Equity Scanning And Watchlists

Equity Scanning and Watchlists

1. Overview + routing

This skill is the workflow ENTRY for equity day trading (screening stages 1 and 2 merged): it builds the tradeable universe and the ranked watchlist. day-trading-patterns CONSUMES the watchlist to work a chosen symbol; this skill selects WHICH symbols.

Routing: chart-pattern recognition on a chosen symbol → day-trading-patterns; social/news sentiment inputs → reddit-signal-mining / gdelt-event-mining / financial-sentiment-analysis; historical validation → trading-strategy-backtester; order handling on a halted/resuming name → equity-broker-execution §6.

2. Universe definition + data sources

What is IN the universe: NMS common stocks (vs OTC), with explicit handling of ETF/ADR/warrant/right, recent-IPO and symbol-change cases; use stable instrument identifiers; and run a broker tradability/shortability check before a name becomes actionable.

Data sources (verify tiers/cadence at use — see research/equity-day-trading-skills/scanning-journaling-research.md):

Source	Fields	Cadence
Polygon.io	quotes/trades, reference tickers, shares outstanding, splits/dividends	real-time (paid) / delayed
Financial Modeling Prep	float, shares outstanding, earnings calendar	daily-ish
Finnhub	quotes, basic financials, earnings	mixed
Nasdaq Data Link	curated short-interest datasets	dataset-specific
Broker scanners (IBKR / Schwab)	movers + tradability/shortability	broker-authoritative

Record, per field, WHICH source it came from and its update cadence.

3. Gapper scan procedure

Pre-market and intraday variants over a threshold template: gap %, price band, minimum dollar-volume, float band, minimum-history requirement, and exchange-calendar / early-close awareness.

CORPORATE-ACTION-ADJUSTED baselines are mandatory. A 10-for-1 split must not read as a −90% gap, nor a reverse split as a +900% gap. Adjust the prior-close baseline by corporate-action factors before computing the gap.

def gap_pct(prev_close: float, premarket_price: float, split_factor: float = 1.0):
    """Corporate-action-adjusted gap. split_factor adjusts the baseline."""
    adjusted_prev = prev_close / split_factor
    if adjusted_prev <= 0:
        raise ValueError("invalid adjusted baseline")
    return (premarket_price - adjusted_prev) / adjusted_prev * 100.0

4. Catalyst calendar ingestion

Ingest earnings, FDA/PDUFA-class binary events, and macro prints; each modifies watchlist rank. Sources: earnings calendars (FMP/Finnhub/Nasdaq/Zacks), biotech (BioPharmaCatalyst / FDA), macro (BLS/BEA/Fed / Trading Economics / Econoday).

5. Float / short-interest / days-to-cover

Lookup with explicit field lineage:

PROHIBITION (binding). FINRA short interest is a twice-monthly POSITION snapshot, published on a lag (settlement mid-month and month-end; take the publication date from FINRA's annual schedule — 2026 dates are generally about seven business days after settlement, not a fixed invariant). Daily short-sale VOLUME (FINRA daily files / exchange Reg SHO daily files) is an ENTIRELY DIFFERENT FACT and must NEVER be substituted for short interest. High daily short volume does NOT mean high short interest.

• Days-to-cover = short interest ÷ average daily volume; define the ADV denominator window explicitly (e.g. 20-day).
• Expose the settlement date AND publication date of any short-interest figure in the output so staleness is visible.

6. Relative volume (RVOL)

Standard form: cumulative volume to the current time-of-day ÷ the average cumulative volume at the same time of day over a lookback (e.g. 20 sessions). Time-of-day normalization is essential — 2× at 09:45 differs from 2× at 15:45. Guard new listings / thin history with a minimum-history check.

7. LULD / halt status tracking

Track pre-open halt lists and resumption status. LULD bands and pauses apply during regular hours only (WATCH as of 2026-07-14 — an SEC overnight price-band amendment, SR filing 34-105596, is pending; re-verify before overnight use). For ORDER handling on a halted/resuming name, hand off to equity-broker-execution §6 — this skill tracks status, it does not place orders.

8. Ranked watchlist construction

Combine gap, RVOL, float, and catalyst weight into a scoring template; enforce a max-size discipline (a bounded watchlist). The output artifact carries REQUIRED per-field metadata:

watchlist_item:
  symbol: str
  score: float
  fields:
    gap_pct:        {value, source, as_of, latency_class, adjustment_state}
    rvol:           {value, source, as_of, latency_class}
    float_shares:   {value, source, as_of}
    short_interest: {value, source, settlement_date, publication_date}

Every field carries source, as-of timestamp, latency class, and adjustment state — a number with no provenance is not actionable.

9. Unattended-mode caveat (TW-4) — first-class

Attended mode is the default framing — a human bridges gaps in seconds, so attended failure is SAFE (worst case: no trades).

UNATTENDED mode FAILS CLOSED on more than emptiness. Halt the pipeline and ALERT — never trade off — when the watchlist is:

• empty (no candidates),
• stale (as-of older than the freshness threshold),
• degraded (missing required fields or a required source is down),
• divergent (sources disagree beyond tolerance), or
• anomalously large (output far bigger than the historical norm).

A stale watchlist that LOOKS healthy is more dangerous than a silent no-op — it invites trading on yesterday's world. Fail closed and alert; do not proceed on a non-empty-but-unverified list.

Anti-Patterns

Anti-Pattern	Why It Fails	Correct Approach
Scanning without RVOL normalization	Raw volume favors large caps and misreads time-of-day	Use time-of-day-normalized RVOL over a lookback window
Passing daily short-sale volume off as short interest	They are different facts; the derived signal is wrong	Use the twice-monthly SI snapshot for SI; never substitute daily short volume
Unadjusted gap baselines across splits/dividends	A split reads as a huge phantom gap	Apply corporate-action-adjusted baselines before computing gaps
Treating stale short interest as current	SI publishes on a lag; the position may have unwound	Expose settlement + publication dates; judge staleness explicitly
Unbounded watchlists	Too many names dilute focus and rank	Enforce a max-size discipline in construction
Sentiment-only universes	Social buzz without price/float/volume structure is noise	Build the universe on price/volume/float; treat sentiment as an input, not the universe
Unattended trust in a non-empty list	Stale/degraded/divergent lists look healthy but are wrong	Fail closed on empty/stale/degraded/divergent/anomalous; alert, never trade off it