Remove deprecated Hosted Training val config surface

[zahlmann]

Summary

• remove the deprecated Hosted Training [val] block from prime train init
• stop exposing val in prime train configs and let schema validation reject old [val] configs
• stop CLI-created runs from sending a val payload and avoid regenerating [val] from legacy run metadata

Context

I had Codex do this sweep after Prime support told me [val] is deprecated and [eval] should be used instead.

prime-rl removed orchestrator.val in PrimeIntellect-ai/prime-rl#2193 and recommends registering validation-style checks as eval environments. This aligns the Prime CLI config surface with [eval] / [[eval.env]].

Tests

• uv run --package prime pytest packages/prime/tests/test_rl_config.py packages/prime/tests/test_train_cli.py packages/prime/tests/test_rl_api.py -q
• uv run --package prime pytest packages/prime/tests/test_lab_view.py -q -k 'training_config or config_builder_preserves_deprecated_val or train_configs or shared_eval'
• uv run --package prime ruff check packages/prime/src/prime_cli/commands/rl.py packages/prime/src/prime_cli/api/rl.py packages/prime/src/prime_lab_app/training_config.py packages/prime/tests/test_rl_config.py packages/prime/tests/test_train_cli.py packages/prime/tests/test_lab_view.py

---

Note

Low Risk
Config-surface removal with tests; run reads may still expose legacy valConfig on RLRun, but new runs no longer send val.

Overview
Removes the deprecated Hosted Training [val] block from the user-facing config path so validation-style checks align with [eval] / [[eval.env]].

CLI & API: Drops ValConfig, the val field on RLConfig, template comments, run-summary output, and val_config on RLClient.create_run (no val key in create payloads). RLConfig with extra="forbid" now rejects TOML that still contains [val].

Lab: training_config no longer maps legacy val_config into exported TOML; rerun copies omit [val] even when old run metadata had it.

Tests lock in: init template and prime train configs omit val, load_config fails on [val], lab export strips val_config, and the config builder still surfaces [val] only to produce a clear schema error.

^{Reviewed by Cursor Bugbot for commit 1d273ea. Bugbot is set up for automated code reviews on this repo. Configure here.}

[zahlmann]

Docs note from the sweep: the public Hosted Training docs still show [val] in Advanced Configurations and End-to-End Training Run, but I couldn't find those Mintlify source pages in this repo, prime-rl, or verifiers. Those docs likely need a parallel update to point users to [eval] / [[eval.env]].

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ff49734cab

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".