Repository navigation
docs(langchain): Document TypeSafeClassifier evaluation tracing - #19879
Conversation
Co-Authored-By: Andrei Borza <[email protected]>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
1 Skipped Deployment
|
|
The change adds classifier evaluation tracing docs to three JavaScript integration pages. The plan checks client errors on those pages, site-wide client errors, and missing-page counts. An issue escalates when those page errors appear or the site error rate rises. Services: PlanWhat changedReaders of the LangChain, LangGraph, and TypeSafe JavaScript integration pages now see TypeSafeClassifier evaluation tracing. Those pages document RiskWrong MDX or PlatformSection markup can break render on those three pages. New code samples can throw in the shared code-tab client. New cross-links can 404. The same code-tab and platform-selector code runs on other JavaScript integration pages. Intended effectThis change is documentation text. Telemetry cannot show that the new sections are on the page. Treat intended effect as not queryable. Do not use a missing-error result as proof the text shipped. Regression watchWrong markup fails first as client errors on the changed integration pages. Shared code-tab and platform-selector failures show as a site-wide error-rate rise. Bad new links show as a missing-page-rate rise. Existing production errors are mostly TypeError and Event. Hold the measured rates. Do not treat that baseline noise as a new fault.
If the error rate rises, inspect code-tab and platform-selector errors on the changed pages. Not observableWhether the new prose is present on the live pages. Whether the documented SDK spans match production SDK behavior. Page-specific web vitals: the baseline window had 37 sampled pageload and navigation spans on those URLs, and p95 was 11.7 s. That sample is too small for a hold band. |
![sentry[bot]](https://avatars.githubusercontent.com/in/12637?s=60&v=4){kind=small}
Co-Authored-By: Andrei Borza <[email protected]>
Co-Authored-By: Andrei Borza <[email protected]>
Co-Authored-By: Andrei Borza <[email protected]>
This branch was successfully deployed
2 participants
DESCRIBE YOUR PR
Document tracing of
TypeSafeClassifierruns from@langchain/typesafe, which now show up asgen_ai.evaluatespans. This applies to standalone calls, calls from LangGraph nodes, and the classifiers that the@langchain/typesafeagent middleware creates.The LangChain page gets a new section with a setup example, the data captured, and privacy controls, plus the supported
@langchain/typesaferange. The LangGraph page gets a short section with aStateGraphnode example that points to the LangChain section. The TypeSafe page now links to it. This follows the Mastra classifier section from #19878. The new sections are hidden on Cloudflare because the SDK has no test coverage there.Shipped in Sentry JavaScript SDK 11.6.0 via getsentry/sentry-javascript#25120, with agent and LangGraph coverage in getsentry/sentry-javascript#25132.
IS YOUR CHANGE URGENT?
PRE-MERGE CHECKLIST
via Andrei Borza.
--
View Junior Session [Sentry]