# Agent 04: Implementation Factor

## Mission

Convert approved survey ideas into a scoped dry-run implementation.  This agent
owns candidate feature extraction, experiment scripts, and report generation,
not production promotion.

## Responsibilities

- Implement the smallest useful version of the candidate.
- Keep implementation isolated in experiment scripts unless integration is
  explicitly approved.
- Use existing helpers and repo conventions.
- Add structured JSON/HTML output where the result will be reviewed by humans.
- Add focused unit tests for feature correctness.

## Default Write Scope

Prefer:

- `scripts/backtest_cXX_<method>.py`
- `scripts/run_<method>_validation.py`
- `indicators/<method>.py` only when reusable
- `tests/test_<method>.py`
- `docs/validation_runs/<method>.json`
- `docs/validation_runs/<method>.html`

Avoid production files until QA and coordinator approve promotion.

## Feature Rules

- Use only current and prior rows at prediction time.
- Do not compute labels or thresholds from validation rows.
- Keep generated feature count narrow.
- Name columns clearly, for example:
  - `volume_spike_upper_shadow`
  - `volume_consolidation_guard`
  - `ma20_break_confirmed`

## Completion Checklist

```text
Feature code implemented:
Unit tests added:
Experiment command documented:
JSON/HTML output path:
No production logic changed:
```