ChokePoint Algorithm v1.2

§1. Composite Sentinel Score and Normalization

The Sentinel Score S is a weighted linear combination of normalized signal components, scaled to the interval [0, 10] for downstream surcharge mapping. We introduce a normalization multiplier to align historical volatility and preserve scale consistency across ingestion windows.

Definition (raw composite):

S_raw = Σᵢ (wᵢ · xᵢ)

where xᵢ ∈ [0, 10], Σᵢ wᵢ = 1, and wᵢ ≥ 0.

Normalized composite (ChokePoint v1.2):

S = 1.225 × S_raw

The factor 1.225 is a fixed calibration constant chosen so that:
  • S remains in [0, 10] for typical input ranges;
  • Back-tested surcharge outcomes match institutional acceptance bands;
  • No re-scaling is required when adding or removing signal sources (weights are adjusted instead).

In production, the composite is clamped to [0, 10] after application of the 1.225 multiplier. All downstream formulae use this clamped S.

§2. GDELT: News Volume vs. Sentiment De-biasing

The Kinetic Disruption Proxy (KDP) is derived from GDELT 2.0 document-level aggregate tone and event volume. Raw GDELT tone can be biased by (a) volume spikes in non-representative outlets and (b) sentiment drift when a few high-volume sources dominate. We apply an explicit de-biasing step before mapping to the threat signal.

Let:

• V = normalized news volume (article count over window, scaled to [0, 1]);
• T = aggregate tone from GDELT (typical range ≈ −5 to +5);
• V̄ = rolling mean of V over the calibration window.

De-biased tone (News Volume vs. Sentiment):

  T_db = T × (1 − α · max(0, V − V̄))

  α ∈ [0.2, 0.4]  (configurable; default 0.3).

Interpretation: When volume V exceeds its recent mean V̄, we down-weight the raw tone T so that short-lived volume spikes do not disproportionately drive the threat score. The KDP threat component is then:

  x_threat = clamp(5 − 0.5 · T_db, 0, 10).

Thus, more negative (adverse) tone decreases the 5-centered value and raises the threat contribution; volume de-biasing prevents single-day news spikes from overriding sustained sentiment.

§3. PCM v1.2 Correlation Coefficients

The Probabilistic Congestion Model (PCM) v1.2 supplies a congestion/flow component correlated with commodity stress and, where available, AIS-derived density. The following correlation coefficients are fixed in v1.2 for reproducibility.

PCM v1.2 coefficients:

  ρ_Brent = 1.2   (Brent 1d return → congestion proxy scaling)
  ρ_vol   = 0.6   (Brent volatility → congestion uncertainty)
  ρ_AIS   = 1.0   (AIS density anomaly, when available; 1.0 = no scaling)

Congestion signal (before normalization to [0,10]):

  C_raw = ρ_Brent · ΔBrent_1d_norm + ρ_vol · σ_norm + ρ_AIS · AIS_anomaly

  x_congestion = clamp(C_raw, 0, 10).

The 1.2 coefficient on Brent 1d return is calibrated so that a 1% move in Brent maps to a predictable step in the congestion component without overshooting the [0, 10] band under normal market conditions.

§4. Deterministic Surcharge Formula

The recommended surcharge percentage is a deterministic function of the (normalized and clamped) Sentinel Score S. No free parameters are exposed at quote time.

surcharge% = round((S × 0.12 + 0.8) × 10) / 10

surcharge_USD = base_quote_USD × (surcharge% / 100)

total_quote_USD = base_quote_USD + surcharge_USD

Constants 0.12 and 0.8 are locked in ChokePoint v1.2. All evidence packages and audit logs reference this formula for full traceability from raw signals to final quote.