transformations.
Explore Groundwork →- Source records
- Shared definitions and checks
- Prepared analytical tables
Make computed results useful
Insights connects headline measures to definitions, trends, and breakdowns. The same scoped metrics can serve connected AI tools through MCP.
Explore Insights →
View full-size image (new tab) Follow a result back through the work
Provenance records instrumented stages, checks, and source evidence. Review the calculation and its inputs alongside the result.
Explore Provenance →
View full-size image (new tab) Groundwork
Source mapping, shared definitions, and validation in your cloud. Assess what is ready, resolve the gaps, and build the analytical foundation.
Insights
Physician performance, revenue cycle, and operational measures with trends, drill-ins, and governed access for connected AI tools.
Provenance
A record of instrumented workflow stages, checks, and evidence for review.
Signal Forward Groundwork Clinical & practice systems Billing & payment records Finance & operating reports
A data foundation your team can build on
Connect the source records, settle the definitions, and make the data ready for the decisions ahead.
We start inside your existing warehouse and reporting estate. Groundwork brings source mapping, data-quality checks, analytical models, and governance into a scoped delivery plan in your cloud.
Find where the numbers diverge
Different reports can apply different definitions to the same source. We trace the joins, filters, and calculation rules before deciding what needs to change.
Revenue cycle
Charges, payments, adjustments, denials, and accounts receivable.
Can payments be tied to the work that generated them?
Close and reconciliation
Ledger totals, compensation calculations, and management reporting.
Which differences are timing, and which are logic?
Illustrative assessment questions. Source availability and definitions vary by organization.
Preserve the source. Make the meaning consistent.
The architecture separates incoming records from shared entities and reporting measures. Existing components can stay where they already serve the work.
- 01 · Landed
Source records
Retain source-shaped data with load dates and source identifiers.
Check completeness and control totals. - 02 · Conformed
Shared entities
Map providers, encounters, charges, and appointments to consistent keys and definitions.
Check linkage, code sets, and rejected records. - 03 · Ready for analysis
Defined measures
Prepare reporting tables and versioned metric definitions at the required grain.
Reconcile results to agreed reference totals.
A record of the work
Where instrumented with Provenance, runs record inputs, transformations, validation results, and lineage across these steps.
Reference architecture, scoped per engagement. Prepared data can support existing BI tools, Insights, and governed AI access through Insights MCP.
Scope the build around what the data can support
A readiness assessment maps the measures you need to the sources you hold. The findings distinguish available data from missing feeds, weak links, and outside dependencies.
- Ready
- Computable from the available fields and history.
- Caveat
- Available, with a limitation such as short history or incomplete linkage.
- Derived
- Requires a documented calculation from source events or sequences.
- External
- Requires a third-party feed, licensed source, or other outside input.
Assessment categories, not a score for a particular organization.
How the work moves
Assess, design, build
Start with one domain and carry it from source to a working analytical view.
- Assess the estate. Profile the available data, review existing reporting definitions, and identify readiness gaps.
- Design the first delivery. Map source fields, agree on definitions and owners, and sequence the dependencies.
- Build and reconcile. Deliver the transformations, validation checks, and reporting tables with a documented handoff.
Our work includes Azure data-serving delivery, warehouse assessment, and governance and lakehouse design. Platform selection follows your existing environment and the requirements of the engagement.
The Glass Box approach
Inspect the assumptions. Follow the calculations. Review the checks.
The data foundation holds the definitions and prepared data. Reporting presents the results. The workflow record supports review of how a result was produced. Validation of the method and its intended use remains part of the work.
Healthcare reference data
Consistent definitions beneath the analysis
Diagnosis, procedure, and DRG classifications help organize claims into clinical and service-line views.
75,238
diagnosis codes in the reference taxonomy
~100K
procedure codes across CPT, PCS, and HCPCS
771
MS-DRGs with CMS weights
These figures describe the reference-data build represented here; coverage varies with code-set version.
Our reference-data work combines published sources, configured taxonomies, and versioned classification tables. Prepared tables support repeatable lookups; checks identify gaps and inconsistent mappings. We review the definitions for the intended use, including any inferred complexity measures.
From source definitions to repeatable lookups
The reference build connects four-level taxonomies through 19 service lines, with five-tier complexity and acuity measures. Published AHRQ and CMS sources are combined with configured categories and AI-assisted assignments during preparation. Once prepared, classification uses versioned tables without a model call for each lookup.
CSV and Parquet reference files, SQL generators, and a Python scorer support use in analytical workflows. Eight tiers of validation check the build for coverage and consistency. A 15-organization evaluation produced an average complexity index of 2.65; that result describes the study sample, not a universal clinical benchmark.
Get started
Ask for a demonstration
Our working demos use synthetic data. We can walk through a run and a report, then discuss the requirements for your environment.