A correction is useful when the next check shows what improved and what still needs attention. The October Records update brings that comparison into the normal Validation pane and extends the path from dataset-quality findings to a return package for the data holder.
The recorded workflow shows the current interface on one synthetic Observation. It follows a reviewed Claude Code correction, independent Records revalidation, an intentional regression and recovery. The film was recorded on 22 October 2026 and includes English captions and a transcript.
Read the changes beside the finding
New, Resolved and Changed markers stay with the finding's validation aspect and field. A resolved finding remains inspectable without contributing to the current severity total. Changed findings retain the before-and-after explanation.
The comparison depends on where you work:
- Resource revalidation: the last two complete checks of the same current resource. The browser holds a temporary comparison; reloading clears it.
- Sandbox: the first complete check is the baseline for the temporary draft.
- Run comparison: an explicitly selected baseline run is compared with another recorded run. Issue types and occurrence counts remain separate: fewer occurrences do not necessarily mean a resolved issue type.
Incomplete checks and changed validation inputs can prevent a comparison. A temporary resource comparison does not retain immutable profile and terminology snapshots, so observed finding changes alone do not establish that the resource edit caused every difference.
Check a repair against both requirements
The film starts with a synthetic oxygen-saturation Observation that has the laboratory category but lacks vital-signs. The selected oxygen-saturation profile requires the latter; the laboratory profile still requires the former.
The reviewed correction adds vital-signs and keeps laboratory. After the source write and readback, Records checks the resource again. A deliberate negative control then removes laboratory, and the next check exposes the new laboratory-profile finding. Restoring both categories completes the demonstration.
The source measurement remains unchanged. A separate unresolved extension-definition finding remains visible throughout; this is not a demonstration of complete conformance or clinical correctness. See AI-assisted FHIR workflows for the film and the separate tool-integration example.
Open SYNDERAI Release 5 in the public demo
The demo page now provides direct entries to the synthetic documents on Medplum and Health Samurai. Start in Bundle, then inspect the embedded resources.
The sources have different scopes. Medplum includes laboratory reports, patient summaries and discharge reports. Health Samurai includes laboratory reports and patient summaries; its discharge reports remain excluded after an R4 status-binding rejection. An embedded Patient or Observation is not necessarily available as a separate searchable resource.
Return dataset-quality findings to the data holder
The native MII-oriented workflow connects a selected population and versioned quality programs to findings, affected resources and CSV ZIP or Excel return tables. Read each indicator with its numerator, denominator and not-evaluable cases. A zero denominator is not a perfect score.
Alongside independently defined indicators, the workflow includes selected CQL-derived resource functions. The dated scope summary separates the native function comparison from the synthetic workflow example. It does not qualify whole Measures, full original CQL execution, original R execution or customer acceptance. The return-table format remains provisional.
The research-data workflow explains how the receiving team can use those results to arrange a correction and a comparable follow-up run.
Choose the available integration
The website's Records product reference is now 0.14.2. This does not identify every hosted deployment. Public package availability remains separate: the npm validator is 0.9.18 and the CLI is 0.1.1 at the 23 October release check. The source CLI version 0.1.3 is not substituted into installation examples.
Use the developer guide to choose the CLI, embedded validator, Agent Tools or MCP surface. Workspace access remains staged and is agreed around a specific evaluation.
