judgevet.adapters.outbound.response_translation

Translate a raw Jev response into a parsed answer or a JevError.

The functions take a status code, the raw body bytes and a status line, so they need no HTTP client. An HTTP adapter passes response.content and the string form of its status error.

The API error detail field is polymorphic:

• Array for validation errors (422): [{"type", "loc", "msg", "input"}]
• Object for auth errors (401/403): {"error_type", "message"}
• Object for an oversized request (400 observed): {"error_type"} equal to "max_tokens_exceeded", returned as JevMaxTokensExceededError
• Absent, malformed or unrecognised: falls back to the status line alone.

The input key in validation errors contains the caller's request payload and is deliberately excluded from error messages to avoid leaking user data.

Examples:

from judgevet import JevError
from judgevet.adapters.outbound.response_translation import (
    parse_success,
    translate_status,
)

error = translate_status(302, b"", "Found")
assert isinstance(error, JevError)

body = (
    b'{"model": "jev-1.13.0", "answers": {}, '
    b'"usage": {"input_tokens": 10, "output_tokens": 0}}'
)
parsed = parse_success(200, body, {})
assert parsed.model == "jev-1.13.0"