Technical guide

For people changing this code, integrating with it, or working out why a result looks the way it does.

• How it is put together — the processes, what runs where, and why the renderer's privileges are load-bearing.
• Protocols on the wire — ASTM E1381 framing and E1394 records, HL7 over MLLP, and the failure modes that are not in the standards.
• Analyzers as they behave — what each model actually sends, taken from captures of live laboratories rather than documentation.
• What is stored, and where — the tables, the migrations, and the rule that governs every value in them.
• Result webhook — the JSON request an HTTP receiver gets for each batch of results, and when a result counts as delivered.
• Usage statistics — what is recorded locally and what leaves the machine.
• Building and releasing — the gate, the release, and the identifiers that must not move.

The one rule worth reading first

A value is stored exactly as the analyzer sent it. No rounding, no locale normalisation, no unit conversion, no < 40 turned into 40. A laboratory can choose otherwise for its own LIS with result rules, and the value as sent is still kept beside the stored one.

The analyzers disagree with each other about almost everything — decimal commas, TND versus Not detected versus Target Not Detected, mantissas with power-of-ten units — and each of those disagreements is the laboratory's to resolve, not this tool's. A tool that tidied values on the way past would be a tool whose output nobody could check against the analyzer's own printout, and the person at the end of that chain is a patient on treatment.

Everything else in this guide is downstream of that.