feat(ios): order a hundred analytes by what is furthest from its range, filter them, and derive what a sheet did not print

[ma3u]

A hundred and one analytes in a list is a list nobody reads to the end, and the seven that sit outside a published range are the seven worth seeing.

Order and filter

Three orders, with the choice at the top of the screen: furthest from its published band, most measured, by name. The first is the default.

Four filter chips, one per place, each carrying how many sit there. The counts are of everything rather than of what is showing, so turning a filter on does not change the numbers beside the filters. An empty selection means all of them: a filter that can hide everything is one people get stuck in.

Deliberately not "high risk" and "risk", which is what was asked for. A risk is a statement about a person, and that is the line section 5 of #186 draws. The order is exactly the same; the claim is one the app can show its working for.

Values worked out from other values

Derived at read time and never stored, so they follow the values they came from rather than drifting from them.

Derived	From	Band
Non-HDL cholesterol	total − HDL	ESC/EAS 2019, already in the table
Remnant cholesterol	total − HDL − LDL	none quoted
Triglycerides / HDL	both, same unit	none quoted
LDL / HDL	both, same unit	none quoted
ApoB / ApoA1	both, same unit	none quoted

Two rules, both tested:

• Only from one report. A triglyceride from March over an HDL from September is a ratio of nothing.
• Only within one unit. Triglycerides over HDL is a different number in mg/dL and in mmol/L, by about a factor of two, so what it was computed in travels with it.

A figure the laboratory printed itself is never replaced by a computed one.

Why the ratios carry no code

LOINC's nearest terms are a molar ratio where this is computed in whatever the sheet printed (55607-6), HDL over LDL the other way round (16616-5), and a percentile rather than a ratio (9340-1). A code that is nearly right is wrong.

Their UCUM unit is {ratio}, not 1. A bare 1 as a unit spelling makes every stray number on a page a measurement, and it did: a page footer reading Seite 1 von 2 started arriving as a value, which two existing tests caught.

Gates

Gate	Result
swift test	239 tests, 42 suites
Scripts/run-ui-tests.sh	22 tests, 0 failures
AnalyteParity	1105 codings, 66 units
epa-ingest Vitest	160 tests
check-localization.sh	241 strings, every prose string has German

Installed on the phone.

Refs #186.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LFMtTkqc3btcaWNur2A93P