Add EQ-Bench3 synthetic training environment

[poofeth]

Summary

• add an eq_bench3 environment for EQ-Bench-style emotional intensity prediction
• include a deterministic generator for uncontaminated synthetic dialogue/emotion prompts
• commit a 64-row HF Dataset-compatible JSONL sample with question, answer, and info fields
• add deterministic scoring for JSON emotion-intensity predictions
• add focused tests covering generator determinism, dataset loading, reward scoring, and environment construction without external API calls

Bounty

• Algora: https://algora.io/PrimeIntellect-ai/bounties/UZGP8mWodUUMpUM8
• Claiming the EQ-Bench3 + train set bounty.

Validation

• uv run pytest tests/test_eq_bench3_environment.py -q
• uv run ruff check environments/eq_bench3 tests/test_eq_bench3_environment.py
• uv run ruff format --check environments/eq_bench3 tests/test_eq_bench3_environment.py
• git diff --check

---

Note

Low Risk
Low risk: changes are additive (new eq_bench3 environment, sample data, and tests) with no modifications to core framework or security-sensitive code paths.

Overview
Adds a new environments/eq_bench3 SingleTurn environment for EQ-Bench-style emotion-intensity prediction, including JSON parsing + a continuous reward (emotion_score_reward) that scores per-emotion intensity error.

Includes a deterministic prompt generator script and commits a 64-row synthetic JSONL sample dataset, plus packaging metadata (pyproject.toml) and focused tests for generator determinism, dataset loading, reward correctness, and load_environment construction.

Updates environments/README.md to list eq_bench3 among available SingleTurn examples.

^{Reviewed by Cursor Bugbot for commit 32cf899. Bugbot is set up for automated code reviews on this repo. Configure here.}

[poofeth]

Validation evidence for the EQ-Bench3 bounty PR:

• uv run pytest tests/test_eq_bench3_environment.py -q passed: 4 tests.
• uv run ruff check environments/eq_bench3 tests/test_eq_bench3_environment.py passed.
• uv run ruff format --check environments/eq_bench3 tests/test_eq_bench3_environment.py passed.
• git diff --check passed.

The sample dataset is generated from local deterministic templates (source=synthetic-uncontaminated-v1) rather than copied from the upstream EQ-Bench question set.

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 1698173. Configure here.}

[poofeth]

Addressed the Bugbot review in commit 32cf899:

• added eq_bench3 to environments/README.md
• added an upper-bound guard for synthetic generation (num_examples <= 96) to avoid duplicate-exhaustion loops
• added regression coverage for the generator bound

Validation after the fixes:

$ uv run pytest tests/test_eq_bench3_environment.py -q
.....                                                                    [100%]

$ uv run ruff check environments/eq_bench3 tests/test_eq_bench3_environment.py
All checks passed!

$ uv run ruff format --check environments/eq_bench3 tests/test_eq_bench3_environment.py
3 files already formatted

$ git diff --check
# no output

[poofeth]

/claim https://algora.io/PrimeIntellect-ai/bounties/UZGP8mWodUUMpUM8