[arifBurakDemiray]

Stacked on #275. Your uncommitted P7 diff (log gathering, connection test tests, changelog) applies cleanly on top of this branch (git apply --check).

What

Fix	Why
Replace, don't merge, the stored settings (port of the reference SDK's settings merge fix, 2026-10-19)	Each answer was merged into the stored settings, so a setting removed in SDK Manager kept its last server value for good: a removed tracking:false, eb, jte, lkl or rqs stayed in force across restarts. Now each valid answer replaces the stored c, and a missing key resolves to the developer value or SDK default. removeListingFilterKeysFromConfig is gone.
cr never lifts the developer's requirement	Cross-SDK rule decided 2026-10-21: resolved = developer requiresConsent OR a cr true. A server, stored or provided cr:false lifted setRequiresConsent(true): everything tracked without consent and onConsentRemoval was ignored.
Journey trigger on a device ID change	deviceIdChanged ends timed events after the new ID is set, so a jte trigger (or a full queue) force-sent the previous user's events under the new ID. Those events now go out together under the previous ID, with no content refresh for the new user.
Store file cut short	JsonFileStorage.save() truncated and rewrote in place; a process stopped mid-write left a file that made every later Countly.init throw JSONException into the app. Saves now go through a temp file and an atomic move, an invalid store starts empty, and a leftover temp file is deleted at open.

Verification

• 9 new tests; each fails on 49c7e03 (cr lifted, removed setting kept, events under new_device_id, JSONException out of init).
• Full :sdk-java suite, JDK 8, isolated user.home: 675/675, twice.
• End to end against a live Countly server, whose real settings answer is {}: with setRequiresConsent(true) and provided cr:false, feat: SBS #275 sent events from two runs without consent and this branch sent none. A tracking:false the server dropped in its next answer kept tracking off on feat: SBS #275, not here. The journey trigger events went out under the new device ID on feat: SBS #275 and under the previous one here. A cut-short store file made Countly.init throw on feat: SBS #275, not here.
• Tests that relied on the merge now carry the settings they need in each response (Content, Filters, Limits, Network, ModuleConfiguration).

Left for you

• updateConfigVariables javadoc still says "merged settings": your P7 diff edits the adjacent line, so changing it here would conflict.
• Several Gates/Filters tests push responses without networking(false), so networking turns back on mid-test. They pass because sends to the dummy URL fail and keep the files; only the journey trigger test asserted pending triggers, which is fixed here. GatesTests also overlaps your P7 edits, so it is untouched.
• cr:true from the server still makes the feature facades return null on the next start (D2), so host code like events().recordEvent(...) throws an NPE. That needs a design decision (non-null no-op facades vs documenting it).
• Deeper device ID fix: end the old user's events and views, save the profile, then switch the ID (the reference order) in ModuleDeviceIdCore. It would also send views().startView end events under the old ID. Larger blast radius, so not here.
• Changelog lines for P8: "Mitigated an issue where a setting removed from the SDK behavior settings kept its last value instead of going back to the configuration." and "Mitigated an issue where a store file cut short by a stopped process made initialization fail."

🤖 Generated with Claude Code