1.0.1

Fixed

• The install event, and anything else tracked before the app first reaches the foreground, now
  carries the same session_id the first session.start carries. The coming session's id is
  minted at configure whenever the next foreground will start a new session, and persisted until
  a session.start adopts it, so even a launch that never reaches the foreground hands it to the
  next one. It used to be minted only at the first foreground, which left those early events
  without a session and made the server create a second session row on every first launch.
• Calling configure again while the app is in the foreground (the documented way to apply a
  consent change) resumes the session immediately. The replacement client used to stay inactive
  until the app was backgrounded and reopened.
• A device whose build fingerprint is just unknown (some OEM and custom ROM builds) is no
  longer classified as an emulator on that alone. The hardware and product checks still catch
  real emulators that report it.

Added

• Configuration validation at construction: heartbeatInterval at least 15 seconds,
  maxBatchSize between 1 and 500, flushInterval and sessionTimeout positive. A zero
  interval was a tight send loop; failing fast with a clear message beats misbehaving quietly.
• Exponential backoff between automatic delivery retries after a transient failure, jittered and
  capped at 60 seconds, honoring a numeric Retry-After on 429 as the floor. A queue past
  maxBatchSize used to retry on every new event with no throttle. An explicit flush() still
  sends immediately, and the first successful send resets the backoff.

Changed

• Environment detection checks the debuggable flag before the emulator heuristics, the order the
  platform contract lists. A debuggable build on an emulator is now tagged debug rather than
  emulator; both are kept out of your numbers by default, so nothing changes unless you send
  from one with debug = true.

1.0.0

First public release, feature-equivalent with the Swift SDK 1.0.0 and sharing its wire format.

Added

• AppGlance.configure(context, apiKey[, debug]), or configure(context, Configuration) for
  full control: intervals, enabledEnvironments, environment for beta channels,
  trackAppLifecycle, debug, endpoint.
• track, trackScreen, identify, setUserProperties, reset, setActive, flush;
  AppEnvironment, Signal, UserProperty.
• A random install id in SharedPreferences (inside Auto Backup, so a reinstall on the same
  account usually keeps it). install is recorded exactly once, first, stamped at configure.
• Sessions and presence via ProcessLifecycleOwner, with a session_id on every event and the
  same five-minute timeout the dashboard uses; a heartbeat every 60 s while foregrounded; a flush
  on background. The device's region setting as the country (collectsCountry); and anything you
  track.
• Optional user properties, merged and clamped to the server's limits (20 keys, 40 / 200
  characters), sent as user.identify only on change; reset on sign-out.
• Environments: every event is tagged production, beta, emulator or debug (stored by the
  ingest beside the Apple tiers). Emulator runs and debuggable builds are detected; a Play testing
  track is opt-in via environment = AppEnvironment.BETA. enabledEnvironments defaults to
  {PRODUCTION, BETA}.
• Debug mode (debug = true): the current build sends whatever its environment (the tag stays
  truthful, so those events appear under All in the dashboard and never in Live) and logs to
  logcat. Without it, a gated build logs one line explaining why nothing is sent.
• Delivery: events are persisted to noBackupFilesDir as they are tracked, sent oldest-first in
  slices of 100 through one serial sender, and retried after transient failures. Every event
  carries a client-minted event_id and the ingest ignores replays. Presence pings, which the
  server folds into rollups on arrival, are re-sent only when it provably never saw them -
  including after a process killed mid-request, since the on-disk queue never holds an in-flight
  ping.
• Calls apply strictly in call order on one background thread; calls made before configure, or
  before the user's first unlock under Direct Boot, are held (up to 200) and replayed.
• Permanent 4xx responses drop the slice instead of retrying forever; 413 halves it.
• minSdk 21, one dependency (androidx.lifecycle:lifecycle-process), explicit-API mode,
  ktlint-clean.