[turnipdabeets]

💡 Motivation and Context

Two more capture v1 follow-ups on top of #903, to match the capture spec (PostHog/sdk-specs#146):

• A $process_person_profile property no longer turns off the personless default. When person processing is off, the SDK now always sets process_person_profile: false. A passed or registered $process_person_profile property only fills that option when it's unset, so it can't override the SDK's false. To change person processing for one event, pass the process_person_profile option. Before, any non-null legacy property stopped the SDK from setting the option, so a registered $process_person_profile: true sent anonymous events with person processing on.
• A 2xx whose body isn't {"results": {...}} drops the batch. Before, the SDK counted it as delivered. Now it's a terminal failure: not retried, logged (debug) with only the status and the event count. This includes a body that's cut off, over 1 MiB, or slower than the read deadline. A proxy that rewrites the response body now drops events, so the migration guide's proxy section says to pass the body through.

Migration guide and the 4.0 changeset updated.

💚 How did you test it?

• New and updated unit tests for each change. They fail without the fix and pass with it.
• ./gradlew apiCheck spotlessCheck :posthog:test :posthog-android:testDebugUnitTest pass.

📝 Checklist

• All commits are signed (see Commit signing)
• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran pnpm changeset to generate a changeset file (updated the existing 4.0 major changeset)

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

• Claude Code checked the v1 branch against the capture spec and made these fixes. Stacked on fix: match capture v1 rules for empty names, drop logs, retry identity and typed values #903 because both change the per-event results reader in PostHogApi.kt.
• Three existing tests asserted the old behavior, where the legacy property won, and now assert the spec's.
• feat: let wrapper SDKs run their own beforeSend on the fully built event #904 moves the personProcessingOptions(options, mergedProperties) call. This PR drops the second argument, so whichever lands second has a one-line conflict.
• Not changed here: captureException in PostHogStateless still sets the legacy $process_person_profile property for a personless exception instead of the option. The result on the wire is the same, because the property is hoisted into the unset option.

🤖 Generated with Claude Code