Account

What's new

Every notable change, newest first. Releases are calendar-versioned — the number in the header of every page tells you which one you're on. You're running v2026.08.

2026.08

August 2026you're hereShare
New

the phone does the work now, not just the reading

  • The iPhone/iPad app could show you an aircraft and record a handful of things; anything that needed *correcting* sent you back to a laptop. It doesn't any more. Every screen that shows you a number now lets you fix it, offline, with the change visible before anything reaches the server.
  • Review a scanned page on the phone. Entries appear beside the scan (in a drawer you swipe up on the phone, in the pane beside it on an iPad). Fields the extractor was unsure of carry a Check this chip; tap the ◎ beside a value and a ring lights the exact ink it was read from, and stays lit while you edit. Confirm N clean accepts everything the model was confident about in one tap, the same rule the web reviewer uses. Entries the extractor missed can be keyed in, and an entry that runs onto the next page can be merged into the one before it — with the equipment, AD and document links following it across.
  • Inspections, directives and equipment are managed, not just listed. Add an inspection or edit its interval, mark any item done, or set up the standard Part 91 items on an aircraft that tracks nothing. Track an AD and record compliance against it with the date, hours, method and the entry that proves it. Record an installation, mark a component removed on a date — and the directive that only applied because of that component turns itself off, with the reason written down.
  • A VOR check still writes a real 91.171(d) entry, and a marked-done item's counter resets before the sync, not after.
  • Documents go in from the phone — from Files, iCloud, the camera, or by dragging a PDF out of the Files app onto an iPad. They queue on the device if there's no signal, cap at 25 MB each, and can be attached to the log entry they belong with.
  • Squawks are resolved where you find them. Swipe one right to resolve it, name the entry that cleared it, reopen it, edit it, or delete it. Resolved squawks keep their own openable list.
  • Ask your logbook works from the phone and the iPad, with the entries the answer came from listed beneath it and the day's remaining answers on the composer.
  • Add an aircraft from the phone. A tail number brings back make, model, serial and registrant from the FAA registry; a tail the registry doesn't know can still be added by hand rather than dead-ending.
New

the iPad stops pretending to be a large phone

  • Full screen, an iPad now shows a 200pt sidebar — tail, type, current status, the four tabs, Ask and Account — and two panes: the verdict beside every tracked item, a page beside its entries, a squawk beside its detail. Scans splits three ways, with the page rail, the scan and the entries all readable at once.
  • A Magic Keyboard drives it: ⌘1–4 for tabs, ⌘K for Ask, ⌘N for a new squawk, ⌘←/⌘→ to turn pages, ⌘↩ to confirm. Inside a text field ⌘← still moves the caret.
  • Split View falls back to the phone layout below 700pt, so an iPad beside ForeFlight is the app you already know rather than two slivers.
New

light appearance, and a sign-in that survives a relaunch

  • The app was dark only, which is wrong on a ramp in daylight. Appearance in the account menu offers Light, Dark, or Match my phone — and it repaints while the app is open rather than waiting for a relaunch.
  • Your session now lives in the iOS Keychain rather than webview storage. Force-quitting, installing a new TestFlight build, or opening the app in airplane mode no longer drops you at a sign-in screen you can't complete without signal. Existing testers are moved across once, silently.
  • Signing out now clears the device. The local copy of your records, the sync position and anything still queued go with it, so a shared iPad doesn't open onto the previous owner's fleet.
New

due and overdue arrive as a notification

  • The daily check that emails you before something comes due now also sends it to the phone, grouped per aircraft the same way the digest is: *"N4321J — 2 items coming due"*. It's the same once-per-due-cycle rule, so running the check twice in a day sends nothing twice.
  • Turning notifications off in the app removes the device, and says so if it couldn't.
Fixed

a change made in two places no longer picks a winner quietly

  • The phone's writes now carry the version of the row they were made against. If someone edited the same entry, reading, squawk or maintenance item on the web while your phone was offline, the upload stops and shows you both versions — yours and theirs, with the differing lines marked — instead of one silently overwriting the other. Decide later leaves it queued rather than forcing a choice on the ramp.
  • Anything the server refused keeps its reason in Waiting to upload until you deal with it. Documents still on the device are counted there too, so the screen can no longer claim everything has reached the server while a POH sits in the queue.
  • A viewer's edit used to report itself saved and change nothing. Every write now reads its own effect back and says *"You don't have edit access to this aircraft"* when it has none.
  • Correcting a reading, marking an item done, or resolving a squawk went through one code path on the web and another on the phone. There is now one implementation per write, which is why the two agree.
Fixed

inspections due on airframe hours now count down

  • An item due on airframe time showed no hours countdown unless someone had written an airframe total in the entry. Most logbooks note "TT" only occasionally and a tach reading every time, so in practice those items counted down on nothing.
  • Where a logbook's total time and tach have never disagreed — checked against every entry that recorded both — the tach now fills in the total, and airframe items count down like any other. Values derived this way are marked as estimated wherever the app already distinguishes a measured reading from a derived one.
  • If any entry recorded both and they differ, nothing is filled in: that aircraft keeps the two apart deliberately, and its airframe items behave exactly as before.
New

airframe pages show AFTT, and you can sort by it

  • Pages in an airframe logbook are now labelled AFTT (airframe total time), which is what an airframe book is kept in.
  • AFTT joins upload order, entry date and tach in the sort menu, and can be applied as a logbook's stored page order like the others.
  • Most airframe books never record a separate airframe total — the tach is the instrument that total is kept in — so where none was written down, the page's tach reading is shown as its AFTT, and sorting by AFTT uses it. Without that, sorting an ordinary airframe book by AFTT would send every page to the bottom as a blank.
  • Maintenance countdowns are unaffected and stay stricter: an item due on airframe hours still shows no countdown unless an airframe time was actually recorded.
Fixed

a page said "✓ reviewed" the moment it was extracted

  • The badge reads how many entries are still unconfirmed, and the pages list forgot to update that count when an extraction finished — so a page nobody had looked at reported itself reviewed, and dropped out of the Needs review filter.
  • Nothing was ever actually confirmed; the entries were in the review queue the whole time. A reload showed the truth. The count now comes from the server.
  • The same fix corrects the entry count on a re-extracted page, which previously reported only the newly-read entries — visible in the delete button's warning.
New

select several pages and delete them at once

  • Tick pages in the list, or Select all shown, and delete them in one action. The confirmation names both the pages and the entries that go with them. Previously a mis-uploaded batch meant one confirm click per page.
  • Selection is limited to pages currently visible, so a filter or a logbook choice can never hide something from what you are about to delete.
New

see your daily AI allowance before you hit it

  • The profile now shows calls used against the daily cap, for everyone. Usage was previously visible only if you had added your own API key, so anyone on the shared allowance met "Daily AI limit reached" with no warning.
  • It also says what the number counts: a page takes about 2 calls to read, so the shared allowance is roughly 50 pages a day, and calls free up individually 24 hours after each one — not all at once at midnight.
Fixed

searching Documents on iOS appeared to do nothing

  • The field filtered only the Everything else list, and that list is defined as everything *not* in AROW — so searching for a registration or an airworthiness certificate, the four documents people reach for most, could never return them.
  • A query now searches the whole vault and shows one list of matches with a count. A query that matches nothing says so, instead of silently removing a heading.
  • Titles, type labels, reference numbers, and file names are all searchable, and several words narrow the result rather than having to be typed in order.
  • Added a clear (×) button, and Enter now dismisses the keyboard covering the results.
Improved

the scan sits beside its entries when you review everything at once

  • Review all now shows each page's scan pinned next to that page's entries, the way the single-page reviewer always has. Checking a transcription used to mean opening the 48px thumbnail full-screen, reading it, closing it, and repeating for the next entry.
  • Tapping a field's ◎ there now spotlights exactly where on the scan that value was read from. The affordance already existed but had never been reachable from this view.
  • Imported entries with no scan behind them keep the full-width single column.
Fixed

tapping Search in Documents zoomed the iOS app

  • Same cause as the squawk composer below: a 13.5px field, under iOS's 16px threshold for form-control zoom. Every editable control in the app is now at or above it.
Fixed

opening New Squawk zoomed and horizontally displaced the iOS app

  • Focusing the 14px description field triggered WKWebView's native form-control zoom. The modal then appeared wider than the screen, and dismissing it left the entire app horizontally pannable until restart.
  • The field now uses iOS's 16px no-zoom threshold. Earlier overflow and width constraints treated the visible symptom but could not stop the focus zoom.
  • Verified on-device in TestFlight and released as iOS 1.3 build 202608280002.
Improved

mobile writes sync when connected

  • iOS now sends meter readings, oil additions, squawks, maintenance completions, and scanned pages immediately when connected. Writes still land in SQLite first for crash safety; offline work remains queued, appears in the pending notice, and retries when connectivity returns.
New

owner guides and private growth measurement

  • Added practical public guides for digitizing aircraft logbooks, maintenance tracking, and AD tracking.
  • Added an RSS feed and stable share links for every What's New release.
  • Added an admin-only activation funnel derived from existing product records, plus one-time summary share/export milestones. No page-view tracking or third-party analytics.
New

shareable maintenance snapshot

  • Share summary uses the device's native share sheet for a compact text snapshot of status, due items, meters, and weight & balance. It omits scans, serial numbers, reporter names, and squawk descriptions; browsers without a share sheet copy the same redacted snapshot instead.
Fixed

Fixed

  • A deleted aircraft now disappears from the phone. change_log was written so hard deletes reach the offline app, and the tombstones were being recorded correctly — they just weren't *readable*, because the row-level policy checked access by looking up the aircraft that the delete had removed. The one row announcing a deletion was hidden from the one device that needed it. Deleting a page or an entry always worked, which is why this went unnoticed.
  • Existing phones repair themselves once. The sync feed only moves forward, so a device that passed an unreadable tombstone could never learn of it by syncing again. The local copy is therefore dropped and rebuilt from the server a single time on upgrade, and "Rebuild from the server" now sits in the account menu for any future drift. Anything recorded but not yet uploaded is kept.
Improved

the iOS app is rebuilt around "can I fly today?"

  • A verdict comes before any list. Status now opens with one countdown ring and one sentence — "One item due soon", "Grounded — annual overdue" — then the single item that needs doing, then everything healthy collapsed into one row. Seven items used to render as near-identical cards, so an engine overhaul due in 2038 competed for attention with a VOR check due in 26 days.
  • A tab bar replaces the Back-stack. Status, Log, Records, Squawks. The tail number in the header switches aircraft in place, so the fleet stops being somewhere you have to back out through.
  • Records is one tab, not three buttons — Documents, Scans and History behind a segmented control, with scanning moved in beside the pages it produces.
  • Scanning is a sheet that already knows the aircraft. It used to ask which tail you were scanning — from inside an aircraft whose header was showing the tail number while the question was on screen. Now it opens over the Scans grid, asks only which logbook, and says where the pages land ("Adds to the end — 24 pages so far"). It no longer says "queued", "pending" or "routes to vision extraction": pages are "saved on your phone" and "upload on the next sync".
  • Log a flight has steppers, shows the delta as you go ("+2.4 hours this flight"), and commits meters *and* oil with one Save to logbook. It used to be two buttons that said "Queue". Either meter can be left out — only one is required, and recording a number you didn't read would claim that meter still shows its old value today. Oil takes any amount, so a half-quart top-up is a half quart rather than being rounded into a bucket.
  • Squawks lead, the composer is a sheet. It used to sit above the list, so opening the keyboard hid the squawks you were checking. Severity now reads Low / Watch / Ground — "high" doesn't tell you whether the aircraft flies.
  • Maintenance history is readable. Entries get a written title instead of shouted imported text, a one-line summary, and the same job recorded by two sources collapses into one marked "2 merged".
  • Scans are grouped by logbook with page numbers, instead of a hundred identical thumbnails; the page viewer swipes and has a thumbnail strip instead of two buttons that took a third of the screen.
  • Documents lead with a Carry aboard card that answers "is my AROW current?".
  • One warm dark palette and two brand faces throughout; monospace is retired.
Security

`unsafe-eval` is gone from the production Content-Security-Policy

  • It was only ever there for OpenCV.js, which capture no longer loads. Removing it means a script injected into a page can no longer reach eval() or new Function() to build code at runtime.
  • Kept in development only: the dev server compiles modules through eval() for hot reload, and these headers apply in dev too — dropping it everywhere looks like a pure hardening win while quietly breaking local development.
  • pdf.js needed no change; version 6 doesn't use eval at all.
New

crop and clean up a page after the fact

  • A page photographed on a desk comes with the desk. The review screen now has Crop & clean up: drag the corners to cut out the background, rotate, and switch on a scan look for a black-and-white document finish rather than a snapshot. (On iPhone the app now uses Apple's own scanner and does this as you shoot — this is the web equivalent, since a browser has no VisionKit.)
  • It runs once, on an image you already have. That's the whole difference from the live auto-crop this replaces, which did the same work on every preview frame and made capture unusable on a phone.
  • The contrast pass is deliberately mild: hard thresholding looks more "scanned" but eats faint pencil, which is most of what a logbook is.
  • Editing replaces the stored scan (the paper remains your legal record), and says so. Entries already extracted came from the old image, so it prompts you to re-extract if the crop changed what's readable.
New

capture straight into the logbook you're looking at

  • Tap a logbook on Aircraft → Pages and there's now a Capture into <logbook> button: the camera opens with that book already selected. Previously you left the logbook you were browsing, went to a separate Capture page, and picked it again out of a dropdown — easy to get wrong when the airframe and engine books look alike.
New

put a logbook back in order in one step

  • Uploaded the second volume before the first? Open that logbook's pages, sort by Entry date, and hit Save this order — it renumbers the whole logbook to match what you're looking at. Previously the only way to reorder was nudging one page up or down at a time, which is fine for a stray page and hopeless for fifty of them.
  • Offered only when it's a fact rather than a guess: one logbook in view, and every page in it already extracted. An un-extracted page has no date, and sorting would quietly dump it at the end and renumber the book around it.
  • Reordering a large logbook used to be one database round trip per page, which made a 150-page book a minutes-long wait. Now batched.
Improved

the iOS app now scans with Apple's own document scanner

  • Scanning uses VisionKit — the scanner Apple Notes uses. Automatic edge detection, perspective correction and the black-and-white document look, done natively and instantly. A beta user pointed at Notes as the thing to match; it turned out to be literally available.
  • You can correct the crop before keeping a page, in Apple's own editor — which is the part the old in-browser auto-crop never offered.
  • Up to 24 pages in one session, so a whole logbook goes in without reopening the camera between pages.
  • Under the hood this needed Capacitor 6 → 8 (no document scanner supports 6). Nothing in the app's own code needed changing for it, but it is a native rebuild, and the on-device database library moved two majors with it.
Fixed

you couldn't get rid of the demo aircraft, and phone page lists had no dates

  • The demo aircraft had no exit. It's auto-shared read-only with every new account, so you're a viewer rather than its owner, and nothing in the app let a viewer drop their own grant. Every aircraft shared with you — the demo included — now has Remove from my dashboard. It removes your access and nothing else: not the aircraft, not a single record, and not anyone else's access. Whoever shared it can share it again.
  • Page lists on a phone showed no dates. The date and tach column is hidden below 640px for width, which left the list undated even though it can be *sorted* by date. They now fold into the line under each page instead of disappearing.
Fixed

the hobbs→tach estimate could be anchored on a mis-keyed reading

  • Still a wrong burn rate after the fix below: 666 hrs/qt from two ordinary top-offs. The estimate that converts a hobbs-only top-off into tach was built from *raw* readings, so a mis-keyed entry with the same number typed into both the hobbs and tach fields counted as a real pair — anchoring the conversion at hobbs == tach and throwing it out by thousands of hours.
  • normalizeReadings() exists to discard exactly that, and every other hours calculation already ran through it. The converter now lives with the rest of the meter maths, on the same normalized, meter-reset-stitched readings, so no caller can skip the step.
Fixed

oil burn rate could be computed across two different meters

  • A top-off logged with tach only and the next logged with hobbs only were subtracted from each other. The meter was picked per row, so ~4141 (tach) and ~965 (hobbs) ended up in the same series: 19 hours of flying on one quart was reported as 3176 hours and 453 hrs/qt — and attributed to the wrong date and the wrong quantity, because sorting on the mixed scale reversed the order too. Every part of that error read reassuringly, which is the wrong direction to be wrong about oil consumption.
  • The whole trend is now measured on one meter: tach when it can carry the series (it's the engine-time meter oil burn actually follows), hobbs only when tach can't. The chart says which meter it used — the same aircraft looks healthier measured on hobbs, because hobbs runs on the ground.
  • A hobbs-only top-off is now bridged into tach rather than dropped, using this aircraft's own *measured* hobbs↔tach ratio, so the trend stays on the meter oil burn actually follows and nothing you log goes missing from the chart. The estimate is derived when the page is read and never stored — log a real tach later and it replaces the estimate by itself. A generic default ratio is refused: that's a constant, not your aeroplane.
  • A top-off that still can't be measured is counted and explained instead of silently vanishing from the chart, and any interval resting on an estimate is marked as one.
Fixed

camera capture on a phone was unusable

  • Reported from the field: on an iPhone the in-browser camera was glitchy, often wouldn't capture at all, and when it did it took far too long. It was not the phone. Capture was pulling a 9 MB, ungzipped OpenCV.js build from a CDN before it would work, running a full edge-detection pipeline on the main thread twice a second to draw the live outline, and then — on the shutter — detecting and perspective-warping at the camera's full resolution synchronously, freezing the page for seconds. (The library it used also leaks a matrix on every call, which iOS Safari is unforgiving about.)
  • Capture now opens your phone's own camera app. You get its autofocus, HDR and stabilisation, the shot is instant, and the photo is generally *better* than the frame-grab it replaces. Nothing is downloaded and nothing is processed on the critical path; a captured page now takes the exact same route as an uploaded one.
  • Auto edge-detect, deskew and crop are gone with it. They only ever worked when that 9 MB download succeeded, which on a phone often it didn't — and the extractor reads the page out of a plain photo perfectly well. Fill the frame and hold steady; it doesn't need to be square.
  • Consequently the app now loads no third-party scripts at all. The OpenCV and jscanify CDNs are removed from the Content-Security-Policy, which is the whole external script surface gone.
Fixed

the top bar and the meters page disagreed about your hours

  • Two different implementations of "current hours" existed. The aircraft shell (the top bar on every aircraft page) hand-rolled its own "newest date wins" pick, and it never even *selected* the source column — so it could not tell an accepted ADS-B estimate from a MyFlightBook reading. On N9363V it showed 964.4 (an estimate) directly beside a provenance line reading *"as of 2026-08-11 · from MyFlightBook"*, where the real MyFlightBook value was 965.1. The number and its own explanation were coming from two different code paths.
  • The duplicate is deleted. The shell now uses the same toReadings() + currentMetersFrom() the meters, status, maintenance and compliance pages — and the iOS app — already use. Face values (what the instrument physically reads, after a meter replacement) are still distinct from stitched total time; that conversion now happens once, in one place.
Fixed

an ADS-B estimate could outrank your own recorded hours

  • An accepted ADS-B estimate kept beating a later MyFlightBook sync of the same flights. Reported on N9363V: the app showed 964.4 hobbs (the estimate) while MyFlightBook had 965.1 for those flights. Tach was right, which is what made it visible — that estimate carried no tach, so it never competed there.
  • The principle was always "your own records win; ADS-B is only a fallback observer", but that was enforced only where suggestions are *raised*, never where current hours are *chosen*. The estimated flag existed and was set correctly; the meter-selection code simply never looked at it.
  • The tie-break made it systematic rather than unlucky: on a shared date the reading closest to the previous value wins, and ADS-B airborne time under-reads by construction (it excludes taxi and runup) — so the estimate was reliably the closer one. A rule meant to reject outliers was quietly picking the guess over the measurement, every time.
  • An estimate is now discarded as soon as any real reading reaches or passes it: meters are cumulative, so a measured value already contains whatever the estimate was guessing at. An estimate above every real reading still counts — that's the case it exists for.
New

The iOS app can now tell you if you're legal, and record what you did

  • Status, computed on the device. Current tach/hobbs with provenance, then every maintenance item and recurring AD worst-first — offline, in the hangar, where there's no signal. The aircraft list flags the worst one so a problem is visible before you open anything. None of the airworthiness math is a second implementation: the app imports the *same* pure compliance code the web runs, because two copies is how a phone and a website start disagreeing about whether an annual is due.
  • Record it there and then — meter readings (pre-filled, and it warns when the value is lower than the last one, which is either a meter swap or a typo), oil added, squawks, and marking a recurring item done. All queue on device and upload on the next sync.
  • The VOR check writes a real record. 91.171(d) wants the place, the bearing error and a signature, so the app asks for them and writes a log entry as well as resetting the counter. A tick-box that only moved a due-date would leave you non-compliant while telling you that you were fine.
  • AROW documents offline — airworthiness certificate, registration, POH/AFM and weight & balance pinned together for a ramp check, each either present or explicitly missing. PDFs open in the app, page by page: they used to say "open on the web app", which is the wrong answer on a taxiway, and a registration or airworthiness certificate is usually a PDF.
  • Nothing recorded disappears quietly. A "Waiting to upload" list shows what hasn't reached the server and keeps the reason on anything refused. Retries are safe: every queued action carries an id that becomes the server row's key, so a re-send after a dropped connection writes nothing the second time.
Fixed

ADS-B passive hours never actually ran

  • The sweep had not made a single successful call since it shipped. Every daily run failed with a 10-second timeout, for every opted-in aircraft, and no flight was ever recorded. The cause was not our code: OpenSky blackholes Google Cloud egress. Measured from a Cloud Run job in us-east4 *and* us-central1, DNS resolves but the TCP handshake to their host never completes (connect=0.000000s, dropped rather than refused), while api.github.com answers in 27 ms from the same container. Raising the timeout could never have fixed a blackhole.
  • The sweep moved to a GitHub Actions job, which reaches OpenSky in under a second. The runner holds no database credentials: it asks /api/cron/adsb which aircraft to look up and what windows to ask for, then posts the results back for the server to write. Opt-in is re-checked at write time, so switching ADS-B off mid-sweep still means no rows.
  • Overlapping observations no longer inflate the estimate. OpenSky emits more than one record for a single flight when receiver coverage breaks up — the real data that surfaced this had a 23-minute segment sitting inside a 72-minute one. Summing them reported 3.4 h for 3.0 h of flying, on the very number we suggest adding to a tach. Overlapping spans are now merged and counted once.
  • /help now says when the check runs (17:30 UTC, daily) and that it looks back three days, so a flight this afternoon is expected tomorrow rather than tonight.
Improved

A landing page that says what the thing does

  • The front page listed six features; the app has closer to thirty, and the ones people actually pick it for — that it is free with no billing code in it at all, that it is MIT-licensed and self-hostable, that extraction is automated and self-serve rather than a transcription service you mail your books to, that it answers questions instead of just storing files, and that it backs itself up to a Dropbox or Google Drive *you* own — were absent altogether. They are on the page now.
  • It opens with a scenario rather than a slogan, and sets data the way the app does: tail numbers, tach readings, AD numbers and dates in the mono instrument face, inline in the prose.
  • All six product screenshots are used, each at full width under the paragraph it illustrates, instead of four rotating in a carousel that showed one at a time. They are captures of dense UI, so a half-column reduced most of them to a dark smudge; at full width every one is readable. The carousel component is gone.
  • Starting is a primary action again. Account creation is the main call to action for signed-out visitors (/login handles sign-in *and* sign-up); signed-in visitors get a route straight to their hangar in the same places.
  • The other public pages are properly reachable — FAQ, How it compares, Coming from MyFBO, Help, the API docs and What's new are in a top bar and a footer, rather than one line of small print. The same two additions were made to the shared marketing footer.
  • It says what the product does not do, in its own section: no accuracy percentage (nothing measures one — you get per-field confidence and the scan beside the entry), no scheduling/dispatch/invoicing/billing, iOS is a TestFlight beta with no Android app, and CSV import is CSV only. The index-not-the-legal-record notice (14 CFR 91.417) stays.
New

Import a CSV

  • Bring maintenance history in from a spreadsheet or another platform, without printing it to PDF first. Pick the logbook, upload the CSV, and the entries are created directly.
  • The columns are mapped, not the rows. There is no importer per vendor and there won't be: one AI pass reads your header plus a few sample rows and proposes what each column means — "Tach Out" → Tach, "A&P" → Signature, "Invoice #" → don't import. You confirm that once, and every row is then converted in plain code. The same file always imports the same way, and a wrong import is explainable by pointing at the mapping rather than at a model.
  • Dates are never guessed. 03/04/2026 is 3 April or 4 March depending on who exported it, and getting it wrong would shift a maintenance date by up to eleven months — which then drives annual-due, 100-hour and AD compliance. The whole date column is scanned and the first date with a day past the 12th settles the reading for the entire file; you're asked only when *every* row genuinely reads both ways, and then you're shown what the first few dates become each way. A column that's internally inconsistent is reported as a broken file rather than quietly coerced.
  • You see the count before anything is written, along with every row that can't be read and why. An unreadable date or an implausible tach fails that one row instead of being imported as a zero, and never takes the rest of the file with it.
  • Imported entries land unconfirmed and show up in Review all grouped as "imported (no scan)" — a foreign spreadsheet hasn't earned the right to drive a reminder or a forecast until you've looked at it. Importing into an aircraft that already has entries is the usual way to create duplicates, so the finish line points at Fix duplicates.
  • Reads what spreadsheets actually emit: comma, semicolon and tab separated files, a UTF-8 BOM, CRLF, and quoted fields containing commas or line breaks. CSV only (save an XLSX as CSV in one step), up to 5 MB / 5,000 rows.
New

Cloud backups to Google Drive

  • Google Drive joins Dropbox as a backup destination, and you can connect both at once — each has its own cadence, schedule and history, so a problem with one doesn't take the other down with it. That's the difference between a backup and a second copy.
  • We ask for the narrowest Drive permission that exists: access to *files this app creates*, and nothing else. Your existing Drive is invisible to us by construction, not by promise. Backups land in a MyTailLog folder, nested per aircraft, and we only ever *add* — nothing is renamed, replaced or deleted.
  • Uploads are resumable: a dropped connection part-way through a large archive picks up from exactly where the server actually got to, rather than starting again or — worse — quietly writing a truncated file.
New

Automatic cloud backups (Dropbox)

  • Your records back themselves up to storage you own. Connect Dropbox from Profile and the same re-importable .zip you can download by hand — records plus every original scan — is pushed to your account monthly or quarterly, one dated file per aircraft at MyTailLog/<TAIL>/<date>-<TAIL>.zip.
  • App-folder access only. We ask for permission to write files in our own folder and nothing else, so the rest of your Dropbox stays invisible to us — and we only ever *add* files. Nothing in your account is renamed, replaced, or deleted; retention is your call.
  • You can tell whether it's working. Profile shows the last run, its result, and its size, and you get an email if two runs in a row fail — a backup that quietly stopped six months ago is worse than none.
  • Very large aircraft are reported as too large to upload rather than failing mysteriously, with a pointer to the manual download.
New

AD discovery by model

  • Find ADs by your actual variant, not just your make. The AD explorer now searches by model and by free-text keyword alongside the existing manufacturer-wide search (both run — the make is the broad net, the model is the sharp one).
  • Every result lists the models the AD names, parsed from its title and summary, with the ones covering your model highlighted; those results sort first. You can see whether your variant is actually named instead of guessing from the manufacturer.
  • Pre-1994 legacy ADs now surface. Model and keyword searches also query the FAA's Dynamic Regulatory System, whose archive reaches back past the Federal Register's 1994 start. If either source is unreachable the other still returns.
  • "Track this AD" in one click, or with a recurrence: one-time vs recurring, an interval in hours and/or calendar months, and a next-due — which feeds the maintenance forecast and the Status grid like any other tracked AD.
  • Search results remain a starting point, not a determination: the parsed model list can be incomplete, and applicability often turns on serial numbers. A scanned A&P AD compliance report is still the ground truth.
New

Maintenance due dates you can plan around

  • "Due in 38.4 hours" now also reads "due ≈ 14 Mar." Hours-based items — 100-hour, hour-interval ADs, oil, component TBO — are projected onto a calendar date from how much you actually fly, using your own logged tach readings.
  • Every projection carries a confidence (from how many readings back it and how far apart they are) and shows the window it was computed over. Below the threshold you see hours only, exactly as before — one reading never becomes a forecast.
  • Intervals spanning a meter replacement are excluded, so a tach swapped in at airframe total can't inflate the rate. Projections are planning estimates, never a substitute for the record: a calendar limit like the annual is always the date it says.
New

Maintenance summary (print / PDF)

  • A one-page maintenance summary per aircraft — status at a glance, inspections and recurring ADs, open squawks, full AD/SB compliance, installed equipment, and current weight & balance. The document you hand a buyer, an insurer, or an IA at annual. Figures come from the same engine as the Status page, so the summary can't disagree with the app.
  • Every .zip backup now carries a `README.txt` manifest describing each file and column, so the archive still explains itself years later.
New

What's new

  • This changelog is now published at [/whats-new](https://mytaillog.com/whats-new), and the version chip in the header links to it.
New

Questions, comparisons, and getting your records out

  • New public pages: /faq (what it costs, what happens to your data, who can see it, what happens if the project stops), /compare (how six different ways of keeping records actually stack up, including where MyTailLog loses), and /switch/myfbo for owners whose platform is shutting down.
New

Attach documents to log entries

  • Link a Records Vault document to a specific log entry — a 337, an 8130-3, an invoice — from either review flow, and see the linked entry from the Vault side. Attachments show on the timeline and travel with the .zip backup.
Fixed

Fixed

  • Restoring a backup that contained an attached document failed outright. The importer remapped every other entry reference but not the document's, so the restore hit a foreign-key violation and rolled back the whole archive — not just the attachment.
  • Removing an attachment deleted the file. The only control on an attached document deleted it from the Vault instead of unlinking it. It now unlinks and the document stays.
  • Printing came out nearly blank. The design tokens are dark-only and browsers drop backgrounds, so printed pages rendered pale ink on white; the app shell also clipped print output to a single screenful.

2026.07

July 2026Share
New

Native iOS app (offline-first, beta)

  • Offline logbook in your pocket. A native iPhone/iPad app (Capacitor) that syncs an aircraft once, then works fully offline — browse every log entry, document, and original scanned page with no signal — and captures new logbook pages offline that upload when you're back online. Currently in TestFlight beta. Built on a self-hosted sync engine (a Postgres change feed → /api/sync/pull → on-device SQLite + a filesystem scan cache), no third-party vendor. See `docs/mobile-and-sync.md`.
Fixed

meter & status accuracy

  • Oil change no longer shows falsely overdue when its last-done was recorded in tach but it counts down on hobbs, and a stray/mis-keyed hours reading (e.g. a duplicate MyFlightBook value) can no longer hijack "current hours."
  • 100-hour inspection no longer shows falsely overdue — a normal gap between the maintenance date and the nearest hours reading is no longer mistaken for a meter mismatch, and the annual reset is preserved.
  • Backup export/import fixed — exports no longer swap pages and log entries (a restore of an old backup could fail); re-export to get a clean archive.
Improved

infrastructure

  • Blob storage moved to Google Cloud Storage (off Supabase Storage's free-tier egress cap; consolidated onto GCP). Access still gated by RLS through the app's serving routes.
  • Monorepo. The repo is now apps/web (this app) + apps/mobile (the iOS app) + packages/. No user-facing change.
New

Records, squawks & engine health

  • Records Vault — a categorized home for the aircraft's permanent records (airworthiness certificate, registration, radio station authorization, POH/AFM, weight & balance, STCs, 337s, 8130-3s, ICAs, manuals), stored alongside the logbook scans. Upload PDFs or photos up to 25 MB; a document can also be attached to a specific maintenance entry.
  • Squawks — pilot-reported discrepancy tracking. Anyone with access can report an issue with a severity (including a shared pilot); editors resolve, reopen, or delete. Open until a mechanic clears it.
  • Oil consumption — log each oil top-off ("added 1.5 qt") with the tach/hobbs and see your burn-rate trend (hours per quart) between top-offs — separate from the lab wear-metal analysis.
New

Open API & integrations

  • OAuth 2.1 API. MyTailLog is now its own Authorization Server + Resource Server (Panva oidc-provider, Authorization Code + PKCE). Third-party apps can read an aircraft's airworthiness / AD / inspection status, equipment, hours, oil, and weight & balance — read-only, and only with the owner's consent. Endpoints under /api/v1; RFC 8414 discovery at /.well-known/oauth-authorization-server.
  • Account-wide sharing. Consent defaults to sharing all your aircraft (including any you add later, so an app keeps working as your fleet grows), with "only the ones I pick" still available. A brand-new account can authorize an app before adding any aircraft (it just sees an empty list until you add one).
  • Self-serve developer portal (/developers) — register public (PKCE) or confidential (client-secret) apps, with docs at /developers/docs.
  • Connected apps in Profile — see and revoke any app's access at any time.
  • Bidirectional MyFlightBook — MyTailLog pulls your hobbs/tach *from* MFB, and MFB (or any consented app) can pull airworthiness *from* MyTailLog. Integration guide: `docs/mfb-integration.md`.
Improved

platform

  • Upgraded to Next.js 16 (Turbopack is now the default build; middleware → proxy), React 19.2, and TypeScript 6. ESLint moved to flat config (eslint.config.mjs).
Security

Security

  • MyFlightBook credentials moved out of browser reach. The per-user MFB OAuth client_secret and access/refresh tokens (already encrypted at rest) were readable as ciphertext by the browser role through row-level security, which scopes rows but not columns. They now live in a private schema Postgres doesn't expose, reachable only through SECURITY DEFINER functions — the same lockdown applied earlier to users' Anthropic keys. No re-entry or key rotation: the ciphertext and encryption key are unchanged. Any credential still stored as legacy plaintext (from before at-rest encryption existed) is now re-encrypted automatically on first use.
  • Fixed a critical cross-tenant authorization gap in the OAuth grant path: the per-aircraft grant now verifies aircraft ownership at both write and read time (RLS + app-layer + a read-time recheck), so a token can only ever read aircraft its owner consented to. The Resource Server authorizes every request explicitly (RLS does not apply to OAuth tokens).
  • Confidential client secrets are encrypted at rest (AES-256-GCM), same as MyFlightBook credentials and users' own Anthropic keys; pinned the GCM auth-tag length.
  • Added Semgrep and Dependabot to CI; SHA-pinned all GitHub Actions.
  • Full-app security audit hardening. Closed an AI-budget race (atomic reservation replacing a check-then-act), moved BYOK Anthropic-key ciphertext into a private schema reachable only via SECURITY DEFINER functions (a browser-role read was possible before), scoped form-action to the consent flow, patched sharp/libvips CVEs, gated the maintenance forecast to owner-confirmed entries, split the document table's write policy to editors-only (a read-only viewer could previously write documents), and added an executable RLS-isolation regression suite plus broad unit coverage.
New

Earlier in 2026.07

  • Oil analysis — import a Blackstone/AVLab lab report (PDF or photo); AI reads every sample and charts wear metals over time against the lab's universal average.
  • Find duplicates — flags likely-duplicate scans and entries (by date, tach, and work text) so re-captures don't pile up.
  • Bring-your-own Anthropic key with usage/cost transparency, and shared-key cost caps.

Follow updates in your feed reader through the RSS feed. MyTailLog is MIT-licensed and open source — the full engineering history is in the git log.