v6.20.0 proposal

[dd-octo-sts]

Features

• Dynamic Instrumentation: Add an evaluation time budget #10156
• Feature Flags: Add privacy-protected flag evaluation telemetry #10464
• LLM Observability: Add BaseEvaluator support for Experiments #9848
• LLM Observability: Add remote evaluators #10487
• OpenTelemetry: Add OTel consistent probability sampling state to spans exported through OTLP #10189
• OpenTelemetry: Otel sampling support in distributed tracing #10117

Fixes

• aws-sdk: Handle region provider rejections #10583
• Dynamic Instrumentation: Bound pause duration telemetry while the application is busy #10607
• Dynamic Instrumentation: Classify worker failures in telemetry #10393
• Dynamic Instrumentation: Drop probe results that stay over the size cap #10345
• Dynamic Instrumentation: Redact sensitive identifiers in log templates and capture expressions #10612
• General: Restore legacy RC capability bits #10573
• LLM Observability: Don't corrupt literal "\u" sequences when encoding payloads #10609
• LLM Observability: Keep LLMObs tracing after an aborted stream #10577
• LLM Observability: Preserve structured experiment output #10472
• LLM Observability: Properly record runtimeContext in newer v7 versions #10629
• OpenTelemetry: Return non-function targets unchanged from ContextManager.bind() #10544

Internal (CI, Testing, Benchmarking)

• AppSec: Update native-appsec to 11.0.3 #10625
• Feature Flags: Shorten evaluation workloads #10588
• Feature Flags: Verify bundled provider behavior and align test versions #10545
• General: Shorten replacement injection #10587
• General: Update one-pipeline version #10610
• postgres: Remove identity hooks #10649
• supabase: Remove identity hooks #10511

Contributors

[dd-octo-sts]

Overall package size

Self size: 9.44 MB
Deduped: 10.2 MB
No deduping: 10.2 MB

Dependency sizes

| name | version | self size | total size | |------|---------|-----------|------------| | import-in-the-middle | 3.5.1 | 127.66 kB | 531.94 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-datadog-prod-us1-2]

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

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🔄 Datadog retried 3 tests - 3 passed on retry

🎯 Code Coverage (details)
• Patch Coverage: 99.40%
• Overall Coverage: 98.27% (+0.03%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 306e37f | Docs | View more details | Give us feedback!}

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-10-06 18:45:55

Comparing candidate commit 306e37f in PR branch v6.20.0-proposal with baseline commit ee31dc7 in branch v6.x.

📊 Benchmarking dashboard

Found 29 performance improvements and 14 performance regressions! Performance is the same for 2284 metrics, 76 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 ----------------------------------'

scenario:llmobs-encode-unicode-ascii-24

• 🟩 cpu_user_time [-136.876ms; -28.281ms] or [-11.967%; -2.473%]

scenario:llmobs-evaluations-categorical-20

• 🟩 cpu_user_time [-240.676ms; -97.212ms] or [-9.366%; -3.783%]
• 🟩 execution_time [-240.848ms; -96.552ms] or [-9.310%; -3.732%]
• 🟩 throughput [+2157.436op/s; +5590.116op/s] or [+3.459%; +8.962%]

scenario:llmobs-evaluations-categorical-24

• 🟩 instructions [-2.3G instructions; -1.0G instructions] or [-13.593%; -5.699%]

scenario:llmobs-evaluations-reasoned-mixed-20

• 🟩 cpu_user_time [-181.223ms; -75.147ms] or [-7.587%; -3.146%]
• 🟩 execution_time [-185.689ms; -77.032ms] or [-7.705%; -3.196%]
• 🟩 instructions [-1396.2M instructions; -581.1M instructions] or [-6.850%; -2.851%]
• 🟩 throughput [+401.491op/s; +968.858op/s] or [+3.207%; +7.740%]

scenario:llmobs-evaluations-reasoned-mixed-24

• 🟥 max_rss_usage [+2.285MB; +5.555MB] or [+3.137%; +7.626%]
• 🟩 cpu_user_time [-219.205ms; -91.013ms] or [-9.868%; -4.097%]
• 🟩 execution_time [-212.879ms; -87.138ms] or [-9.479%; -3.880%]
• 🟩 instructions [-1.5G instructions; -0.6G instructions] or [-7.887%; -3.077%]
• 🟩 throughput [+522.876op/s; +1285.577op/s] or [+3.884%; +9.549%]

scenario:llmobs-evaluations-reasoned-mixed-26

• 🟥 max_rss_usage [+2.389MB; +5.712MB] or [+3.258%; +7.791%]
• 🟩 cpu_user_time [-219.309ms; -91.365ms] or [-11.113%; -4.630%]
• 🟩 execution_time [-222.289ms; -91.397ms] or [-11.064%; -4.549%]
• 🟩 instructions [-2.0G instructions; -0.8G instructions] or [-11.126%; -4.755%]
• 🟩 throughput [+688.262op/s; +1679.083op/s] or [+4.563%; +11.133%]

scenario:llmobs-evaluations-score-20

• 🟩 cpu_user_time [-241.052ms; -98.932ms] or [-9.991%; -4.100%]
• 🟩 execution_time [-244.372ms; -99.981ms] or [-10.046%; -4.110%]
• 🟩 throughput [+2561.760op/s; +6269.189op/s] or [+4.121%; +10.086%]

scenario:llmobs-evaluations-score-24

• 🟩 instructions [-2.2G instructions; -0.9G instructions] or [-13.712%; -5.378%]

scenario:openfeature-scale-full-20

• 🟥 max_rss_usage [+3.331MB; +17.653MB] or [+2.267%; +12.014%]

scenario:openfeature-stress-full-20

• 🟥 cpu_user_time [+218.073ms; +1060.092ms] or [+2.452%; +11.918%]
• 🟥 execution_time [+225.458ms; +1085.743ms] or [+2.290%; +11.030%]
• 🟥 throughput [-6.787op/s; -1.412op/s] or [-11.060%; -2.300%]

scenario:openfeature-stress-full-24

• 🟥 cpu_user_time [+210.319ms; +1032.896ms] or [+2.447%; +12.018%]
• 🟥 execution_time [+216.634ms; +1043.436ms] or [+2.266%; +10.912%]
• 🟥 max_rss_usage [+7.703MB; +37.539MB] or [+2.477%; +12.070%]
• 🟥 throughput [-6.874op/s; -1.429op/s] or [-10.874%; -2.260%]

scenario:openfeature-stress-full-26

• 🟥 cpu_user_time [+205.783ms; +1003.651ms] or [+2.399%; +11.698%]
• 🟥 execution_time [+208.874ms; +1028.348ms] or [+2.394%; +11.789%]
• 🟥 throughput [-8.043op/s; -1.640op/s] or [-11.585%; -2.362%]

scenario:openfeature-typical-full-20

• 🟥 instructions [+0.9G instructions; +4.5G instructions] or [+2.417%; +11.550%]

scenario:propagation-inject-replacements-20

• 🟩 cpu_user_time [-1408.639ms; -285.421ms] or [-11.610%; -2.352%]
• 🟩 execution_time [-1413.429ms; -285.534ms] or [-11.634%; -2.350%]
• 🟩 instructions [-11.5G instructions; -2.4G instructions] or [-11.407%; -2.387%]
• 🟩 throughput [+2897.318op/s; +14480.808op/s] or [+2.333%; +11.661%]

scenario:propagation-inject-replacements-24

• 🟩 cpu_user_time [-1244.850ms; -256.965ms] or [-10.794%; -2.228%]
• 🟩 execution_time [-1248.447ms; -257.072ms] or [-10.805%; -2.225%]
• 🟩 instructions [-10.7G instructions; -2.2G instructions] or [-10.936%; -2.292%]
• 🟩 throughput [+2896.492op/s; +14116.446op/s] or [+2.219%; +10.816%]

Unstable benchmarks

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

scenario:appsec-iast-with-vulnerability-iast-enabled-default-config-20

• unstable max_rss_usage [-33.467MB; +29.823MB] or [-9.716%; +8.658%]

scenario:appsec-iast-with-vulnerability-iast-enabled-default-config-24

• unstable max_rss_usage [-17.580MB; +21.205MB] or [-4.802%; +5.793%]

scenario:debugger-line-probe-with-snapshot-minimal-26

• unstable cpu_user_time [-423.869ms; +500.738ms] or [-4.805%; +5.677%]

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

• unstable cpu_user_time [-955.006ms; +164.355ms] or [-11.079%; +1.907%]
• unstable execution_time [-959.202ms; +182.943ms] or [-9.818%; +1.873%]
• unstable throughput [-21.714op/s; +106.564op/s] or [-1.862%; +9.136%]

scenario:debugger-line-probe-without-snapshot-26

• unstable cpu_user_time [-1123.348ms; +1645.056ms] or [-13.011%; +19.053%]
• unstable execution_time [-1141.079ms; +1691.731ms] or [-11.684%; +17.322%]
• unstable throughput [-173.684op/s; +115.700op/s] or [-14.856%; +9.897%]

scenario:encoders-0.4-immediate-flush-20

• unstable max_rss_usage [-12.410MB; +7.825MB] or [-14.651%; +9.238%]

scenario:llmobs-encode-unicode-ascii-24

• unstable execution_time [-145.799ms; -29.848ms] or [-12.586%; -2.577%]
• unstable throughput [+546.344op/s; +2659.373op/s] or [+2.604%; +12.676%]

scenario:llmobs-encode-unicode-ascii-26

• unstable cpu_user_time [-185.970ms; -38.462ms] or [-20.672%; -4.275%]
• unstable execution_time [-190.641ms; -39.563ms] or [-20.636%; -4.282%]
• unstable instructions [-1593.5M instructions; -334.0M instructions] or [-22.278%; -4.670%]
• unstable throughput [+1134.005op/s; +5498.731op/s] or [+4.262%; +20.665%]

scenario:llmobs-encode-unicode-mixed-20

• unstable cpu_user_time [-1.822s; -0.380s] or [-16.472%; -3.435%]
• unstable execution_time [-1.824s; -0.380s] or [-16.429%; -3.426%]
• unstable instructions [-17.1G instructions; -3.6G instructions] or [-17.853%; -3.744%]
• unstable throughput [+74.777op/s; +359.493op/s] or [+3.422%; +16.451%]

scenario:llmobs-encode-unicode-mixed-24

• unstable cpu_user_time [-2.054s; -0.428s] or [-19.458%; -4.051%]
• unstable execution_time [-2.062s; -0.429s] or [-19.461%; -4.048%]
• unstable instructions [-18.3G instructions; -4.3G instructions] or [-19.388%; -4.510%]
• unstable throughput [+92.948op/s; +447.882op/s] or [+4.042%; +19.479%]

scenario:llmobs-encode-unicode-mixed-26

• unstable cpu_user_time [-1.741s; -0.364s] or [-18.278%; -3.826%]
• unstable execution_time [-1.747s; -0.365s] or [-18.207%; -3.805%]
• unstable instructions [-17.5G instructions; -3.7G instructions] or [-19.667%; -4.123%]
• unstable throughput [+96.473op/s; +461.782op/s] or [+3.806%; +18.216%]

scenario:llmobs-evaluations-categorical-24

• unstable cpu_user_time [-385.020ms; -161.400ms] or [-22.110%; -9.269%]
• unstable execution_time [-386.588ms; -158.225ms] or [-21.932%; -8.976%]
• unstable throughput [+8387.152op/s; +20664.948op/s] or [+8.978%; +22.120%]

scenario:llmobs-evaluations-categorical-26

• unstable cpu_user_time [-421.465ms; -176.647ms] or [-27.253%; -11.422%]
• unstable execution_time [-451.307ms; -181.984ms] or [-28.433%; -11.465%]
• unstable instructions [-3.4G instructions; -1.3G instructions] or [-24.752%; -9.704%]
• unstable throughput [+12271.048op/s; +29679.284op/s] or [+11.617%; +28.097%]

scenario:llmobs-evaluations-score-24

• unstable cpu_user_time [-378.002ms; -158.445ms] or [-22.746%; -9.534%]
• unstable execution_time [-378.798ms; -158.627ms] or [-22.612%; -9.469%]
• unstable throughput [+8765.070op/s; +20961.935op/s] or [+9.497%; +22.711%]

scenario:llmobs-evaluations-score-26

• unstable cpu_user_time [-405.783ms; -169.836ms] or [-27.743%; -11.611%]
• unstable execution_time [-408.166ms; -170.851ms] or [-27.576%; -11.543%]
• unstable instructions [-3.3G instructions; -1.4G instructions] or [-25.401%; -10.656%]
• unstable throughput [+12255.282op/s; +29349.041op/s] or [+11.572%; +27.713%]

scenario:openfeature-scale-full-20

• unstable cpu_user_time [+142.896ms; +965.516ms] or [+1.756%; +11.863%]

scenario:openfeature-scale-full-24

• unstable cpu_user_time [+0.951s; +4.541s] or [+11.976%; +57.207%]
• unstable execution_time [+0.956s; +4.563s] or [+11.808%; +56.384%]
• unstable instructions [+3.1G instructions; +15.0G instructions] or [+5.707%; +27.338%]
• unstable max_rss_usage [-54.718MB; -11.267MB] or [-18.699%; -3.850%]
• unstable throughput [-2067.195op/s; -433.105op/s] or [-56.587%; -11.856%]

scenario:openfeature-scale-full-26

• unstable cpu_user_time [+0.874s; +4.171s] or [+11.417%; +54.508%]
• unstable execution_time [+0.884s; +4.220s] or [+11.338%; +54.110%]
• unstable instructions [+3.1G instructions; +14.9G instructions] or [+5.693%; +27.191%]
• unstable throughput [-2036.439op/s; -426.655op/s] or [-54.282%; -11.373%]

scenario:openfeature-stress-full-20

• unstable max_rss_usage [+5.096MB; +24.445MB] or [+3.257%; +15.625%]

scenario:openfeature-stress-full-26

• unstable max_rss_usage [+6.248MB; +30.093MB] or [+2.805%; +13.509%]

scenario:openfeature-typical-20

• unstable max_rss_usage [+3.918MB; +18.872MB] or [+3.039%; +14.639%]

scenario:openfeature-typical-24

• unstable cpu_user_time [+259.256ms; +1522.773ms] or [+2.647%; +15.546%]
• unstable execution_time [+278.484ms; +1609.745ms] or [+2.795%; +16.157%]
• unstable throughput [-8987.394op/s; -1612.758op/s] or [-16.081%; -2.886%]

scenario:openfeature-typical-26

• unstable max_rss_usage [+5.004MB; +24.241MB] or [+2.786%; +13.496%]

scenario:openfeature-typical-full-20

• unstable cpu_user_time [+267.895ms; +1599.527ms] or [+3.041%; +18.157%]
• unstable execution_time [+0.335s; +1.846s] or [+3.588%; +19.791%]
• unstable max_rss_usage [-0.146MB; +23.455MB] or [-0.108%; +17.381%]
• unstable throughput [-5288.820op/s; -981.788op/s] or [-19.801%; -3.676%]

scenario:openfeature-typical-full-24

• unstable cpu_user_time [+0.525s; +2.565s] or [+5.790%; +28.278%]
• unstable execution_time [+0.615s; +2.990s] or [+6.523%; +31.709%]
• unstable instructions [+1.6G instructions; +7.6G instructions] or [+3.697%; +17.787%]
• unstable throughput [-15856.431op/s; -3219.785op/s] or [-31.888%; -6.475%]

scenario:openfeature-typical-full-26

• unstable cpu_user_time [+0.440s; +2.240s] or [+4.818%; +24.550%]
• unstable execution_time [+0.535s; +2.666s] or [+5.653%; +28.174%]
• unstable instructions [+1.4G instructions; +6.7G instructions] or [+3.120%; +15.153%]
• unstable max_rss_usage [+5.560MB; +26.937MB] or [+3.063%; +14.838%]
• unstable throughput [-13787.057op/s; -2820.088op/s] or [-28.076%; -5.743%]

scenario:propagation-inject-replacements-26

• unstable cpu_user_time [-1359.090ms; -280.617ms] or [-12.670%; -2.616%]
• unstable execution_time [-1364.574ms; -279.358ms] or [-12.609%; -2.581%]
• unstable instructions [-12.5G instructions; -2.6G instructions] or [-13.007%; -2.722%]
• unstable throughput [+3596.946op/s; +17624.468op/s] or [+2.577%; +12.627%]

[codecov]

Codecov Report

❌ Patch coverage is 99.95224% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.86%. Comparing base (ee31dc7) to head (306e37f).
⚠️ Report is 39 commits behind head on v6.x.

Files with missing lines	Patch %	Lines
.../src/openfeature/writers/flag-evaluation-worker.js	97.40%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             v6.x   #10603      +/-   ##
==========================================
+ Coverage   98.84%   98.86%   +0.01%     
==========================================
  Files        1045     1058      +13     
  Lines      168309   171887    +3578     
  Branches       74       74              
==========================================
+ Hits       166365   169936    +3571     
- Misses       1944     1951       +7     

Flag	Coverage Δ
ai-guard	63.75% <44.69%> (-0.40%)	⬇️
apm-capabilities	54.72% <50.23%> (-0.07%)	⬇️
apm-integrations	80.98% <61.23%> (-0.26%)	⬇️
appsec	57.88% <59.51%> (+0.03%)	⬆️
debugger	69.87% <67.76%> (-0.03%)	⬇️
instrumentation	55.36% <57.23%> (+0.04%)	⬆️
llmobs	80.70% <92.70%> (+0.27%)	⬆️
openfeature	67.33% <77.54%> (+1.28%)	⬆️
platform	68.54% <56.55%> (-0.19%)	⬇️
profiling	65.98% <44.42%> (-0.44%)	⬇️
serverless	65.59% <57.01%> (-0.12%)	⬇️
test-optimization	81.85% <65.01%> (-0.25%)	⬇️

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.

[vjfridge]

LGTM for Feature Flags! #10464 and #10545 look correct in the release notes. We validated #10464’s final commit against master using system-tests and ffe-dogfooding, with no observed regression in regular evaluations and passing EVP/privacy checks.

Leaving overall release approval to the JS team 🙇‍♀️

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 306e37fff8

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]

Preserve undefined manual sampling tags in addTags

manual.keep and manual.drop intentionally treat any value other than false as enabled, including undefined, but these checks now suppress setPriorityFromTags for that valid input. Consequently, span.addTags({ 'manual.keep': undefined }) (including an options-tags object) leaves the trace on its automatic decision and can drop a trace that the caller explicitly requested to keep; previously the own-property check invoked the sampler for this case.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Add ESM rewrites for AI SDK v7 runtime context

These new v7 descriptors only target dist/index.js. Applications importing ai as ESM execute dist/index.mjs—which the existing v4–v6 descriptors explicitly register separately—so their generateText and streamText calls never publish the includeRuntimeContext channel. As a result, v7 ESM users with runtimeContext silently lose those metadata fields from the emitted LLM observability spans.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Redact dynamically selected sensitive fields

When a log template or capture expression uses a dynamic lookup such as user[key] and key resolves to password or token, this helper only checks whether the variable name key is sensitive. It returns no redaction error, so the compiled expression evaluates the lookup and sends its string result verbatim instead of passing it through inspectSegment; this leaks values covered by the default or configured identifier list. Treat dynamic member/index keys as unsafe or check their resolved key before rendering.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-06T21:30:58.627384Z	306e37f	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to 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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.