[bot] OpenAI legacy text completions API (`client.completions`) not instrumented

[braintrust-bot]

Summary

The OpenAI legacy text completions API (/v1/completions) is available in both the official openai gem and the ruby-openai gem but is not instrumented. This is the original OpenAI generation API, still used today for fine-tuned instruct models (gpt-3.5-turbo-instruct, babbage-002, davinci-002) and other text completion workflows. It is distinct from the chat completions API (/v1/chat/completions), which is already instrumented.

What is missing

Official openai gem

OpenAI::Resources::Completions exposes two generative execution methods, neither of which is patched:

• client.completions.create(model:, prompt:, max_tokens:, temperature:, top_p:, n:, stop:, ...) — Synchronous text completion; returns a Completion object with choices[].text, usage, model, and id.
• client.completions.create_streaming(model:, prompt:, ...) — Streaming variant; yields raw SSE chunks aggregating into the same final structure.

ruby-openai gem

OpenAI::Client exposes the equivalent via:

• client.completions(parameters: { model:, prompt:, max_tokens:, ... }) — Calls /v1/completions directly. Accepts the same parameters as the official SDK.

What a span should capture

• Input: the prompt (text string or array of strings/tokens)
• Metadata: model, max_tokens, temperature, top_p, n, stop, frequency_penalty, presence_penalty, seed, logprobs, provider, endpoint (/v1/completions)
• Output: choices[].text and finish_reason for each completion choice
• Metrics: prompt_tokens, completion_tokens, total_tokens from the usage object; time_to_first_token

The pattern is identical to the existing ModerationsPatcher — a new CompletionsPatcher wrapping create() and create_streaming() on the respective resource class.

Braintrust docs status

not_found — The Braintrust OpenAI integration docs at https://www.braintrust.dev/docs/integrations/ai-providers/openai and the tracing guide at https://www.braintrust.dev/docs/guides/tracing focus exclusively on chat completions. The legacy /v1/completions endpoint is not mentioned.

Upstream sources

• Official OpenAI Ruby SDK: https://github.com/openai/openai-ruby — OpenAI::Resources::Completions with create and create_streaming methods in lib/openai/resources/completions.rb
• OpenAI Legacy Completions API docs: https://platform.openai.com/docs/api-reference/completions — documents the endpoint, all parameters, and usage fields
• ruby-openai gem: https://github.com/alexrudall/ruby-openai — client.completions(parameters: {}) documented in the README

Local repo files inspected

• lib/braintrust/contrib/openai/patcher.rb — defines ChatPatcher, ResponsesPatcher, ModerationsPatcher; no CompletionsPatcher for OpenAI::Resources::Completions
• lib/braintrust/contrib/openai/instrumentation/ — contains chat.rb, responses.rb, moderations.rb, common.rb; no completions.rb
• lib/braintrust/contrib/ruby_openai/patcher.rb — defines ChatPatcher, ResponsesPatcher, ModerationsPatcher; no CompletionsPatcher
• lib/braintrust/contrib/ruby_openai/instrumentation/ — contains chat.rb, responses.rb, moderations.rb, common.rb; no completions.rb
• test/fixtures/vcr_cassettes/alexrudall_ruby_openai/completions.yml — recorded cassette showing a POST https://api.openai.com/v1/completions call with gpt-3.5-turbo-instruct, confirming the endpoint is accessible via the ruby-openai gem
• Grep for Completions (outside of Chat::Completions), instruct, /v1/completions across lib/braintrust/ returns zero instrumentation matches