signal-scanner

name	signal-scanner
description	Signal scanning pipeline including Stage 2 analysis, VCP detection, sector strength, PEAD, forensic accounting, advisory model, quality gates, and strategy ensemble. Use when working on signal_scanner.py, stage_analysis.py, sector_strength.py, scan logic, signal ranking, quality gates, diagnostics, or any trading strategy code.

Signal Scanner

Two-Stage Pipeline

Stage A — Fast Structural Filters (Parallel)

Workers: SCAN_STAGE_A_MAX_WORKERS (default 4), timeout: SCAN_STAGE_TASK_TIMEOUT_SEC (120s)

1. Stage 2 Analysis: price within STAGE2_52W_PCT (85%) of 52W high + 200 SMA rising for STAGE2_SMA_UPWARD_DAYS (20) days
2. VCP Detection: VCP_DAYS (4) consecutive days with volume below 50-day average
3. Sector filter (optional): SECTOR_FILTER_ENABLED — stock's sector must outperform SPY

Output: shortlist of SIGNAL_TOP_N * SCAN_STAGE_A_SHORTLIST_MULTIPLIER tickers, capped at SCAN_STAGE_A_SHORTLIST_CAP (40).

Stage B — Enrichment (Parallel)

Workers: SCAN_STAGE_B_MAX_WORKERS (default 4)

1. PEAD: boost/penalty based on recent earnings surprise (PEAD_SCORE_BOOST, PEAD_SCORE_PENALTY)
2. Forensic accounting: Sloan ratio, Beneish M-Score, Altman Z-Score checks (FORENSIC_FILTER_MODE)
3. SEC enrichment: filing analysis, score hints (SEC_ENRICHMENT_ENABLED)
4. Advisory model: calibrated P(up in 10 days) overlay (ADVISORY_MODEL_ENABLED)
5. Quality gates: filter weak signals (QUALITY_GATES_MODE: off/shadow/soft/hard)
6. Strategy ensemble: breakout + pullback weighted by regime (STRATEGY_ENSEMBLE_MODE)
7. Breakout confirmation: optional intraday price check (BREAKOUT_CONFIRM_ENABLED)

Regime Gate

Before scanning, checks SPY vs 200 SMA. If SPY is below and SCAN_ALLOW_BEAR_REGIME=false (default), scan is blocked entirely. Diagnostics will show scan_blocked_reason: bear_regime_spy_below_200sma.

Diagnostics Dict

Every scan returns (signals, diagnostics). Key diagnostic counters:

diagnostics = {
    "scan_blocked": 0,
    "scan_blocked_reason": None,
    "watchlist_size": 0,
    "df_empty": 0,
    "too_few_candles": 0,
    "stage2_fail": 0,
    "vcp_fail": 0,
    "no_sector_etf": 0,
    "sector_not_winning": 0,
    "breakout_not_confirmed": 0,
    "exceptions": 0,
    "self_study_filtered": 0,
    # ...quality gate and plugin counters
}

Always increment relevant counter when rejecting a ticker. This powers the dashboard's "why no signals?" explanation.

Signal Ranking

Composite score from: Stage 2 proximity, VCP quality, sector RS, PEAD, guidance tone, SEC hints, advisory P(up). Sorted descending, top SIGNAL_TOP_N (default 5, 0=unlimited) returned.

Each signal carries strategy_attribution.top_live labeling the dominant strategy.

Quality Gates

Mode	Behavior
off	Disabled
shadow	Track would-filter counts only
soft	Filter when QUALITY_SOFT_MIN_REASONS (2) weak reasons exist
hard	Filter on any single weak reason

Exception: weak_breakout_volume is always hard regardless of mode.

Plugin Modes (OFF → SHADOW → LIVE)

All plugins follow the same pattern. Current plugins:

• Execution Quality (EXEC_QUALITY_MODE): spread/slippage checks, limit orders
• Exit Manager (EXIT_MANAGER_MODE): partial TP, breakeven stops, time stops
• Event Risk (EVENT_RISK_MODE): earnings/macro blackout windows
• Regime v2 (REGIME_V2_MODE): score-based sizing multipliers
• Correlation Guard (CORRELATION_GUARD_MODE): pairwise correlation limits
• Strategy Ensemble (STRATEGY_ENSEMBLE_MODE): regime-weighted strategy blending

Promote one at a time. Hold shadow for 3-5 sessions minimum. See Brain/Runbooks/Canary Rollout.md.

Key Files

• schwab_skill/signal_scanner.py — main pipeline, scan_for_signals_detailed()
• schwab_skill/stage_analysis.py — is_stage_2(), check_vcp_volume(), add_indicators()
• schwab_skill/sector_strength.py — get_sector_heatmap()
• schwab_skill/advisory_model.py — calibrated probability scoring
• schwab_skill/config.py — all env var getters with defaults