One practical addition: if an e-mail from the board never arrives, tell us here from the account it should have gone to.

Registration mail is the single most common thing to go missing — some providers discard forum mail silently, and it lands in a spam folder often enough to be worth checking first. If a resend does not fix it, that is a real problem on our side and we would rather chase it than have somebody quietly give up at the registration step.

Since this is the bug category, the shape of a report we can actually act on:

• What you did, in the order you did it.
• What you expected, and what happened instead.
• Where: which view, which browser or the phone app, roughly when.
• Screenshots with the numbers cropped out if you would rather not share them — the layout is usually what we need, not your data.

What not to put in a public thread: your account e-mail, your device serial number, or a photograph of anything from a clinic. If a report needs your account looked at, the in-app feedback button is attached to it and this board is not.

The cancel worked. The wait is the cleanup, and it is the part worth understanding, because the instinct to "just upload again" is the one thing that makes it worse.

What happens when you cancel:

1. Cancellation is acknowledged immediately — the importer stops at the next safe boundary rather than in the middle of writing a session.
2. Then every session, event, waveform and file from that upload is removed, in the background. On a large import that removal legitimately takes minutes. The indicator stays visible saying it is removing data, with the button disabled.
3. The page refreshes itself when cleanup finishes, so the calendar and charts reflect reality rather than what was there mid-delete.

Do not upload again while "Cancelling…" is still showing. A second upload landing on top of a running cleanup is exactly the way to end up unsure what state you are in.

Cancelling only removes the upload being cancelled. Data imported earlier is never touched — there is no "cancel" that reaches backwards.

If you did upload again mid-cleanup: wait for both to settle, then check the calendar. If what you see does not match what you expect, that is worth a report — with the profile and the approximate time, not with any account details.

Fair, and short.

Silver is the core experience: uploads, the dashboard tiles, the Daily / Overview / Statistics views, and the printable reports. Everything about getting your data in, seeing it, and taking a document to an appointment works on Silver.

Locked on Silver, unlocked on Gold:

• the Charts waveform view,
• SomniScan (deep analysis) and SomniPattern (pattern detection),
• the SomniDoc AI analysis, which needs an eligible tier,
• the clinician brief and its sharing options (print, PDF, tracked e-mail, expiring links).

A locked tile is not hidden — it tells you what is behind it and what it would cost. No feature quietly disappears from a menu.

Two mechanics worth knowing, because they are the ones that surprise people:

• The free trial clock is per user, not per subscription. New accounts get a trial of the paid experience, and choosing or changing a tier gives a fresh trial window before billing starts — but the total is tracked against you, not the account. Deleting an account and making a new one does not mint new trial days.
• Extra profiles beyond the two free ones are a paid add-on, shown and confirmed before anything changes, and removed from the billing automatically when you delete the extra profile.

Cancelling locks the paid features when the paid period lapses; it does not delete your data. If a charge does not match what you expect, use the in-app feedback button with the account e-mail — billing is checked against the payment provider's own records, and that is not something to post publicly.

Almost certainly an app version behind rather than a fault, and the fix is a Play Store update.

The newer BMC G3-platform cards — that is the Luna G3X and recent BMC G3 firmware — use a card layout that older builds of the Android app did not recognise. Support for it landed in app version 1.12; anything older sees an unfamiliar folder, declines to guess, and offers the DropBox review route instead. The web app has recognised the layout for longer, which is why the same card behaves differently on a laptop.

What to do:

1. Update the app from Google Play, then upload the card again. De-duplication means the retry costs you nothing.
2. Or upload the same card from the web app in the meantime — it is the same import pipeline and the same profiles.

Two things worth knowing while you are there:

• When you pick the folder, Android asks you to grant access to it. Grant it at the card root, not at a folder inside the card, or the picker hands the app only part of the data.
• If the card does not appear in the picker at all, that is usually the reader rather than the app: look under the reader's own entry in the picker's side menu.

They look at the same night for three different things, and they do not compete.

SomniScan — "did the machine miss something?" Sub-threshold event detection. Breathing disturbances that did not meet the device's scoring criteria — too short, or the airflow did not drop quite far enough — but still interrupted the night. Long flow drops are additionally tagged by whether the device flagged the same moment or not; a notable count of drops the device did not flag suggests the device is under-reporting, which is an observation to take to a clinician, not a diagnosis. Zero findings on a clean night is a normal, good result, not a failed scan.

SomniPattern — "is the breathing cycling?" Periodic-breathing detection: repeating waxing-and-waning patterns across the night, reported in neutral language with when and for how long.

Somni-UARS — "is there resistance without scoreable events?" The upper-airway-resistance angle: the pattern where the airway narrows enough to fragment sleep without producing apneas or oxygen drops big enough to be counted.

Practical notes:

• All three need imported high-resolution flow data for that night. A night without waveforms cannot be deep-analysed and will say so.
• The last seven nights are scanned automatically in the background, so recent nights are usually instant; older nights compute on demand and take a few seconds.
• Their findings feed the AI analysis automatically — you do not have to run them first and paste anything anywhere.
• They are observations, not diagnoses, and they are deliberately written that way.

Nothing is failing. That is the shape of the answer by design, and it is worth knowing why so the expectation matches the tool.

A wide-scope answer is built as a night-level narrative for the most recent night, plus a trend block for the period. The trend block is deliberately compact: an average, a direction of travel, the event counts that moved. It is a trend summary, not a night-by-night table — a 90-row table restated in prose would be less useful, not more.

On Week and wider you also get a "7-day view" paragraph drawn from the automatic background scan of your last seven nights, always with a coverage line ("6 of 7 nights scanned") so you can see what it is speaking from. If the background scan has not finished, it says so rather than guessing — ask again a minute later.

If what you actually want is per-night numbers over a period, the analysis panel is the wrong tool and there are three right ones:

• the Statistics page for tables,
• the Overview charts / Trends tile for the shape of a period,
• the Reports page for something printable to take to an appointment.

One honest limitation, stated because the tool states it too: pattern aggregates wider than the 7-day window are not built yet, so a "90-day tally" of the deep-analysis engines is not something it can produce today. It will tell you that instead of inventing a number.

Working as designed, and the design is deliberate: nights run noon to noon.

A session that starts after midnight belongs to the previous evening's night. Go to bed at 01:00 on the 15th and the night appears under the 14th — because that is the sleep that started on the evening of the 14th. Every CPAP data tool has to make this decision, and the alternative (splitting one night across two calendar days at midnight) makes a mess of both of them: two half-nights, two usage figures, two AHIs, none of them meaning anything.

Practical consequences:

• A shift worker who sleeps 09:00–16:00 gets that sleep split at noon — that is the one case where noon-to-noon is the wrong boundary, and it is a known limitation rather than a fault in your data.
• Two sessions in one night (mask off at 3 a.m., back on at 4) stay on the same night and are shown as separate sessions of it, not as two nights.
• Usage hours for a night are the sum of that night's sessions, which is why a night with a long break can show good total hours and a fragmented chart.

If a night is genuinely absent — not shifted, absent — the usual causes are the ones from the import threads: the card had not recorded it yet, or it went into a different profile.

Two different guards, two different answers.

"Another upload is already in progress in a different tab." The app allows one upload at a time per browser, so that two uploads cannot race each other into the same profile. If a tab crashed, was killed, or was closed mid-upload, the lock can outlive the tab that took it. It expires by itself — wait about ten minutes and it clears. If you are in a hurry: close every SomniCharts tab, then open one. Do not clear cookies or site data to get around it; you will just sign yourself out.

"SomniCharts is open in another tab" (full screen). Different rule: only one tab per browser may be the active one, so two tabs cannot fight over the same session. Click Use SomniCharts here and this tab takes over — the other one blocks itself instead. One caution: if an upload really is running in the other tab, taking over warns you first, and taking over mid-upload interrupts that upload.

And the case that looks like a bug but is not: the browser asking "Leave site? Data you entered may not be saved" during an upload. That is the app protecting an upload in flight. Click Cancel — stay on the page — and the upload carries on. Only leave if you actually want to abandon it.

That message means de-duplication worked: every file and every session on the card you just uploaded is already in this profile. It is a success message, not an error — but if you genuinely expected new nights, one of three things is true.

1. The wrong profile was active. This is the common one. Data imports into the profile named at the top right, and a card uploaded into the wrong profile is a duplicate there and missing here. Switch profile and upload the same card again — nothing is lost, and nothing is duplicated.

2. The card is not the card you think. Two machines, two cards, one drawer. Or the card was swapped and the new nights are on the other one.

3. The machine did not write the nights you expected. A card that was out of the machine, full, or unreadable to the machine records nothing. The machine keeps a limited number of nights of detailed data on the card and only summary data further back, so a card left out for weeks and then re-inserted has a gap that nothing can recover.

Diagnosing it is quick: look at the calendar in the profile you uploaded into and find the most recent night that has data. That date is where the card's history actually ends.

Re-uploading is always safe. De-duplication runs at three levels — file content, upload manifest and session start time — so uploading the whole card again after a holiday is the right habit: only the new nights import.

And the practical division of labour, because it saves everybody time:

• Something in the app is broken, or your account or billing needs looking at — use the in-app feedback button (My Feedback in the right rail), not a public thread. That panel is attached to your account, so we can actually look at the right data. A forum thread is not, and we will not ask you to paste account details here.
• Something about the app confuses you, and the answer is useful to other people — SomniCharts, and the answer stays where the next person can find it.
• Something on this board is broken or annoying — Site Feedback.
• Something about therapy, machines, masks or the numbers — Community, or the reply box under the relevant Library topic.

Select the top-level folder of the card itself — the whole card. Not a folder inside it, and not individual files. The importer walks the card and finds what it needs.

What it is looking for, by brand:

• ResMed (AirSense / AirCurve 10 and 11, S9) — the card root that contains DATALOG, SETTINGS, STR.edf and the identification files.
• Philips Respironics (DreamStation, System One) — the card containing the P-SERIES folder. The card root works; so does P-SERIES itself.
• Löwenstein prisma — the folder prismaTS produced. DCM folders and dcm.zip archives both import as they are; you do not have to unpack the zip.
• BMC / Luna — three different layouts, all supported, all selected the same way: the card root. Recent G3-platform cards hold files named after the machine's own serial number (an index file, an events file, a settings file and numbered waveform files); older Luna G2S cards use a slightly different events file; Luna II and Hoffrichter POINT cards carry THERAPY/RECORD folders instead. Extra items you may see next to them — language-pack files, a System Volume Information folder — are normal and ignored.

Picking the wrong thing does not break anything. If the folder is not a card the importer recognises, it does not guess: it offers the DropBox route instead, which sends the folder to the team for a look rather than importing garbage. And if you upload a card you have already imported, de-duplication skips everything it has seen before — worst case, nothing new arrives.

Two things worth checking before you press Upload, because they are the two that actually cause grief:

1. The right profile is active — data always imports into the profile named at the top right. Check it before, not after.
2. One upload at a time per browser. Wait for the upload to finish; the import then runs on the server and you can close the tab, log out, or go to bed.