Stockbee Momentum Burst Screener
Screen US equities for Stockbee-style short-term Momentum Burst candidates. The skill is a candidate-generation and setup-quality workflow, not a signal service or an auto-execution system.
When to Use
- User asks for Stockbee / Pradeep Bonde style Momentum Burst screening
- User wants 4% breakout, dollar breakout, or range expansion candidates
- User asks for short-term 3-5 day swing momentum setups
- User wants to review whether a daily breakout has A/B/C setup quality
- User provides a symbol list, universe file, or historical OHLCV JSON for screening
- User wants candidate outputs to feed into
technical-analyst,position-sizer, ortrader-memory-core
Prerequisites
- FMP API key for live universe and historical OHLCV screening:
export FMP_API_KEY=your_api_key_here - Install
requirements.txt, including the pinned exchange-calendar dependency. - Optional no-API path: provide
--prices-jsoncontaining daily OHLCV bars by symbol; historical fixtures need an explicit completed--as-of YYYY-MM-DD. - Run only after the market-regime workflow allows new swing risk, or mark output as manual-review-only.
Workflow
Step 1: Choose Input Mode
Use one of three modes:
Mode A: FMP universe scan
python3 skills/stockbee-momentum-burst-screener/scripts/screen_momentum_burst.py \
--fmp-universe \
--max-symbols 300 \
--output-dir reports/Mode B: Explicit symbols
python3 skills/stockbee-momentum-burst-screener/scripts/screen_momentum_burst.py \
--symbols NVDA SMCI PLTR TSLA \
--output-dir reports/Mode C: Offline OHLCV JSON
python3 skills/stockbee-momentum-burst-screener/scripts/screen_momentum_burst.py \
--prices-json data/daily_ohlcv.json \
--as-of 2026-11-06 \
--output-dir reports/Completed-session freshness gate
Every input mode must pass the same freshness gate before scoring. The default cutoff is the most recently completed NYSE session, frozen before collection using the New York exchange calendar. Before the close, today's unfinished bar is excluded; weekends, holidays, DST and early closes follow the shared calendar.
--as-of YYYY-MM-DDselects the last session on/before that date. It must not exceed the actual latest completed session, even if today was specified before the close. Historical dates only clip the available input window: live FMP requests do not guarantee that older data will be available or point-in-time.--max-bar-age-sessions Npermits at most N subsequent exchange sessions (nonnegative integer; default 0). A wider tolerance deliberately admits older observations; it does not make them current. Keep the default for Day 1 screening.- Bars after the cutoff are excluded. Missing/noncanonical dates anywhere in
supplied rows produce
REJECTED/invalid_price_date. In--prices-json, each row'sdate(ortimestamp) must be exactlyYYYY-MM-DD; values such as2026-11-06 00:00:00or2026-11-06T00:00:00Zare rejected. A latest eligible bar on a non-session day is also rejected. No eligible bar producesno_completed_price_data. Too-old bars producestale_price_data, score 0 and a do-not-trade downstream action. Short usable history stays rejected. - With
--max-bar-age-sessionsabove 0, a setup on an older (but tolerated) bar is capped atMANUAL_REVIEWwith astale_bar_toleratedsoft tag, neverACTIONABLE_DAY1. The cap changes onlystate;ratingandsetup_scorekeep their computed values, and the tag is added only when the cap applies. Treatbar_age_sessions > 0as the authoritative stale-bar signal, and do not route onratingalone. When every processed symbol is stale (typically FMP has not yet published today's EOD data), the report carries a top-levelwarningsentry and a## Warningssection in the markdown. - Calendar failure exits 1 before reports are written; invalid flag syntax exits
- A future/uncompleted as-of exits 1 before any provider request. Neither path replaces existing reports. The FMP universe additionally requests active equities and discards rows explicitly marked inactive; absent flags never substitute for the bar freshness gate.
JSON candidates include latest_bar_date (null if unplaceable/no eligible bar),
expected_session, bar_age_sessions (null when unknown) and
max_bar_age_sessions. Report metadata includes the expected session, allowed age,
latest_bar_dates by symbol and stale_price_data_count. The Markdown summary
shows cutoff, tolerance and stale count even without --include-rejected.
Dates and freshness do not establish provider identity, completeness, adjustment
or historical availability, and this screen never authorizes an order.
Step 2: Run the Screening Pass
The script detects these trigger families:
- 4% Breakout:
close / previous_close >= 1.04, volume above previous day, and volume above the liquidity floor - Dollar Breakout:
close - open >= 0.90, volume above the liquidity floor - Range Expansion: current daily range exceeds the prior three daily ranges while the prior day was not already extended
It then scores setup quality using:
- Trigger strength
- Volume expansion
- Prior base / range contraction quality
- Close location near the high of day
- Risk distance to the trigger-day low
- Failure filters such as prior 3-day run-up or recent 4% breakdown
- Market gate alignment
Step 3: Review Output
Read the generated JSON and Markdown reports. For each candidate, present:
- Trigger type and all matched trigger tags
- Day gain, dollar gain, volume ratio, and close-location percentage
- Prior base length and base width
- Entry reference, stop reference, and risk percentage to stop
- Setup score, rating, state, and reject reasons
- Suggested downstream action
Step 4: Send Survivors to Trade Planning
Use the output conservatively:
- A / A- candidates: send to
technical-analystfor manual chart validation, thenposition-sizer - B candidates: watchlist or smaller-risk review only
- Watch-only candidates: keep in model book; do not plan a trade unless chart review upgrades the setup
- Rejected candidates: retain for post-analysis, not for execution
Output
stockbee_momentum_burst_YYYY-MM-DD_HHMMSS.json- Structured candidate list, metadata, thresholds, score components, and rejectsstockbee_momentum_burst_YYYY-MM-DD_HHMMSS.md- Human-readable report grouped by rating/state
Resources
references/momentum_burst_methodology.md- Stockbee-style method summary and implementation boundariesreferences/scoring_system.md- Component weights, state thresholds, and failure filtersreferences/entry_exit_rules.md- Entry reference, stop, sizing handoff, and exit template
Shared FMP client
Live FMP access uses the bundled generated scripts/fmp_client.py, loaded only
when an FMP run starts, so offline modes still work without requests. CLI
key/environment precedence, the API-call budget and successful output shapes are
preserved. Requests use stable endpoints only; a denied, empty or failed stable
request does not fall back to retired v3 URLs. The shared client masks raw and
URL-encoded key echoes, retries malformed JSON once, and retries a 429 at most once,
after a 60-second wait, before stopping on a repeated rate limit.