feat(llmobs): add BaseEvaluator support for Experiments

[mehulsonowal]

Stack

1 of 2 — base: master. The RemoteEvaluator follow-up is #10487.

What does this PR do?

Adds dd-trace-js LLM Experiments parity features for project configuration and the core class-based evaluator API:

• Adds llmobs.projectName and DD_LLMOBS_PROJECT_NAME configuration.
• Supports project overrides for datasets, experiments, and dataset pulls.
• Adds reusable BaseEvaluator and BaseSummaryEvaluator classes with evaluator contexts; their evaluate() methods may be synchronous or asynchronous.
• Adds rich EvaluatorResult and MultiEvaluatorResult values.
• Supports evaluator reasoning, assessment, metadata, and tags on emitted evaluation metrics.
• Updates TypeScript declarations, documentation examples, tests, and the experiments example.
• Keeps the existing mlApp/service fallback behavior.

Evaluator scope

This PR intentionally provides the evaluator primitives and result model, not a catalog of evaluator implementations:

• Included: base row and summary evaluators, contexts, and metadata-capable result types.
• Deferred: Python-style built-in evaluators, LLMJudge, and structured-output evaluator helpers.
• Remote managed evaluators are implemented separately in stacked PR feat(llmobs): add remote evaluators #10487.

CSV upload remains out of scope.

Motivation

Bring the JavaScript Experiments API closer to the dd-trace-py LLM Experiments setup while keeping the initial evaluator surface focused on reusable primitives. Applications can implement their own evaluator logic and attach rich metadata to emitted metrics without committing the JavaScript SDK to a built-in evaluator catalog.

Additional Notes

• Updated with the latest master.
• Targeted evaluator/facade/SDK tests: 252 passing, 2 pending.
• Changed-file ESLint passes.
• Two VCR-backed facade cases require the local replay service at 127.0.0.1:9126 and were excluded from the local run.

[dd-octo-sts]

Overall package size

Self size: 9.34 MB
Deduped: 10.07 MB
No deduping: 10.07 MB

Dependency sizes

| name | version | self size | total size | |------|---------|-----------|------------| | import-in-the-middle | 3.5.1 | 127.66 kB | 513.61 kB | | opentracing | 0.14.7 | 194.81 kB | 194.81 kB | | dc-polyfill | 0.1.11 | 25.74 kB | 25.74 kB |

_{🤖 This report was automatically generated by heaviest-objects-in-the-universe}

[datadog-official]

✅ All CI checks and tests passed. Datadog automation helped this PR pass.

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🔄 Datadog retried 2 tests - 2 passed on retry

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 98.26% (+0.05%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 5b50359 | Docs | View more details | Give us feedback!}

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-10-01 19:26:37

Comparing candidate commit 5b50359 in PR branch mehul/llmobs-experiments-parity with baseline commit 0a5e10c in branch master.

📊 Benchmarking dashboard

Found 0 performance improvements and 0 performance regressions! Performance is the same for 2300 metrics, 13 unstable metrics.

Explanation

This is an A/B test comparing a candidate commit's performance against that of a baseline commit. Performance changes are noted in the tables below as:

• 🟩 = significantly better candidate vs. baseline
• 🟥 = significantly worse candidate vs. baseline

We compute a confidence interval (CI) over the relative difference of means between metrics from the candidate and baseline commits, considering the baseline as the reference.

If the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD), the change is considered significant.

Feel free to reach out to #apm-benchmarking-platform on Slack if you have any questions.

More details about the CI and significant changes

You can imagine this CI as a range of values that is likely to contain the true difference of means between the candidate and baseline commits.

CIs of the difference of means are often centered around 0%, because often changes are not that big:

---------------------------------(------|---^--------)-------------------------------->
                              -0.6%    0%  0.3%     +1.2%
                                 |          |        |
         lower bound of the CI --'          |        |
sample mean (center of the CI) -------------'        |
         upper bound of the CI ----------------------'

As described above, a change is considered significant if the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD).

For instance, for an execution time metric, this confidence interval indicates a significantly worse performance:

----------------------------------------|---------|---(---------^---------)---------->
                                       0%        1%  1.3%      2.2%      3.1%
                                                  |   |         |         |
       significant impact threshold --------------'   |         |         |
                      lower bound of CI --------------'         |         |
       sample mean (center of the CI) --------------------------'         |
                      upper bound of CI ----------------------------------'

Unstable benchmarks

These benchmarks have a confidence interval too wide to call a change; treat them as noise rather than signal.

scenario:debugger-line-probe-with-snapshot-default-20

• unstable max_rss_usage [-10.442MB; +6.291MB] or [-7.623%; +4.592%]

scenario:debugger-line-probe-without-snapshot-24

• unstable cpu_user_time [-627.758ms; +1101.614ms] or [-7.043%; +12.360%]
• unstable execution_time [-632.791ms; +1116.232ms] or [-6.308%; +11.128%]
• unstable throughput [-114.327op/s; +64.170op/s] or [-10.058%; +5.646%]

scenario:dogstatsd-aggregated-26

• unstable cpu_user_time [-99.170ms; +155.352ms] or [-6.049%; +9.476%]
• unstable execution_time [-97.509ms; +157.171ms] or [-5.892%; +9.496%]
• unstable throughput [-1023662.969op/s; +639168.505op/s] or [-6.991%; +4.365%]

scenario:encoders-0.4-immediate-flush-20

• unstable max_rss_usage [-1.333MB; +10.532MB] or [-1.614%; +12.758%]

scenario:plugin-aws-sdk-eventbridge-inject-detail-26

• unstable execution_time [-174.457ms; +294.372ms] or [-4.377%; +7.385%]
• unstable throughput [-102032.801op/s; +60474.161op/s] or [-6.753%; +4.003%]

scenario:plugin-graphql-long-with-depth-off-26

• unstable max_rss_usage [-30.673MB; +17.981MB] or [-11.901%; +6.976%]

scenario:plugin-graphql-long-with-depth-on-max-26

• unstable max_rss_usage [-5.964MB; +21.295MB] or [-2.314%; +8.261%]

scenario:plugin-mongodb-core-binary-hash-24

• unstable execution_time [-150.909ms; +171.517ms] or [-5.094%; +5.789%]

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.85%. Comparing base (0a5e10c) to head (5b50359).
⚠️ Report is 30 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #9848      +/-   ##
==========================================
+ Coverage   98.78%   98.85%   +0.06%     
==========================================
  Files        1042     1048       +6     
  Lines      167109   169089    +1980     
  Branches       74       74              
==========================================
+ Hits       165081   167146    +2065     
+ Misses       2028     1943      -85     

Flag	Coverage Δ
ai-guard	63.93% <31.61%> (+0.40%)	⬆️
apm-capabilities	54.90% <31.61%> (+0.06%)	⬆️
apm-integrations	81.17% <31.61%> (-0.09%)	⬇️
appsec	57.94% <31.61%> (+<0.01%)	⬆️
debugger	69.59% <31.61%> (-0.34%)	⬇️
instrumentation	55.41% <31.61%> (+0.18%)	⬆️
llmobs	80.57% <100.00%> (+0.39%)	⬆️
master-coverage	98.85% <100.00%> (?)
openfeature	65.87% <31.61%> (-0.19%)	⬇️
platform	68.72% <31.61%> (-0.04%)	⬇️
profiling	66.15% <31.61%> (-0.26%)	⬇️
serverless	65.71% <31.61%> (+0.51%)	⬆️
test-optimization	82.00% <31.61%> (-0.15%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: da49206da1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9bfdf6c977

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[sabrenner]

just one small nit, and i think some of the codex comments probs need addressing, but otherwise lgtm! approving to unblock once codex comments are addressed

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6faeb6f120

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".