chore(decisions): remove the System One converters and OpenAI spec types left dead by #45214

[devin-ai-integration]

TLDR

Problem this solves:

• feat(decisions): add OpenAI as a Decisions provider behind a shared decisions format #45214 moved both decision formats onto one shared translation
• The older System One converters and the OpenAI spec types stayed behind with no caller
• Two dead modules still cost lint, type checks, and reader attention

How it solves it:

• Deletes litellm/llms/base_llm/decisions/systemone.py and litellm/types/openai_decisions.py
• Deletes the two unit test files that only exercised those modules
• Changes no request path: nothing imported either module outside its own tests

User Flow

Before: a developer classifying a support ticket gets answers from both decision routes, and nothing here is broken

1. They send POST https://litellm-domain/v1/decisions with "model": "gpt-6-luna", an input string, and a questions list, and get 200 with one answer per question
2. They send POST https://litellm-domain/v1/systemone with "model": "typesafe/jev-1.13.0", a state string, and a questions map, and get 200 with one answer per question name

After: the same two calls answer the same way; the only change is on the maintainer's side, where two unused modules are gone

1. They send the same POST https://litellm-domain/v1/decisions and get the same 200 with one answer per question
2. They send the same POST https://litellm-domain/v1/systemone and get the same 200 with one answer per question name

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests (this PR only deletes tests that covered the deleted modules)
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/unit/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

This PR deletes code nothing imports, so the expected result is that both legs answer the same. Both did, on real OpenAI and real TypeSafe calls

Shared setup, config.yaml (the env holds OPENAI_API_KEY, TYPESAFE_API_KEY, and a random LITELLM_MASTER_KEY, and no DATABASE_URL):

model_list:
  - model_name: gpt-6-luna
    litellm_params:
      model: openai/gpt-6-luna
      api_key: os.environ/OPENAI_API_KEY
  - model_name: typesafe/jev-1.13.0
    litellm_params:
      model: typesafe/jev-1.13.0
      api_key: os.environ/TYPESAFE_API_KEY
general_settings:
  master_key: os.environ/LITELLM_MASTER_KEY

decisions_openai.json:

{"model": "gpt-6-luna", "input": "Ticket (billing): The export job hangs at 99% and never finishes", "questions": [{"type": "predicate", "name": "is_defect", "instructions": "Is this a defect?"}, {"type": "choice", "name": "severity", "instructions": "How severe is it?", "choices": [{"value": "low", "description": "cosmetic"}, {"value": "high", "description": "blocks users"}]}]}

systemone_typesafe.json:

{"model": "typesafe/jev-1.13.0", "state": "Ticket (billing): The export job hangs at 99% and never finishes", "questions": {"defect": {"type": "noul", "instructions": "Is this a defect?"}, "severity": {"type": "choice", "instructions": "How severe is it?", "criteria": {"low": "cosmetic", "high": "blocks users"}}}}

Before (0721cff)

Proxy booted from a worktree at 0721cffab2: python litellm/proxy/proxy_cli.py --config config.yaml --host 127.0.0.1 --port 42027 --num_workers 2 --timeout_worker_healthcheck 60, one instance, 2 uvicorn workers, no database

POST /v1/decisions on an OpenAI model (gpt-6-luna)

1. curl -sS -w '\nHTTP %{http_code}\n' -X POST http://127.0.0.1:42027/v1/decisions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" --data @decisions_openai.json
2. Observed:

{"model":"gpt-6-luna","answers":[{"type":"predicate","name":"is_defect","probability":0.84},{"type":"choice","name":"severity","choice":"high","probabilities":[{"value":"low","probability":0.02},{"value":"high","probability":0.98}],"confidence":0.96}],"usage":{"input_tokens":275,"input_tokens_details":{"cached_tokens":0,"cache_write_tokens":0},"output_tokens":0,"output_tokens_details":{"reasoning_tokens":0},"total_tokens":275}}
HTTP 200

POST /v1/systemone on a TypeSafe model (jev-1.13.0)

1. curl -sS -w '\nHTTP %{http_code}\n' -X POST http://127.0.0.1:42027/v1/systemone -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" --data @systemone_typesafe.json
2. Observed:

{"model":"jev-1.13.0","answers":{"defect":{"type":"noul","noul":0.87},"severity":{"type":"choice","choice":"high","confidence":1.0,"probabilities":{"high":1.0,"low":0.0}}},"usage":{"input_tokens":336,"output_tokens":48}}
HTTP 200

After (b9cbbe4)

Proxy booted from a worktree at b9cbbe4aa3: python litellm/proxy/proxy_cli.py --config config.yaml --host 127.0.0.1 --port 41758 --num_workers 2 --timeout_worker_healthcheck 60, one instance, 2 uvicorn workers, no database

POST /v1/decisions on an OpenAI model (gpt-6-luna)

1. curl -sS -w '\nHTTP %{http_code}\n' -X POST http://127.0.0.1:41758/v1/decisions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" --data @decisions_openai.json
2. Observed:

{"model":"gpt-6-luna","answers":[{"type":"predicate","name":"is_defect","probability":0.84},{"type":"choice","name":"severity","choice":"high","probabilities":[{"value":"low","probability":0.02},{"value":"high","probability":0.98}],"confidence":0.96}],"usage":{"input_tokens":275,"input_tokens_details":{"cached_tokens":0,"cache_write_tokens":0},"output_tokens":0,"output_tokens_details":{"reasoning_tokens":0},"total_tokens":275}}
HTTP 200

POST /v1/systemone on a TypeSafe model (jev-1.13.0)

1. curl -sS -w '\nHTTP %{http_code}\n' -X POST http://127.0.0.1:41758/v1/systemone -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H "Content-Type: application/json" --data @systemone_typesafe.json
2. Observed:

{"model":"jev-1.13.0","answers":{"defect":{"type":"noul","noul":0.87},"severity":{"type":"choice","choice":"high","confidence":0.99,"probabilities":{"low":0.0,"high":1.0}}},"usage":{"input_tokens":336,"output_tokens":48}}
HTTP 200

Observations:

• Both routes answer 200 with the same answer shape on both legs

• The TypeSafe confidence moved 1.0 to 0.99 between calls: provider variance, same on both legs

• /v1/decisions usage reports output_tokens: 0 on both legs; this PR leaves it alone

• b9cbbe4 passes /audit (skipped: no product code on a request path, a refactor deleting unimported modules)

Type

🧹 Refactoring

Caveats (if any)

Low

• No before=fail leg exists: the PR changes nothing a caller can see, so the proof is a same-answer A/B over both decision routes
• The deleted unit tests covered only the deleted modules; the live System One translation keeps its tests under tests/unit/llms/base_llm/decisions/ and tests/unit/llms/openai/decisions/
• litellm.types.openai_decisions and litellm.llms.base_llm.decisions.systemone stop importing; they only ever shipped in the dev tag v1.106.0-dev.2, never in a stable or rc release, and removing them is the point of this PR

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

• b9cbbe4 passes /live-pr-risk

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Confidence Score: 5/5

[Low impact] This PR appears safe to merge because live Decisions calls do not use the deleted modules

Summary

Removes unused System One converters, old OpenAI Decisions types, and their two test files

• Live Decisions calls already use the retained types and shared conversion code
• Retained tests cover both request formats and their replies
• No actionable issues found. No runtime tests were run

_{Reviews (1) · Last reviewed commit: "chore(decisions): remove the System One ..." · Reviewed by Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_remove_dead_systemone_converters (b9cbbe4) with main (4ed0267)}

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit b9cbbe4. Configure here.}

[mateo-berri]

LGTM