Run Tests

Run benchmark problem tests using the eval-snapshot command instead of raw pytest. This ensures tests run in the correct Docker environment with proper isolation.

Usage: /run-tests <snapshot_path> <problem_name> <checkpoint_index>

Example: /run-tests outputs/run_001/submissions/file_backup/checkpoint_2/snapshot file_backup checkpoint_2

Command

bash

slop-code --quiet eval-snapshot {snapshot_path} \
  -p {problem_name} \
  -c {checkpoint_index} \
  -e configs/environments/docker-python3.12-uv.yaml \
  -o /tmp/eval-output \
  --json

Parameters:

•snapshot_path: Path to the solution directory to test
•problem_name: Name of the problem (e.g., file_backup, execution_server)
•checkpoint_index: Checkpoint to evaluate (e.g., checkpoint_1, checkpoint_2)

Output Location

Results are saved to the output directory (/tmp/eval-output by default):

code

/tmp/eval-output/
├── evaluation.json          # Structured test results
├── evaluation.log           # Detailed execution log
└── quality_analysis/        # Code quality metrics
    ├── ast_grep.jsonl       # AST-grep rule matches
    ├── files.jsonl          # File-level metrics
    ├── overall_quality.json # Aggregated quality scores
    └── symbols.jsonl        # Symbol/function metrics

Reading Results

evaluation.json Structure

json

{
  "problem_name": "eve_industry",
  "checkpoint_name": "checkpoint_3",
  "duration": 34.71,
  "entrypoint": "uv run industry.py",
  "tests": [
    {
      "id": "test_naga",
      "checkpoint": "checkpoint_1",
      "group_type": "Regression",
      "status": "passed",
      "duration_ms": 1029.27,
      "file_path": ".evaluation_tests/test_checkpoint_1.py"
    }
  ],
  "pass_counts": {
    "Regression": 25,
    "Core": 5,
    "Functionality": 5
  },
  "total_counts": {
    "Regression": 26,
    "Core": 5,
    "Functionality": 5
  },
  "pytest_exit_code": 1,
  "pytest_collected": 36,
  "infrastructure_failure": false
}

Key Fields

Field	Description
`tests`	Array of individual test results
`pass_counts`	Passed tests by group type
`total_counts`	Total tests by group type
`pytest_exit_code`	0 = all passed, 1 = some failed
`infrastructure_failure`	True if environment setup failed

Test Group Types

Group	Description
`Core`	Must pass for checkpoint to pass
`Functionality`	Additional coverage (optional)
`Regression`	Tests from prior checkpoints
`Error`	Error handling tests

Interpreting Results

Quick Summary

bash

# Parse with jq to get summary
cat /tmp/eval-output/evaluation.json | jq '{
  passed: .pass_counts,
  total: .total_counts,
  exit_code: .pytest_exit_code
}'

Finding Failed Tests

bash

# List failed tests
cat /tmp/eval-output/evaluation.json | jq '.tests[] | select(.status == "failed") | .id'

Check Pass/Fail by Group

bash

# Core tests (must all pass)
cat /tmp/eval-output/evaluation.json | jq '.pass_counts.Core == .total_counts.Core'

Common Workflows

Run and Check Status

bash

slop-code --quiet eval-snapshot ./snapshot \
  -p file_backup -c checkpoint_1 \
  -e configs/environments/docker-python3.12-uv.yaml \
  -o /tmp/eval-output --json

# Check if passed
if [ $(cat /tmp/eval-output/evaluation.json | jq '.pytest_exit_code') -eq 0 ]; then
  echo "All tests passed!"
else
  echo "Some tests failed"
  cat /tmp/eval-output/evaluation.json | jq '.tests[] | select(.status == "failed")'
fi

Run Multiple Checkpoints

bash

for checkpoint in checkpoint_1 checkpoint_2 checkpoint_3; do
  echo "=== $checkpoint ==="
  slop-code --quiet eval-snapshot ./submissions/$checkpoint/snapshot \
    -p my_problem -c $checkpoint \
    -e configs/environments/docker-python3.12-uv.yaml \
    -o /tmp/eval-$checkpoint --json
  cat /tmp/eval-$checkpoint/evaluation.json | jq '.pass_counts'
done

Troubleshooting

Infrastructure Failure

If infrastructure_failure: true:

•Docker may not be running
•Image build failed
•Check evaluation.log for details

Tests Not Found

If pytest_collected: 0:

•Problem name may be wrong
•Checkpoint doesn't exist
•Test files missing from problem directory

Timeout Issues

Default timeout is 180s per test. For long-running tests, this is controlled in the problem's pytest config.

Notes

•Always use eval-snapshot instead of raw pytest for benchmark problems
•Tests run in isolated Docker containers
•Results include both correctness and quality metrics
•Use --json flag to get machine-readable output
•The -o flag specifies where to save results