Android SDK 7.3.1 and Gradle plugin 4.0.8

[krassx]

Adopts Bugsee Android SDK 7.3.1 (Maven Central) and Bugsee Android Gradle plugin 4.0.8.

Pins and everything derived from them

• native-versions.json: android.sdk 7.3.1, android.gradlePlugin 4.0.8.
• Derived from them: the bare example's plugins block and its guard, plugin/build/native-versions.baked.json, the regenerated android-options-manifest.json (from v7.3.1; only sdkVersion/generatedAt change, and option-keys.json/option-enums.json are byte-identical), the single-source tests, the design/plan version lines, and the README's manual plugins block. A new test keeps that README line on the pin.
• The bare example sets no ndk.debugSymbolLevel. With 4.0.8 the native upload reads the unstripped merged_native_libs, whatever the level is (gradle-plugin feat: Phase 5 — attributes and identity #10/feat: Phase 6.9 — iOS privacy checks and wrapper at module init #13).

Gradle plugin 4.0.8: #8, #9 and #10 confirmed in the published artefact

The Plugin Portal's 4.0.8 marker POM points at com.bugsee:bugsee-android-gradle-plugin:4.0.8. Its jar has bugsee-plugin-version.txt = 4.0.8, CliBinaryResolver.DEFAULT_VERSION = "0.8.1" (#8, floor ≥ 0.7.12), --force in CliUploader (#9), and NativeUploadTask.useMergedNativeLibs/mergedNativeLibsDir (#10). The release-branch tree f24023c equals main@9139989 apart from version.txt, so #11, #12, #13 and #15 are all in it.

Fixed Android bugs: pins flipped (iOS pins untouched)

Test	Was	Now	Fix
lifecycle itDelete	it.failing both	plain it on Android (iOS stays, cocoa#195)	android#214 (#222)
feedback itGreeting	it.failing on Android	it	android#215 (#225)
exceptions case 10, "no second report"	it.failing on Android	it	#192 (unhandled exception files a crash only)
launch-options enum witnesses	expected ordinals 0/2/4	expect values 1/3/21	android#216 (#223)

After #63 landed (rebased onto d957f55), its Android pins are flipped as well:

Test (#63)	Fix
report-dialog-keys RA-02 actionBarColor	android#217 (#224)
feedback-keys FA-01 / FA-11 / FA-20	android#218 (#226). FA-19 inputTextColor was only visible through 7.3.0's buggy hint; it now types into the input (1524 px of #0000ff)
sdk-breadcrumbs FLOW-15	android#219 (#228)
options-readback environment record	android#216 (#223): values, not ordinals
option-effects OPT-025 (native crash, detect.crash=false)	android#220 (#227)
option-effects OPT-081 (out-of-process report)	android#221 (#229)
option-effects OPT-002	was pinned on #219. With crumbs back, capture.breadcrumbs.extras adds no crumbs on Android: it copies the intent extras into system crumbs' data.extras (BreadcrumbSystemEvents), which is now the witness (1 vs 0)

Each of these was first run unchanged on 7.3.1. Every pin above failed with "Failing test passed", and then the flipped tests passed.

Still pinned on Android: exception labels (still ignored), plus the two new SDK issues below.

Behaviour changes found (tests adjusted only where 7.3.1 is the correct behaviour)

• performance.json is now OTLP/JSON (android#207 via #208; bugsee/specs sdk/reporting/bundle/performance.md). span-lifecycle and apm read transactions, which no longer exists. scripts/performance-capture.ts reads both layouts (legacy, and OTLP by the spec's profile-2 reader rules) into the fields the tests already assert. The assertions themselves are unchanged. Focused Stryker: 98.41%, with 2 equivalent survivors.
• The feedback action bar is now painted (android#218): OCR reads the back arrow into the title ("< Feedback"). A leading arrow glyph is dropped before the exact match.
• Network (#231–#243): an oversized body is dropped whole with size_too_large. That is correct, and Android's custom-option now asserts it directly. Regression: a POST that fails before its body is written is recorded without the body (no_data); on 7.3.0 it kept the body. Filed bugsee/bugsee-android#246, pinned it.failing on Android.
• The feedback greeting (#215 / #225) is still not shown on this flow (set right before showFeedbackActivity, offline). The prefs hold it. Filed bugsee/bugsee-android#245, pinned again on Android.
• Request supplements (isOverride) reach the network filter. The Android snapshot exposes a read-only override (cursor[bot] P2).

Device: WOD_LX1, Debug arm64, Bugsee Android SDK 7.3.1 [0aa9ae1ff] (= v7.3.1)

Suite	Result
network / network-filter / add-network-event	3/3, 2/2, 2/2
lifecycle (delete-data)	6/6
span-lifecycle / apm (OTLP reader)	4/4, 2/2
feedback	3/3 (greeting it.failing, #245)
exceptions	13 passed, 21 skipped (case 10 plain it)
launch-options	38 + custom-option and failed-POST (re-run)
smoke	12 passed, 1 skipped
report-dialog-keys / feedback-keys / sdk-breadcrumbs / options-readback / option-effects	22/22, 41/41, 2/2, 81/81, 16 passed + 10 skipped

Release + R8 launch smoke (7.3.1 banner, status=2, relaunch, no fatal): WOD_LX1, emulator bugsee_api24 (SDK 24), bugsee_api36 (SDK 36).

Release + R8, and the native symbol upload (loopback stub only)

assembleRelease -PbugseeE2eMinify=true, 4 ABIs, no debugSymbolLevel anywhere in the app, a non-placeholder fake token, and plugin.endpoint on a 127.0.0.1 stub that records every request (BUGSEE_CLI_AUTO_UPDATE=0):

• uploadBugseeReleaseNative ran bugsee-cli 0.8.1 on build/intermediates/merged_native_libs/release: "collected native libraries libraries=64 uploadable=64", 66 symbol POSTs plus 66 PUTs (139 MB). The 2 extra are the source map and the R8 mapping.
• Every arm64 library's GNU build-id matches its stripped copy in the APK. Sent as dwarf: libbugsee_e2e_native.so (this app's NDK crash library, 3c73ec05…), libappmodules, libreactnative (7e70f59e…, 140 MB unstripped vs 6.9 MB in the APK), libjsi, libfbjni. Sent as symtab: the stripped ones (libhermesvm, libc++_shared, the Bugsee SDK's own). Every POST has replace_if_richer: true.
• AGP's own native-debug-symbols.zip (.so.sym) was not used.

🤖 Generated with Claude Code

[cursor]

Stale comment

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8

Reviewed 908380d against surrounding pins, the Expo plugin, the JS/Android network-filter bridge, 7.3.1 release notes, and specs PR 42 (OTLP profile 2). Did not run tests locally (no node_modules in this environment). GitHub CI lint, typecheck, unit was green at review time; the PR body’s device section is still in progress.

Findings

P2 High — JS network filter cannot see 7.3.1 isOverride

• Location: packages/react-native/android/src/main/java/com/bugsee/reactnative/NetworkFilterBridge.java (snapshot, ~414–444); packages/react-native/src/network/filter.ts (NetworkFilterEvent)
• Problem: 7.3.1 can deliver a second start event for the same request (NetworkEvent.isOverride() == true) that carries the request body. The RN snapshot never includes that flag, so JS filters cannot implement the contract in the 7.3.1 notes (“redact each event by what it carries; the body may be on the additional event”).
• Impact: A filter that only rewrites the first before — including this repo’s installNetworkFilter in examples/bare/scenarios/network-filter.ts — leaves the override event’s URL/body in the report. Per-event {...event, body: redact(event.body)} still works; first-event-wins filters that were correct on 7.3.0 do not.
• Scenario: setNetworkFilter redacts URL/body on the first before. OkHttp/HttpURLConnection then offers an additional start with the body. JS returns that event unchanged (no override field to branch on). The capture keeps the original URL and body.
• Fix: Put override: event.isOverride() on the Android snapshot (read-only; apply already mutates the same native object). Add override?: boolean to NetworkFilterEvent. Document that every event must be redacted on its own. Extend the network-filter e2e past GET-without-body so a second start would fail toHaveLength(1) if left unrewritten.

P3 Low — README still says debugSymbolLevel is required for Bugsee upload

• Location: packages/react-native/README.md (manual android/app/build.gradle block, ~215–230)
• Problem: That block still says AGP extracts none by default, “so there would be nothing to upload”, and requires debugSymbolLevel 'SYMBOL_TABLE'. This PR’s example and plan state that plugin 4.0.8 uploads from merged_native_libs regardless of the level. The new README test only locks the plugins { version } line.
• Impact: Manual-setup docs contradict the reason for 4.0.8. Following the README is harmless (Play symbols still help). Copying the example while overriding gradlePluginVersion back to 4.0.7 would skip native upload.
• Scenario: A reader of the 4.0.8 README vs the example comments gets opposite instructions for the same pin.
• Fix: State that 4.0.8 uploads from merged_native_libs without debugSymbolLevel, and that SYMBOL_TABLE is optional (Play Console), matching the example. Keep the Expo plugin injection if Play symbols are still wanted.

Not bugs

• Pins are consistent (native-versions.json, baked JSON, example Gradle guard, options-manifest sdkVersion, pin tests). option-keys.json unchanged matches 7.3.1 “No new options”.
• performanceTransactions matches the profile-2 rules from specs PR 42 (local-root 0x200, hex case, snapshot vs complete dedup, url.full / db.query.text / file.path / bugsee.description). Unit tests cover the mutants that matter. iOS still takes the legacy transactions path.
• Android it.failing → it flips match 7.3.1 notes: delete-all-generations (#222), feedback greeting (#225), unhandled exception is one crash (#192), enum options recorded by spec value (#223).

---

1. Overall risk: Medium
2. Merge recommendation: Request changes — expose override on the JS filter snapshot before this pin ships. Do not merge until the WOD_LX1 device section is filled in; 7.3.1’s network changes are the part this PR does not cover with new e2e.
3. Most important to fix: P2 network-filter isOverride gap.
4. Positive: Tight single-source pin update; OTLP reader is scoped to e2e and well tested; it.failing flips are tied to named Android fixes rather than weakened.

  

_{Sent by Cursor Automation: Bugsee code review}

[cursor]

Stale comment

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8 (re-review)

Re-reviewed fd60cbc (prior 908380d) against the JS/Android network-filter bridge, the published 7.3.1 AAR on Maven Central, the Expo plugin, and the new device pins. Did not run tests locally (no node_modules). GitHub CI for this head is still queued; the previous ci run was cancelled by this push.

fd60cbc does not touch the wrapper filter snapshot or the README debugSymbolLevel block. Both prior findings still stand.

Findings

P2 High — JS network filter cannot see 7.3.1 isOverride (not addressed)

• Location: packages/react-native/android/src/main/java/com/bugsee/reactnative/NetworkFilterBridge.java (snapshot, 414–444); packages/react-native/src/network/filter.ts (NetworkFilterEvent)
• Problem: Maven Central com.bugsee:bugsee-android:7.3.1 adds NetworkEvent.isOverride() / setOverride(boolean) (absent from 7.3.0). The concrete capture entry and OkHttp interceptor implement them; the interface defaults are false / no-op. NetworkFilterBridge.snapshot() still serializes id/url/method/body/headers/type only — no override — and NetworkFilterEvent has no such field.
• Impact: A filter that only rewrites the first before — including this repo’s installNetworkFilter in examples/bare/scenarios/network-filter.ts — leaves the override event’s URL/body in the report. Per-event {...event, body: redact(event.body)} still works; first-event-wins filters that were correct on 7.3.0 do not.
• Scenario: setNetworkFilter redacts URL/body on the first before. OkHttp then offers an additional start with the body (isOverride() == true). JS returns that event unchanged. The capture keeps the original URL and body. This head’s custom-option e2e already iterates every before of the 64-byte POST, which is the shape that produces the extra start.
• Fix: Put override: event.isOverride() on the Android snapshot (read-only; apply already mutates the same native object). Add override?: boolean to NetworkFilterEvent. Document that every event must be redacted on its own. Extend the network-filter e2e past GET-without-body so a second start would fail toHaveLength(1) if left unrewritten.

P3 Low — README still says debugSymbolLevel is required for Bugsee upload (not addressed)

• Location: packages/react-native/README.md (manual android/app/build.gradle block, ~215–230)
• Problem: That block still says AGP extracts none by default, “so there would be nothing to upload”, and requires debugSymbolLevel 'SYMBOL_TABLE'. This PR’s example and plan state that plugin 4.0.8 uploads from merged_native_libs regardless of the level. The new README test only locks the plugins { version } line.
• Impact: Manual-setup docs contradict the reason for 4.0.8. Following the README is harmless (Play symbols still help). Copying the example while overriding gradlePluginVersion back to 4.0.7 would skip native upload.
• Scenario: A reader of the 4.0.8 README vs the example comments gets opposite instructions for the same pin.
• Fix: State that 4.0.8 uploads from merged_native_libs without debugSymbolLevel, and that SYMBOL_TABLE is optional (Play Console), matching the example. Keep the Expo plugin injection if Play symbols are still wanted.

Not bugs (this head)

• fd60cbc correctly puts itGreeting back on it.failing for this flow (greeting in preferences, empty chat, android#245) and pins control’s refused POST body as it.failing (no_data, android#246). Those are native SDK gaps, not wrapper defects.
• Android custom-option now witnesses #232 directly (before + size_too_large) instead of comparing against a control that 7.3.1 no longer fills. iOS still uses the shrink-against-control check.
• Pins remain consistent (native-versions.json, baked JSON, example Gradle guard, options-manifest sdkVersion). option-keys.json unchanged matches 7.3.1 “No new options”.
• performanceTransactions still matches the profile-2 rules (local-root 0x200, hex case, snapshot vs complete dedup, url.full / db.query.text / file.path / bugsee.description). iOS still takes the legacy transactions path.
• Android it.failing → it flips that remain: delete-all-generations (#222), unhandled exception is one crash (#192), enum options recorded by spec value (#223).

---

1. Overall risk: Medium
2. Merge recommendation: Request changes — expose override on the JS filter snapshot before this pin ships. Do not merge until that wrapper gap is closed; the WOD_LX1 network pins in fd60cbc are the right e2e direction but they do not cover the filter.
3. Most important to fix: P2 network-filter isOverride gap.
4. Positive: Device-driven pins for #245/#246 instead of claiming #225/#232 fixed things they did not; OTLP reader stays scoped to e2e; pin update is still single-source.

  

_{Sent by Cursor Automation: Bugsee code review}

[cursor]

Stale comment

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8 (re-review)

Re-reviewed bcfd6c6 (prior fd60cbc, 908380d). bcfd6c6 is the wrapper/docs follow-up for the two findings from the last pass.

Prior findings

P2 High — JS network filter cannot see 7.3.1 isOverride — fully addressed.
NetworkFilterBridge.snapshot() now writes override: event.isOverride(). NetworkFilterEvent.override?: boolean documents it as Android-only, read-only, and that a supplement must be filtered like any other event. apply() still has no override writer, so a JS echo cannot clear the native flag. The new JVM test covers both snapshot values and that a reply with override: false leaves the event’s flag true.

No POST e2e was added. That is acceptable: installNetworkFilter is a first-before round-trip probe, and its GET has no body so 7.3.1 does not offer a supplement. Residual only: a customer filter that still does first-event-wins on a POST would leak the extra start; the type comment is the contract for that.

P3 Low — README still says debugSymbolLevel is required for Bugsee upload — fully addressed.
The manual android/app/build.gradle block no longer sets a level. It matches the example: plugin 4.0.8 uploads from unstripped merged_native_libs; a level is only for Play. Expo injection remaining until #64 is noted in-thread and is out of this PR.

Not bugs

• Pins, OTLP reader, and the device-driven #245 / #246 / size_too_large pins from fd60cbc are unchanged and still look right.
• The JVM stub does not override setOverride; production apply() never calls it, so the “reply cannot change it” assertion is still the right check against the write list.

---

1. Overall risk: Low
2. Merge recommendation: Approve. No remaining P0–P3 on this head.
3. Most important to fix: Nothing on this PR.
4. Positive: Small, targeted follow-up; snapshot is read-only as recommended; README now agrees with the 4.0.8 example.

  

_{Sent by Cursor Automation: Bugsee code review}

[cursor]

Stale comment

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8 (re-review)

Re-reviewed bcfd6c6 (prior fd60cbc, 908380d) against the JS/Android network-filter bridge (snapshot, apply, JVM tests, setNetworkFilter / replacementJson), the iOS snapshot (no override key, extra keys ignored), the Expo plugin still injecting debugSymbolLevel, pin sources (native-versions.json, baked JSON, example Gradle guard, options manifest), and the OTLP e2e reader vs span-lifecycle/apm. Did not run tests locally (no node_modules). GitHub ci for this head was still queued at review time.

bcfd6c6 closes both prior findings. No new wrapper defect with a credible failure scenario.

Prior findings

P2 High — JS network filter cannot see 7.3.1 isOverride — fully addressed

• NetworkFilterBridge.snapshot() now writes override: event.isOverride() (read-only; apply still only writes url/body/headers/error fields/statusText).
• NetworkFilterEvent.override?: boolean documents the 7.3.1 supplement contract (same id/type, typically the request body).
• JVM test covers both flag values and that echoing override: false in the reply does not change the native event.

P3 Low — README still says debugSymbolLevel is required for Bugsee upload — fully addressed

• Manual setup and the bare example both omit the level and state that plugin 4.0.8 uploads from unstripped merged_native_libs; a level is only for Play.

Findings

None remaining in the wrapper.

Not bugs (this head)

• itGreeting stays it.failing on Android for this flow (greeting in preferences, empty chat, android#245). Native gap, not a wrapper defect.
• Control POST body is it.failing (no_data on refused connect, android#246). Same: native capture regression, honestly pinned.
• Custom-option Android witness matches #232 (every before of the 64-byte POST is size_too_large). iOS still compares against control.
• The network-filter scenario is still a GET with no body, so it does not produce a supplement; installNetworkFilter's first-before rewrite is the round-trip probe, not a POST redaction template. Filters that spread the event and rewrite each before still work; first-id-wins POST filters were already broken on 7.3.1 even before the flag was exposed.
• Expo config plugin still writes debugSymbolLevel 'SYMBOL_TABLE'. Harmless with 4.0.8 (merged_native_libs is independent of the level); author pointed at #64 to stop injecting it.
• Pins stay consistent. option-keys.json / option-enums.json unchanged matches 7.3.1 “No new options”.
• performanceTransactions still matches profile-2 (local-root 0x200, hex case, snapshot vs complete dedup, url.full / db.query.text / file.path / bugsee.description). iOS still takes the legacy transactions path.

---

1. Overall risk: Low
2. Merge recommendation: Approve — prior P2/P3 are addressed; no remaining wrapper issues to fix before merge. Residual native pins (#245, #246) are documented, not hidden.
3. Most important to fix: Nothing in this wrapper. Follow-ups live in bugsee-android (#245, #246) and, if desired, stopping the Expo plugin from writing debugSymbolLevel (#64).
4. Positive: override is read-only on the wire the way apply already worked; README and the example now tell the same 4.0.8 story; device-driven it.failing for #245/#246 instead of claiming native fixes that this flow does not show; OTLP reader stays e2e-scoped and unit-tested.

  

_{Sent by Cursor Automation: Bugsee code review}

[cursor]

Stale comment

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8 (re-review)

Re-reviewed b4d99fa after the rebase onto #63 (prior 02dd607 / bcfd6c6, fd60cbc, 908380d). The new commit is e2e pin flips only, reported as seen on WOD_LX1 with 7.3.1 [0aa9ae1ff].

Prior findings

P2 (isOverride on the JS filter snapshot) and P3 (README debugSymbolLevel) remain fully addressed in 02dd607. Snapshot still writes override: event.isOverride(); the README still omits the level and points at merged_native_libs.

This head (b4d99fa)

No new wrapper or test defect with a credible failure scenario.

The #63 Android it.failing pins that 7.3.1 actually fixed are now plain tests, tied to the named SDK PRs:

• report-dialog RA-02 actionBarColor (#224)
• feedback FA-01 / FA-11 / FA-20 (#226); FA-19 inputTextColor now types into the chat EditText instead of asserting the old 0.5-alpha hint blend
• FLOW-15 SDK breadcrumbs (#228)
• options-readback enums as values, not ordinals (#223)
• OPT-025 native SIGSEGV with detect.crash=false (#227) and OPT-081 out-of-process report (#229)

OPT-025 stays split (JS fatal vs native), so a JS detect.crash=false regression can still go red on its own — that was the remaining P2 on #63.

OPT-002 is not a weakened count check: 7.3.1 records the same crumb kinds with or without extras (WOD_LX1: 16/16), and the witness is now system crumbs whose JSON carries "extras":{" only on the option-on run. That matches BreadcrumbSystemEvents copying intent extras into crumb data.

Not bugs

• itGreeting / control POST no_data remain it.failing (android#245 / #246). Native gaps, still pinned.
• Expo plugin still injects debugSymbolLevel (called out for #64); harmless with 4.0.8.
• FA-19 takes the first EditText in the dump; on this chat flow that is the message field, and the device run found 1524 px of the typed colour.

---

1. Overall risk: Low
2. Merge recommendation: Approve. No remaining P0–P3 on this head.
3. Most important to fix: Nothing in this wrapper. Residual native follow-ups stay #245 and #246.
4. Positive: Pin flips are device-proven and named to Android PRs; OPT-025 is not re-combined; OPT-002’s witness matches the option’s real effect instead of a count that 7.3.1 no longer changes.

  

_{Sent by Cursor Automation: Bugsee code review}

[cursor]

Deep review: Android SDK 7.3.1 / Gradle plugin 4.0.8 (re-review)

Re-reviewed b4d99fa (prior bcfd6c6, fd60cbc, 908380d) against the wrapper network-filter path (Android snapshot/apply, JS NetworkFilterEvent/replacementJson, iOS snapshot which still has no override key), pin sources (native-versions.json, baked JSON, example Gradle guard, options manifest sdkVersion/generatedAt only), the OTLP e2e reader vs span-lifecycle/apm, and the new #63 suite pin flips. Did not run tests locally (node_modules absent). GitHub ci for this head was still queued at review time.

b4d99fa is the rebase onto main after #63 plus device-proven un-pins. No new wrapper defect with a credible failure scenario.

Prior findings

P2 High — JS network filter cannot see 7.3.1 isOverride — fully addressed

• NetworkFilterBridge.snapshot() still writes override: event.isOverride() (read-only; apply still only writes url/body/headers/error fields/statusText).
• NetworkFilterEvent.override?: boolean still documents the 7.3.1 supplement contract (same id/type, typically the request body).
• JVM test still covers both flag values and that echoing override: false in the reply does not change the native event.

P3 Low — README still says debugSymbolLevel is required for Bugsee upload — fully addressed

• Manual setup and the bare example both omit the level and state that plugin 4.0.8 uploads from unstripped merged_native_libs; a level is only for Play.

Findings

None remaining in the wrapper. The new pin flips match 7.3.1 as measured on WOD_LX1:

• report-dialog RA-02 / feedback FA-01, FA-11, FA-20: KNOWN_NOT_PAINTED.android emptied (#217/#218).
• FA-19 inputTextColor now types into the chat EditText (7.3.1 draws the hint in inputTextHintColor).
• FLOW-15 SDK breadcrumbs: plain it (#219/#228).
• options-readback and launch-options enum witnesses: values 1/3/21, not ordinals (#216/#223).
• OPT-025 native crash with detect.crash=false and OPT-081 out-of-process report: plain tests (#220/#221).
• OPT-002: extras attach to system crumbs' data (same crumb count as control); the old length comparison would have been a false fail.

Not bugs (this head)

• itGreeting stays it.failing on Android for this flow (greeting in preferences, empty chat, android#245). Native gap, not a wrapper defect.
• Control POST body is it.failing (no_data on refused connect, android#246). Same: native capture regression, honestly pinned.
• Custom-option Android witness matches #232 (every before of the 64-byte POST is size_too_large). iOS still compares against control.
• Exception case 10 is a plain it (#192); case 7 (labels) stays it.failing. iOS delete stays it.failing (cocoa#195).
• The network-filter scenario is still a GET with no body, so it does not produce a supplement; installNetworkFilter's first-before rewrite is the round-trip probe. Filters that spread the event and rewrite each before still work.
• Expo config plugin still writes debugSymbolLevel 'SYMBOL_TABLE'. Harmless with 4.0.8 (merged_native_libs is independent of the level); stopping that inject is #64, not this PR.
• Pins stay consistent. option-keys.json / option-enums.json unchanged matches 7.3.1 “No new options”. FrameRate.Low/High = 1/3 and VideoMode.DirectBuffers = 21 match option-enums.json.
• performanceTransactions still matches profile-2 (local-root 0x200, hex case, snapshot vs complete dedup, url.full / db.query.text / file.path / bugsee.description). iOS still takes the legacy transactions path.

---

1. Overall risk: Low
2. Merge recommendation: Approve — prior P2/P3 remain addressed; the #63 pin flips are the 7.3.1 behaviour, not weakened assertions. Residual native pins (#245, #246) are documented, not hidden.
3. Most important to fix: Nothing in this wrapper. Follow-ups live in bugsee-android (#245, #246) and, if desired, stopping the Expo plugin from writing debugSymbolLevel (#64).
4. Positive: Rebase onto #63 did not paper over remaining 7.3.0 gaps — labels, greeting-on-this-flow, and refused-POST body stay it.failing where the device still shows them. override stays read-only on the wire; OTLP reader stays e2e-scoped and unit-tested; OPT-002 was rewritten to the actual extras-in-data witness instead of a count that 7.3.1 would not satisfy.

  

_{Sent by Cursor Automation: Bugsee code review}