Quant-Bench

Use the attached task packet to construct a portfolio for masked assets.

The packet contains train_prices.parquet, train_returns.parquet, asset_metadata.csv, constraints.json, data_dictionary.json. Asset identifiers are intentionally masked. Use only the packet data and public constraints in that packet; do not try to infer real-world ticker identities.

Objective: Construct long-only masked-asset weights to maximize hidden-window Sharpe ratio, with full investment, 30% max weight, and at least 4 selected assets.

Hidden scoring objective: maximize_hidden_holdout_sharpe. The public training window is evidence, not the scoring window. The submitted weights are scored on a private holdout objective normalized against a public-constraint-aware baseline and a private hindsight oracle.

Rules:
- Include every packet asset in result.weights.
- Numeric weights are decimal fractions, not percentages.
- Obey constraints.json exactly; public constraints gate score validity.
- You may use any forecasting, shrinkage, risk model, or portfolio-construction method you think is appropriate.
- Optional audit or diagnostic notes may be included outside result, but they are ignored by the scorer.

Return a structured dashboard. In the dashboard markdown, include exactly one fenced code block labeled benchmark_result containing valid JSON:

```benchmark_result
{
  "task_id": "forecasting.masked_max_sharpe_portfolio.v1",
  "result": {
    "weights": {
      "ASSET_01": 0.0,
      "ASSET_02": 0.0,
      "ASSET_03": 0.0,
      "ASSET_04": 0.0,
      "ASSET_05": 0.0,
      "ASSET_06": 0.0,
      "ASSET_07": 0.0,
      "ASSET_08": 0.0,
      "ASSET_09": 0.0,
      "ASSET_10": 0.0,
      "ASSET_11": 0.0,
      "ASSET_12": 0.0
}
  }
}
```