Four sites, four regulatory files, and one device they are all meant to describe.
The device is described once. Each site attaches its own validation evidence to it, so a divergence between sites becomes visible instead of becoming a finding.
- Orbenne Santé
- Dispositif Kelvyn
- 🧩Modèles
- 📐Spécification
- ✅Validations
- par_site4
- essais_cliniques6
- 📈Surveillance post-marché
VAL-Kelvyn — Validation de performance
- dispositif
- Kelvyn — plateforme d'imagerie
- specification_de_reference*
- SPEC-Kelvyn · révision 5
- sites_ayant_valide
- 4
- sites_sur_revision_courante
- 3
- divergence*
- 1 site en retard
- statut_global*
- À aligner
Trois sites ont validé contre la révision 5 de la spécification. Le quatrième a validé contre la révision 4, avant la modification du protocole d'étalonnage. L'écart est visible ici, pas en audit.
| Name | revision_validee* | essais* | statut* |
|---|---|---|---|
| Site A — Ligne pilote | Révision 5 | Complets | Validé |
| Site C — Production série | Révision 5 | Complets | Validé |
| Site D — Sous-traitance | Révision 4 | Partiels | À rejouer |
The customer's workspace, anonymised. Real structure, fields and states.
Anonymised at the customer's request. The organisation, the device, the sites and every figure are changed. The multi-site problem and the model are as they were.
Each site is rigorous. Together they describe four slightly different devices.
One device, four branches of evidence hanging off it.
Validation lives where it was produced
Each site keeps its own file. Nobody holds a view across the four without asking four people.
The device is described once
Specification, configuration and intended use live on the device item, not in each site's copy of them.
Divergence is discovered, not detected
Two sites validating against different revisions of the same specification is normal until an auditor notices.
Each site writes its own evidence
Validation records attach to the device with the site as an attribute, so local rigour is preserved and made comparable.
The lifecycle is split across documents
Trials, certifications and post-market evidence sit in separate documents per site rather than on the device.
Divergence is a computed column
Which sites validated against which revision is a value, visible before an audit rather than during one.
The site is an attribute, not a filing cabinet.
- Orbenne Santé
- Dispositif Kelvyn
- 🧩Modèles
- 📐Spécification
- ✅Validations
- par_site4
- essais_cliniques6
- 📈Surveillance post-marché
VAL-Kelvyn — Validation de performance
- dispositif
- Kelvyn — plateforme d'imagerie
- specification_de_reference*
- SPEC-Kelvyn · révision 5
- sites_ayant_valide
- 4
- sites_sur_revision_courante
- 3
- divergence*
- 1 site en retard
- statut_global*
- À aligner
Trois sites ont validé contre la révision 5 de la spécification. Le quatrième a validé contre la révision 4, avant la modification du protocole d'étalonnage. L'écart est visible ici, pas en audit.
| Name | revision_validee* | essais* | statut* |
|---|---|---|---|
| Site A — Ligne pilote | Révision 5 | Complets | Validé |
| Site C — Production série | Révision 5 | Complets | Validé |
| Site D — Sous-traitance | Révision 4 | Partiels | À rejouer |
What we modelled
One specification, four validations
The view nobody had: which site validated against which revision, on one screen.
Validation status, by site
Four sites against one specification revision. The right-hand column is computed, so nobody has to compare four files.
| Site | Révision validée | Essais | Écart |
|---|---|---|---|
| Site A — Ligne pilote | 5 | Complets | Aucun |
| Site B — Assemblage | 5 | Complets | Aucun |
| Site C — Production série | 5 | Complets | Aucun |
| Site D — Sous-traitance | 4 | Partiels | 1 révision |
ƒ Écart is computed by Koddex, not typed by anyone.
One site validation, in full
Site D — Sous-traitance
VAL-Kelvyn-D- site
- Site D — Sous-traitance
- specification_validee
- SPEC-Kelvyn · révision 4
- essais
- Partiels
- ecart_de_revision
- 1 révisionspec.current_rev − site.validated_rev
What the graph computes
The divergence that used to surface during an audit, available before one.
Makes approving a specification revision a decision with a named cost across the network.
One device description, per-site evidence, and divergence as a value rather than a finding.
The network stopped describing four devices.
| Before | After | |
|---|---|---|
| The device description | One copy per site | One item, four branches |
| Divergence between sites | Found by an auditor | A computed column |
| Local rigour | Preserved, and invisible | Preserved, and comparable |
| Approving a spec revision | A document change | A named list of sites to re-validate |
| A view across the network | Ask four people | One screen |
Three things worth knowing before you defend this internally.
Sites have to agree on one specification
Making divergence visible does not resolve it. The first weeks surface disagreements that four separate files had been quietly absorbing.
It does not perform the validation
Trials and tests happen where they happen. Koddex holds which revision each site validated against and what evidence it produced.
Historic files stay historic
Past validations are referenced rather than re-keyed. The computed divergence is meaningful from the revision you start at, not retroactively.
Frequently asked questions
Does each site lose its autonomy?
No — that was the condition. Each site still produces its own validation evidence, with its own people and its own protocols. What changes is that all of it hangs off one device description, so the four can be compared without four requests.
What happens when a specification is revised?
The sites still validated against the previous revision are named immediately. Approving a revision therefore carries a visible cost across the network, which is precisely the information that used to arrive late.
Is this only for MDSAP-style multi-site audits?
It shows up most sharply there, because an auditor crossing sites is exactly who finds the divergence. But any organisation validating one device in more than one place has the same problem the rest of the year.