Decision Systems for Health Networks
- MSSS — Urgences (horaire)
- MSSS — Répertoire M02
- INSPQ — Soins intensifs
- ASPC — Eaux usées
Every panel is a screen that exists today.
Our software turns a health network’s published files into one model of itself, so a surge response can be compared before it is chosen rather than argued about after the fact.
Four layers, one lineage.
Connect what exists, model it once, run it under load, and compare what you could do about it. Each figure on the last screen traces back to the file on the first.
1
Connect
Sources and syncs
Point at a REST endpoint, a file, or a stream and set an interval. Encodings, delimiters and schedules are handled where they belong — not in a spreadsheet before upload.
2
Model
Pipelines into an ontology
A visual graph turns rows into the objects a network actually has: facilities, beds, territories, and the links between them. Terminology coding runs as a node in that graph, not as a separate migration.
3
Run
The digital twin
The model is put under an observed load and the network responds — beds fill, patients wait, transfers move. Deterministic, so the same question always returns the same answer.
4
Decide
Compare the responses
Two protocols, side by side, on deaths, waiting and cost. Ranked by Pareto dominance, so an option that is worse on every axis is named as dominated rather than hidden inside a weighted score nobody agreed on.
Two sources. One region they agree on.
Nothing is merged and nothing is overwritten. A published file keeps its own vocabulary and its own calendar; what the platform builds is the region where two of them describe the same thing — and that region is the only part a decision can rest on.
- Read, not declared
- A CSV column arrives typed as text whatever it holds. The values are read to find the dates, the measures, the identifiers — which is how a permit number stops being offered as something to sum, and how a date column written the wrong way round gets caught instead of charted.
- The file that arrives is the file that was published
- Encoding, delimiters, unit expansion and coding all happen inside the platform, where they leave a trace. Fixing them in a spreadsheet first is what makes provenance impossible to defend later.
The same arithmetic, whatever the network.
The engine reasons about capacity, occupancy and travel time. Not one of its mechanics knows the name of a city or of a disease, which is why a demonstration built on Montréal open data is a demonstration of the machinery rather than of one region.
- Lives
- Cases that went unserved, and what that cost.
- Waiting
- Patient-days spent waiting for a place.
- Cost
- What the response itself consumed.
Responses are ranked by dominance, never by a weighted score. An option worse on every axis is named as dominated; where two options each win on something, the platform says so instead of inventing an exchange rate between a death and a dollar that nobody agreed to.
What the demonstration runs on.
Not a sandbox. The Studio reads published government files, with their real encodings, their real gaps and their real publication schedules.
MSSS · Québec
The M02 registry of installations and the capacities authorised at permit. Published in Windows-1252, with tabs padding a column name.
312 installations · region 06
MSSS · Québec
The hourly emergency-room census. Four hospitals in sixteen report “pas d’information disponible”, and the platform says so on every chart.
REST · hourly
INSPQ · Québec
Nine daily epidemiological series — admissions, intensive care, deaths, positivity, variants, reproduction rate. Laid over a run to ask whether it reproduced what happened.
2020–2023 · daily
PHAC · Canada
SARS-CoV-2 in wastewater, by collection basin. An early signal that owes nothing to who came forward for a test.
89 sampling days
One of these arrived with its day and month written the wrong way round. The platform caught it, named it, and it was corrected — which is the point of reading values instead of trusting a schema.
See it on your own network.
Obscyro is in its test phase and is not being sold. A demo is a working session on your data or on ours — we would rather understand the decision you cannot defend today than present slides.