stock-pool-entry-signal-scan

name	stock-pool-entry-signal-scan
description	Use when generating training-window entry-signal-analysis datasets for a J-stock stock-pool experiment, especially scanning a broad/full universe with a fixed entry strategy before ticker-level pool selection.

Stock Pool Entry Signal Scan

Generate training data for ticker selection. This skill does not select tickers and does not run portfolio evaluation.

Command Shape

Use entry-signal-analysis for production-style entry signal quality:

uv run python main.py entry-signal-analysis `
  --entry-strategies <ENTRY_STRATEGY> `
  --universe-file <FULL_UNIVERSE_FILE> `
  --start <TRAIN_START> `
  --end <TRAIN_END> `
  --horizons 1 3 5 `
  --primary-horizons 3 5 `
  --primary-horizon 3 `
  --label-mode next_open `
  --ranking-strategy <RANKING_STRATEGY> `
  --entry-filter-mode <ENTRY_FILTER_MODE> `
  --output-dir "G:/My Drive/AI-Stock-Sync/entry_signal_analysis"

Prefer explicit --start and --end for fold-specific training windows. Use --years only when full calendar years are intended.

Parameter Rules

• Resolve production defaults from G:/My Drive/AI-Stock-Sync/config.json, especially production.strategy_groups and production.signal_ranking_strategy.
• Use the same ranking strategy and entry filter mode later used in OOS evaluation unless the experiment intentionally varies them.
• Do not include OOS/test dates in the scan.
• Keep the broad/full universe fixed for training scans across folds unless the experiment explicitly studies full-universe definitions.
• Use entry-exit-validation instead only when the user asks to select tickers by entry-plus-exit behavior.

Required Artifacts

After a run, locate and report:

• entry_signal_analysis_manifest.json
• entry_signal_analysis_candidates_*.csv
• entry_signal_analysis_selected_*.csv
• entry_signal_analysis_summary_*.json
• entry_signal_analysis_report_*.md

Use selected rows for pool selection unless the user asks to score all BUY candidates.

Validation

Check the manifest before handing off to ticker selection:

• start_date and end_date match the training fold.
• universe_size matches the expected broad/full universe.
• entry_strategies, ranking_strategy, entry_filter_mode, and horizons match the plan.
• Candidate and selected counts are nonzero, or the empty result is explicitly reported.