Repository navigation
Releases
ExplorInk GPS 0.3.14 -- on Google Play, and release-signed
ExplorInk GPS is on Google Play as of today:
https://play.google.com/store/apps/details?id=org.explorink.gpsbridge
What changed in the app. Before asking Android for background
location, the app now explains what it sends, where it goes and what
declining costs, and it explains it on both routes rather than only the
first time. A phone that declined once used to be sent to the settings
page with no explanation at all. Google Play requires that disclosure,
and a rider deciding whether to grant it has more use for it than Google
does.
Signing changed, and it breaks updates once. This APK carries our own
release key instead of the shared Android debug key every earlier build
used. Android refuses an update across a different signing key, so an
older sideloaded build has to be uninstalled first. Uninstalling loses
nothing the reader needs: the pairing lives in Android's companion-device
record, and recordings are files you can copy out first.
Play and APK are separate routes. An app published as a bundle is
signed on Google's side, so the Play copy and this APK have different
signatures and can never update each other. Pick one route per phone.
Built from e64abbf. Verified: signer SHA-256
0b500f72803471e2462c9dc8c973648fc9004055f5efe77a3dfd513d04a80f0a, the
same key the Play upload carries. This APK itself was not installed on a
handset; the same code was run as a debug build on a Galaxy S24 Ultra
against an Xteink X4 Pro, and the Play build of it is live.
ExplorInk GPS 0.3.13 -- doc-only: the name collision is per-model
Doc-only release: three code comments said every board shares the same
advertised BLE name. They do not -- only two boards of the same model
collide (confirmed on hardware 2026-11-08: an X3 and an X4 Pro side by
side advertised distinct names, XteinkX3Map and XteinkX4ProMap). No
behavior changed from 0.3.12.
ExplorInk GPS 0.3.12 -- Home shows the paired device, and stops guessing
Home now shows which X4 is paired, and the app never connects to one on
its own.
The device plate used to look identical whether the phone was paired
with the right board or a stale one from before -- nothing anywhere
named the paired MAC address. "Device resting" now carries a short
suffix of it, and the tap dialog leads with the full address. "Forget
device" was already there and already worked.
While checking that, found the other half of the same problem: unpaired,
the app connected to whichever board answered first, with no
confirmation -- live-demonstrated on hardware. Fixed the same way the
app's own pairing code already warned about: unpaired, the app only
notices that something answered ("something is advertising -- tap to
pick it") and leaves the choice to the same system pairing picker as
before. Tested with two boards advertising at once (an X3 and an X4
Pro): the picker listed both, picking one connected only to that one,
the other stayed untouched.
Reviewed before merge; the review caught a stale Forget-device warning
and a wording/color choice that broke two of the app's own design rules,
both fixed before this build.
Hardware-tested on an Xteink X3 and an Xteink X4 Pro.
ExplorInk GPS 0.3.11 -- the app and the device swap versions
The app and the device now tell each other what they are.
On connecting, the app reads the device's firmware revision and model
number off the Bluetooth SIG Device Information Service (0x180A) and
writes its own version back. Both appear as a device_version line in
the device sheet, so a support question answers itself: which build is
on the device, and which board is it.
Because the service is the standard one, nRF Connect and any other
generic BLE app shows the firmware version too, with this app not
installed.
It is an add-on and it cannot cost a connection. Every firmware
built before today answers neither half, and against those builds
nothing is even queued. Verified on hardware against a genuinely old
device build: connect, both tile channels subscribed, MTU negotiated,
position received and the map redrawn, unchanged.
Hardware-tested on an Xteink X3, an Xteink X4 Pro and a LilyGo T5 S3
Pro.
ExplorInk GPS 0.3.10 -- Map areas and Pins look like Home now
Map areas and Pins now match Home's redesign: a sub-screen app bar (back
arrow, no bottom Close button), card-style rows with the right icon, one
hairline divider between rows, section labels instead of ad hoc headings.
Covers everything merged since 0.3.9, which was step 1 of the same pass
(colors, typography, button hierarchy, the Home brand lockup).
Map areas:
- Each queued area gets a state badge derived from its own counts --
Queued, Sending, On device, Needs you -- the same fill-state language
the device plate already uses. - New action on a zone: "Open in Google Maps", so a rider can see what a
coordinate box actually covers before trusting it. - Dropped the zone row's progress bar: the "x/y sent" line already answers
the fraction, a bar under it only repeated the number. - The top status line is now live (lit while a phase is actively running,
grey while only waiting, gone when there is nothing to say) instead of a
static "map areas" headline.
Pins:
- The pin-kind picker is a real dropdown again -- its custom flat styling
had quietly thrown away Android's own dropdown arrow. - Every pin kind shows the same icon the firmware draws on its map
balloon (house, flag, dot, tent, heart, or the baked "P"/digit for
parking and #1-#5), not a generic pin, in both the picker and the
on-device list.
Also in this build: the in-app About screen (app bar overflow -> About) --
what the app does, the permission list with reasons, and the privacy
disclosure Google Play expects in-app for background location -- and the
mountain lockup on Home's hero.
An adversarial review of the Map areas/Pins diff caught two real bugs
before this shipped: the zone badge could never read "On device" for an
area with any sea/no-data square (it tested sent >= tiles; the app's own
definition of "nothing left to do" is sent + noData + stuck >= tiles),
and a disconnect mid-transfer could leave the badge stuck on "Sending"
because the repaint cache key did not see inFlightBytes cross zero.
Both fixed before merge.
Verified on a Galaxy S24 Ultra against a live LilyGo T5 S3 Pro: both
screens, the reachable zone badge states (Queued, Sending, On device --
the error state "Needs you" is code-reviewed, not hardware-observed), the
zone action dialog, the Google Maps intent, the pin-kind picker (all 11
icons, collapsed and open), and one full on-device pin save/delete
round-trip.
473 unit tests. Built from master at 2893370. Debug-signed.
v0.3.9
First visual design pass, step 1 of a staged rework against ExplorInk's
Outline design system: color, typography and button hierarchy, plus the
brand lockup on Home. No feature or protocol changes.
- Colors: the design system's eight-neutral-plus-two-signal token set,
light and dark, replacing the previous plain platform theme. - Typography: Instrument Sans (UI), IBM Plex Mono (every number/counter/
log line), bundled directly in the APK. - Buttons: four kinds (primary, secondary, quiet, destructive), assigned
by role across Home, Pins, Map areas and Share route, each with a real
disabled-state treatment. - ExplorInk mark + wordmark at the top of Home.
Verified on a Galaxy S24 Ultra, light and dark mode, including the
disabled and primary button states live (not just read off the layout).
Bottom navigation, a device status plate and the instrument row are the
next step, not this release.
ExplorInk GPS 0.3.8 -- a failed tile fetch says why
A skip sent back to the device carried one word, nosource, for three unrelated failures: a definite 404, a link that never got an answer, and a tile the CDN served at a content id the freshness index did not promise. That is what made the 0.3.7 content-id bug read as "the CDN does not have these tiles" for an afternoon, while a plain curl fetched every one of them.
TileSource.read now answers a result instead of a nullable byte array, and the device hears which failure it was:
| reason | what happened |
|---|---|
nosource |
a definite 404 -- nobody has built this ground |
unreachable |
no signal, a timeout, a 5xx, a body over the transfer cap |
wrongcrc |
served twice, neither copy at the promised content id |
fmt<N> |
the tile exists, built to a format this device cannot read |
Pairs with a firmware change that re-arms a held tile whose replacement never arrived -- before it, a device that failed one stale fetch never asked about that tile again until it rebooted. That half is not in firmware 0.2.3; it reaches readers with the next firmware release.
473 unit tests. Verified against a T5 S3 Pro: the real freshness check fetched and replaced a stale tile and the panel drew the street names it had been missing.
Built from master at adaccec. Debug-signed.
ExplorInk GPS 0.3.7 -- stale tiles can finally be replaced
The app could never replace a stale tile
Rebuild an area on the tile server, sync, and a device that already held
those squares kept the old ones. It looked like the sync had nothing to do.
It had not: the device reported every out-of-date tile and asked for it, and
the app refused to hand any of them over.
TileHeader computes a tile's content identity to check that what the server
just sent really is the version the index promised. Two faults, either one
enough on its own:
- The fold ran over six slots. The tile builder froze it at fifteen on
2026-08-26, so that adding a map layer would not change the identity of
every tile whose bytes had not moved. This side was never updated, and its
own test pinned the superseded six-slot numbers, so both test suites passed
while the two implementations disagreed about every tile in existence. - The header length was read as 36 bytes for every format version.
Version 4 insertedcoord_shiftand is 37, so the layer count came from the
wrong byte and the directory walk after it was nonsense.
The check then failed, the app retried past the cache, failed again and
answered skip nosource for every stale tile. Missing tiles were never
affected, because that path sends no expected identity and skips the check
entirely, which is why the whole thing looked healthy.
Found on a LilyGo T5 S3 Pro whose street names never arrived after the area
was rebuilt. Verified on that board with this build: it now holds exactly the
content id the published index carries.
Tests
The fixed vectors are copied from the tile builder's own, case for case, and a
real published version-4 header is pinned as a binary fixture. That one
matters: a header the test builds itself would have missed the length fault,
because the test would have built it at the length the code already believed
in. 470 unit tests.
What this build still is
Debug-signed, with the key every Android SDK install shares. A later
release-signed build cannot update it: Android refuses an update across a
signing key, so you would have to uninstall first. Unchanged from previous
releases and still tracked.
ExplorInk GPS 0.3.6
ExplorInk GPS 0.3.6 (versionCode 18)
Share a route from Google Maps into the device
Plan a route in Google Maps with as many stops as you like, Share, pick
ExplorInk. The app reads the stops out of the link, works out a line along the
roads, builds a trip file and puts it on the device's card, where it is listed
under Trips and drawn on the map.
It is a line on a paper map and nothing more. No turn instructions, no
rerouting, no voice: the device still answers where you are and what is around
you, with the trip drawn across it.
Two things the app says out loud rather than hiding:
- Google shares the stops, not the line. No polyline exists in a shared
link at any point, so the route drawn on the device is worked out by a router
and will differ from what you saw in Google Maps. - Every stop is listed with its coordinates before anything is sent. A stop
read wrong bends the whole route, and nothing further down the chain can
notice: the file is valid, its checksums match, the device draws it. That
list is the only check there is.
A share while the device is asleep is the normal case, not an error. The trip
waits on the phone and goes over the next time the device's map or sync screen
is open.
One share-sheet entry instead of two
Sharing a place used to offer two ExplorInk rows, one per destination. There is
now one, and the choice it used to carry -- is this a pin, or an area to cache
-- is asked in a screen that can explain both. A route never raises that
question.
Under the hood
The transfer loop that carries tiles to the device is no longer welded to
tiles: it carries any file, which is what lets a trip use it unchanged. The
trip file is built by a writer asserted byte-for-byte equal to the laptop
writer the device has been fed by since August.
Verified on hardware, 2026-10-27
A six-stop route shared from a Galaxy S24 Ultra reached a LilyGo T5 S3 Pro:
346 points, 2.8 kB, listed in the picker and ridden from it. The tile fetch was
re-checked on the same board after the refactor -- 71 of 76 tiles at 12.7 kB/s.
Host tests 468/468.
ExplorInk GPS 0.3.5
ExplorInk GPS 0.3.5 (versionCode 17)
A point shard older than the device is pushed, not skipped
The device names the highest .tip format version it reads in
NEED_POINTS <count> fmt <version>. This app checked that with equality, which
was right while there was one version.
.tip version 2 ships now. A device on the new firmware says fmt 2, and the
published shards are still mostly version 1 while the tile queue rebuilds them
one area at a time. Under the old test the app would have offered such a device
nothing at all, and a blank point layer is exactly the failure version 2 was
written to fix, arriving from the other end.
The gate is now "newer than the device", not "different from the device". A
shard from the future is still refused: the device cannot size its records and
would refuse the file whole anyway.
Verified on hardware, 2026-09-17
X4 Pro and a Galaxy S22 Ultra. The device advertised NEED_POINTS 6 fmt 2, the
CDN's shard for 561/355 is version 1, and the app pushed it: 6,048 bytes landed,
crc matched on the device. The previous build skipped the same shard as fmt1.
Host tests 425/425.