SILD Pilot — Review your data transfer.
Are you changing an interface, preparing a migration or reviewing a specific transfer? The SILD Pilot compares agreed information in matched source and target data against rules defined in writing, using authorised test data or effectively anonymised data.
You receive a written report, a machine-readable findings list and one limited re-test following an agreed correction. Preserved information, demonstrated deviations and unresolved cases are reported separately.
up to 1,000 unambiguously matched source/target pairs
up to 10 units of meaning agreed in writing · 1 limited re-test
Paid pilot. Binding fixed price once the scope of review has been jointly defined. The 30-minute initial call is free of charge and without obligation.
Do you also need a new adapter, a mapping correction or migration support? These project services are agreed separately. View project services.
What the pilot covers.
The pilot is deliberately limited: it delivers traceable results for one transfer and up to ten agreed review rules. The scope is agreed in writing before work starts.
| Transfer | One defined transfer source → target, one named use case, one message or resource class, one recorded version and configuration state. |
| Data | Up to 1,000 unambiguously matched source/target pairs from released test data or effectively anonymised data. A running production installation is not part of the standard pilot. |
| Formats | Existing export formats and available parsers or mappings. Whether that is sufficient for your transfer is clarified before you commission the pilot. A new proprietary format, a new adapter, operation on your premises or special access require an additional offer agreed in advance. |
| Review rules | Up to ten units of meaning jointly agreed in writing, for example value and unit, code and code system, time period, reference relationship or document class. |
| Deliverables | Review agreement · written report · machine-readable findings list (CSV/JSON) · the input, configuration and tool versions needed for interpretation, with hashes · 60-minute explanatory call · prioritised technical next steps. |
| Re-test | One limited re-test is included: once, on the same transfer, with the same units of meaning and at most the same number of cases, after an agreed correction. The corrected data must be available within 30 calendar days of report handover. The re-test is not a comprehensive re-certification; new questions, adapters or data classes are additional scope. |
| Delivery | Targeted within ten working days of all agreed inputs being fully provided and technical feasibility being confirmed. The offer states the specific start date. |
What exactly is reviewed.
A review rule states which information must be readable again in the target system. The target counts as a whole: a value may sit in another place or only become readable through a code conversion — so we do not merely compare individual fields.
The dose stays the same.
The dose remains unambiguously determinable, for example 12 mg also as 0.012 g. Review rule: value and unit, interpreted together, yield the same dose in the target as in the source.
The code keeps its system.
Review rule: the source code and its code system can be determined unambiguously from the target context.
A complement, not a replacement.
Conformance validators check more than syntax — for example profiles, terminology bindings and invariants (HL7 FHIR R4: Validation) — and can point to problems. Passing the chosen target validation alone, however, does not show that the agreed source information has been preserved.
Five statuses, reported separately.
Each reviewed case receives exactly one status per review rule. We report the denominator, the review coverage and unresolved and non-reviewable cases separately — and no global percentage for “meaning fidelity”.
Review rule met
The agreed reconstruction succeeds for the reviewed unit.
Deviation evidenced
A specific deviation is backed by evidence; the report explains what it means according to the review agreement.
Unresolved
For example because part of the target data or a code conversion (decoder) is missing. A missing expected field is initially a finding or a limit of the review, not a proven loss.
Not reviewable
For example because the source or the source ↔ target mapping was not provided.
Not applicable
The review rule does not apply to this case.
What this looks like is shown in the illustrative sample report (PDF, in German) with six synthetic example cases and the matching findings list (CSV, German column names). It is not a customer report and not evidence of an actual review run.
From the initial call to a traceable report.
Initial call (30 minutes, free of charge)
We clarify the transfer, the source/target pairs needed, up to ten review questions and whether your transfer is suitable for the pilot. Please do not send any patient data for this.
Review agreement and fixed price
Before work starts we agree in writing: permitted data access, the specific source and target, mapping, review rules, versions, responsible persons and a secure transfer channel. If prerequisites are missing, we clarify them first.
Review run
The agreed pairs are reviewed against the agreed rules. Data processing and confidentiality follow the actual set-up and are agreed in advance.
Report and explanatory call (60 minutes)
Result groups with evidence, findings list, versions and hashes, prioritised technical next steps. Findings are assigned to the responsible teams for review.
Re-test (once, limited)
Same transfer, same units of meaning, at most the same number of cases — after an agreed correction, with data provided within 30 calendar days of the report.
Paid pilot, binding fixed price.
| Initial call | 30 minutes, free of charge and without obligation. |
| Pilot | Paid. Binding fixed price once the scope of review has been jointly defined, stated in writing in the offer. One limited re-test is included. |
| Additional scope | New proprietary format, new adapter, operation on your premises, special access, further transfers, questions or data classes: only under an additional offer agreed in advance. |
| Continuous operation | Continuous monitoring in production is not part of the pilot and is currently not a finished standard product. |
| Reference | No obligation to be named. A reference report requires your separate consent — regardless of the result. |
Application example KDL: the switch of German document classification to KDL can be reviewed as one such transfer within the same pilot — see the KDL audit page (in German).
What the pilot does not state.
The SILD Pilot is a technical quality review of data transfers: it checks whether agreed information can be read back completely and correctly in the target after one defined transfer. It makes no clinical assessments and gives no diagnostic or therapeutic recommendations. We only make statements about regulatory status after a separate assessment.
- Sample and defined rules — not a full review of all medical meaning.
- No finding in the sample is not proof of general freedom from loss.
- A deviation does not by itself prove an irreversible loss of information. That additionally requires showing that the agreed information cannot be reconstructed from the complete target under the agreed model and coding assumptions.
- No clinical safety case, no conformance certification, no attribution of fault or liability.
- No promise that errors will be found or that a particular result will occur.
- Hashes show which inputs were reviewed; on their own they prove neither reproducibility nor preservation of meaning.
- Technical handling priorities can be agreed; they are not clinically validated severity levels.
Read first, then talk.
The scope of review and the sample report are available for download (in German). The FM-4 model and the accompanying papers are published on Zenodo as self-archived preprints that have not been peer-reviewed. For today’s technical classification: statements about coverage presuppose a declared model and fixed decomposition and variation assumptions; they verify neither any software nor your transfer.