TypeError: _a.trim is not a function during exposure event

[FabricioFFC]

Expected Behavior

The Experiment Node Server SDK should safely handle non-string values when canonicalizing exposure data before calling .trim(). Invalid or unexpected input types should not cause a runtime exception during variant evaluation or exposure tracking.

Current Behavior

A runtime exception is thrown during exposure tracking:

TypeError _a.trim is not a function 
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure.js:16:93 Exposure.canonicalize
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure-filter.js:16:44 InMemoryExposureFilter.shouldTrack
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure-service.js:23:37 AmplitudeExposureService.<anonymous>
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure-service.js:8:71 
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure-service.js:4:12 __awaiter
    /app/node_modules/@amplitude/experiment-node-server/dist/src/exposure/exposure-service.js:22:16 AmplitudeExposureService.track
    /app/node_modules/@amplitude/experiment-node-server/dist/src/local/client.js:149:87 LocalEvaluationClient.evaluateV2
    .next/server/chunks/7651.js:19:1755 d.fetchVariants

Possible Solution

Improve input validation earlier in the exposure pipeline.

Steps to Reproduce

We weren't able to reproduce it locally.

Environment

• JS Server SDK Version: 1.13.2
• Runtime: Node.js
• Installation Method: yarn
• Browser and Version: N/A

[FabricioFFC]

We found the root cause: in our flow, we're sometimes sending user_id as a number instead of a string. This is a very edge case, and we're implementing a fix to prevent this. So feel free to close this issue.