[bot] Mistral `classifiers.moderate()` and related classification endpoints not instrumented

[braintrust-bot]

Summary

The @mistralai/mistralai TypeScript SDK exposes a classifiers resource with four model-execution endpoints for content moderation and classification. None are instrumented — no channels, no plugin handler, no wrapper proxy, and no auto-instrumentation config. Calls to any classifiers.* method produce no Braintrust spans.

This is directly comparable to OpenAI's moderations.create(), which is fully instrumented across all layers in this repo (wrapper, auto-instrumentation, channel, plugin).

What instrumentation is missing

SDK Method	Description	Model
client.classifiers.moderate()	Moderate text content against policy categories	mistral-moderation-2603
client.classifiers.moderateChat()	Moderate chat conversation content	mistral-moderation-2603
client.classifiers.classify()	Classify text with custom/system categories	Classification model
client.classifiers.classifyChat()	Classify chat content with custom/system categories	Classification model

All four are model inference endpoints — they accept input, run a model, and return generated classification/moderation scores. They are not CRUD or admin endpoints.

No coverage in any instrumentation layer:

• Wrapper (js/src/wrappers/mistral.ts, lines 78–88): The mistralProxy switch only handles chat, fim, agents, and embeddings. No classifiers case.
• Auto-instrumentation config (js/src/auto-instrumentations/configs/mistral.ts): No config for any Classifiers.* methods. Only Chat, Fim, Agents, and Embeddings classes are configured.
• Channels (js/src/instrumentation/plugins/mistral-channels.ts): No channel definitions for classifier endpoints.
• Plugin (js/src/instrumentation/plugins/mistral-plugin.ts): No handler for classifier calls.
• Vendor types (js/src/vendor-sdk-types/mistral.ts, lines 181–187): MistralClient type only declares chat, fim, agents, embeddings. No classifiers property.

A grep for classif or moderate across js/src/wrappers/mistral.ts and js/src/instrumentation/plugins/mistral-plugin.ts returns zero matches.

Precedent in this repo

OpenAI moderations.create() — the directly comparable moderation API from a different provider — is fully instrumented:

Layer	OpenAI Moderations	Mistral Classifiers
Wrapper	oai.ts (moderationProxy)	Missing
Auto-instrumentation	openai.ts config	Missing
Channel	moderationsCreate	Missing
Plugin handler	openai-plugin.ts	Missing
Vendor types	OpenAIModerationCreateParams / Response	Missing

Braintrust docs status

not_found — The Braintrust Mistral integration page at https://www.braintrust.dev/docs/integrations/ai-providers/mistral documents chat completions, FIM, embeddings, agents, audio, and OCR. Classifiers/moderation are not mentioned.

Upstream references

• Mistral Guardrailing docs: https://docs.mistral.ai/capabilities/guardrailing/
• Mistral API classifiers endpoints: POST /v1/moderations, POST /v1/chat/moderations, POST /v1/classifications, POST /v1/chat/classifications
• Mistral TypeScript SDK: https://github.com/mistralai/client-ts — client.classifiers resource
• Model: mistral-moderation-2603 (dedicated moderation model)

Local files inspected

• js/src/wrappers/mistral.ts — lines 78–88: proxy switch (no classifiers case)
• js/src/auto-instrumentations/configs/mistral.ts — full file: no classifier configs
• js/src/instrumentation/plugins/mistral-channels.ts — no classifier channels
• js/src/instrumentation/plugins/mistral-plugin.ts — no classifier handlers
• js/src/vendor-sdk-types/mistral.ts — lines 181–187: MistralClient missing classifiers
• e2e/scenarios/mistral-instrumentation/ — no classifier test scenarios