How the numbers are calculated

Average

Mean of your readings in the period, in mg/dL (mmol/L = mg/dL ÷ 18.0182).

Share of readings in range

Readings in each band ÷ all readings. Bands: very low below your very-low limit (default 54), low below target low (70), in range target low to target high (70–180, inclusive), high above target high up to your very-high limit (250), very high above that. This is a share of readings. A CGM reports continuous time-in-range; finger-prick readings cannot, because glucose between checks is unknown.

GMI (estimated)

3.31 + 0.02392 × average mg/dL. Defined for CGM data; here it is indicative only. It is not HbA1c.

Estimated A1c

(average mg/dL + 46.7) ÷ 28.7, the ADAG estimated-average-glucose relationship. Not a laboratory result and can differ from your lab A1c.

Variability (CV)

Standard deviation ÷ average × 100. A common target is 36% or lower.

Data coverage

Readings in the period, and how many calendar days of the period have at least one reading.

Typical day

Percentiles of readings grouped by hour of day. An hour needs at least 3 readings to appear.

Meal rise

For each meal: the latest reading up to 60 minutes before, and the first reading 1–3 hours after. Rise is after minus before. Meals without both readings are left out.

Active insulin (display only)

Each rapid-acting dose is assumed to wear off in a straight line over the duration you set. Real insulin action is not linear and varies between people. Do not use this figure to decide a dose.

Carbohydrate values

Typical household servings with a likely range. Recipes, brands and portion sizes vary; a package label always takes precedence.

Adherence

Each scheduled dose is matched with a logged dose of the same medicine within two hours either side; one log entry can satisfy only one scheduled dose. A scheduled dose with no match is "not recorded", which is not the same as missed. You can mark a dose as taken earlier or skipped, with a note. Adherence is the share of due doses logged or confirmed as taken. It never judges or advises.

Exercise and sleep

Active days (15+ minutes of logged activity) are compared with other days that have at least 2 readings, and each group needs 5 days. Sleep hours are paired with that morning's fasting reading and compared with a Pearson correlation once there are 7 pairs. These are associations in your own log and do not show cause.

Weekdays and weekends

Median reading by hour of day, with Saturday and Sunday on one line and the other days on the other.

Meal comparison

Meals are grouped by dish (portion counts ignored) and by breakfast, lunch or dinner, using the label if you gave one and the clock time otherwise. Groups of fewer than two meals are not shown.

Search

Questions are interpreted by fixed rules on this device, and the answer states how it was read: the record type, time range, meal context and limits used. Nothing is sent anywhere.

Weekly and monthly comparisons

The last 7 (or 30) calendar days, including today, against the same number of days before them. Differences are simple subtractions, and fewer readings make them less reliable.

Target presets

Starting points from the international consensus on time in range (Battelino et al., 2019). Your clinician's numbers always take precedence.

Importing a meter or CGM file

The file is read in your browser and never uploaded. Columns are matched by their names and shown for you to confirm. Dates like 03/04/2026 are never guessed: if the file does not make the order clear, you choose. Readings outside 20 to 600 mg/dL are skipped, and so are rows already in your log (same kind and value within 10 minutes). Insulin values are imported only when the type is stated or you choose one. Every import is tagged so the whole batch can be undone.

HbA1c comparison

The estimate uses the readings in the 90 days up to the test date and the same formula as elsewhere. It is shown only beside your lab result and is labelled unreliable when there are fewer than 30 readings or 20 days with a reading. HbA1c in % and in mmol/mol convert with mmol/mol = (% − 2.15) × 10.929.

Lab ranges

"Within", "above" or "below" always refers to the reference range that was printed on your lab report and that you entered. The app does not supply or interpret ranges.

Leaving out marked days

On Trends you can exclude whole days covered by an illness, travel, fasting or other marked event from every chart and figure. Nothing is deleted.

Calendar alarms

A web page cannot wake a phone on a schedule, so reminders inside the app only fire while it is open. The calendar file puts the same times in your phone's own calendar, whose alarms fire with the app closed. Times are floating (no time zone), so they stay at the same clock time wherever you are.

Backup to your own storage

Where the browser allows it, you choose a file and the app keeps it up to date. Otherwise the phone's share sheet sends a copy to an app you choose. Automatic copies are not encrypted; encrypted copies need your passphrase each time and the passphrase is never kept.

Syncing devices

Every record has a random id and a last-changed time. When you merge a sync file, a record that exists on both devices takes the more recent change; a record present on one device is added; a record deleted on one device is deleted on the other unless it was changed after the deletion. Identical records created separately on two devices are recognised by their content and not duplicated. The file is encrypted with your passphrase (AES-256-GCM, key from PBKDF2) and nothing is sent anywhere.

Snapshot links

A snapshot is compressed, optionally encrypted with a passphrase, and placed after the "#" of a link. Browsers never send that part to a server, so nobody, including MSRX, can see it. The cost is that a link is a copy that cannot be revoked and does not update; its expiry date is only applied by the viewer page.

Barcodes

The camera picture is processed on your device by the browser. Only the barcode number is sent to Open Food Facts, only when you allow it. Values come from a crowd-sourced database; some products give figures for the prepared food only, which is shown when so.

Each pattern shows the readings it is based on and keeps observations separate from possible explanations. Associations in your own log are not proof of cause. The app never recommends starting, stopping or changing a medicine or dose.

• Optional smart parsing only helps decide what an ambiguous number in a sentence you typed refers to (a reading, carbs or units) and when it was taken.
• It does not diagnose, prescribe, suggest doses or answer medical questions.
• It is off until you turn it on, and if it is unavailable the app uses its built-in rules.

MSRX GlucoNexa is a personal logbook and pattern viewer. It is not a medical device and not a substitute for advice from a clinician. For severe symptoms, confusion, or if someone cannot safely swallow, get emergency help. Never give food or drink to someone who is unconscious.