How the Flow Score works

The five components

The score is the sum of five weighted sub-scores, capped at 100. Each component is independent — a strong reading in one area can't compensate for missing data in another, but a single mediocre area won't kill a strong alert.

Thresholds

The score band determines color, label, and how to interpret the signal:

Inside the Flow sub-score (0–25)

The Flow component is the heaviest weighted. It rewards four institutional-activity signatures:

• OI accumulation (0–8 pts): +8 when call OI rises by >10,000 contracts or >10% of total; +5 at >5,000 / >5%; +3 at >1,000 / >2%
• Unusual contracts (0–6 pts): +6 when 3+ contracts trade abnormal volume; +3 for 1–2
• Block trades (0–5 pts): +5 when 2+ block trades print; +3 for a single block
• Call/put ratio (0–6 pts): +6 when ratio >3; +4 at >2; +2 at >1.5. (Bullish-only — bearish flow is gated separately per M2M long-bias policy)

Source of truth: src/lib/flowAnalyzer.ts · calculateSqueezeScore(). If the code and this page disagree, the code wins.

Conviction Score vs Flow Score

You may see two scores referenced internally — a Flow Score (raw detector output, described above) and a Conviction Score (calibrated probability after post-processing). The user-facing badge on each alert card is the Conviction Score.

The calibration step adjusts for known false-positive risk factors (earnings proximity, FOMC week, opex), repeat-fire history, and regime fit. So a 70 Flow Score during a calendar-risky window may surface as a 55 Conviction Score — the alert is still detected, but the calibration is honest about the elevated noise.

Cross-source confluence (Phase 2 + 3)

Beyond the five Flow Score components, every alert is also checked against six external data sources for confluence signals. When a flow alert + insider buying + institutional accumulation + analyst upgrade + short-interest squeeze fuel + retail-sentiment data all align, the alert carries a composite-uplift badge (✦ Nx confluence).

• Insider-confirmed (Form 4 filings): senior-officer purchase ≥$100K in last 30 days. Weight 30–90 by dollar size.
• Institutional accumulation (13F filings): 3+ funds opened new positions last quarter, net QoQ accumulating. Weight 40–80 by breadth.
• Analyst catalyst (analyst ratings): upgrade in last 7 days, net upgrade-vs-downgrade positive. Weight 25–75 by recency.
• Squeeze precondition (short-interest data): 3+ of { SI≥15%, CTB≥20%, DTC≥5d, Util≥90% } simultaneously hot. Weight 70 (3/4) or 95 (4/4).
• Sentiment fade (social-sentiment data): CONTRARIAN — mention velocity z≥2 + bullish ≥65%. Signals retail FOMO; institutions historically pre-position before the pile-in. Weight 50–85 by intensity.
• Sentiment squeeze fuel (social-sentiment + short-interest data): CONFIRMATION — bullish ≥60% + mention spike + SI ≥15%. Retail discovery driving structural squeeze fuel. Weight 40–80.

Composite uplift formula: max(w1) + 0.4·w2 + 0.15·w3 (diminishing returns past 3 sources). Source: src/lib/flow/crossSourceConfluence.ts.

Limitations

• The score is a hypothesis, not a recommendation. A 75 score can fail; a 50 score can win. Use it to prioritize, not to size.
• Track-record figures (visible in the Flow Scanner Track Record tab and per-card edge pill) are the truth of how each alert type has performed. Cross-reference the score against the track record before committing.
• Sample size matters. Win-rate figures with n<20 are marked "preliminary" — treat them as directional, not validated edge.
• Bearish flow is informational-only (per M2M long-bias policy). The score still computes for awareness, but no trade plan is published.
• Backtest contamination caveat: the scoring weights have been tuned against historical outcomes. Walk-forward / out-of-sample validation is queued as part of the Track Record time-series feature.