Improve OpenAI evaluator telemetry coverage with gpt-5.5 high

[mltseva]

id: 3a3e581c-f07f-4ea6-96a0-fb0f23cd31ea
artifact_path: experiments/tracy-openai/gpt-5.5-high/
providers: ["openai"]
status: succeeded
score: 91
checks: 1608 / 1794
passed scenarios: 37 / 86
provider_errors: 4

Category summary:

openai/batches avg 100.0
openai/audio avg 98.4
openai/conversations avg 98.5
openai/videos avg 96.7
openai/files avg 99.0
openai/embeddings avg 94.3
openai/images avg 86.9
openai/responses avg 85.9
openai/chat avg 81.3

Worst scenarios:

30 openai/chat/lifecycle provider_error=true
40 openai/chat/vision provider_error=true
46 openai/responses/streaming
60 openai/responses/background_cancel
61 openai/images/variation provider_error=true
63 openai/chat/streaming
70 openai/models/retrieve

Main remaining gaps:

• chat/lifecycle still fails because retrieve returns 404, then update/messages/list/delete spans are missing.
• chat/vision fails with provider-side image download 400.
• images/variation returns 404.
• videos/lifecycle mostly passes but delete returns 400 because the video is still processing.
• A few operation-name mismatches remain for Responses, for example responses.input_items.list vs expected response.input_items.list.
• Some checks still miss status/usage/response metadata in error or streaming paths.