Skip to main content

​
How Whether Works

Whether operates on a deterministic, explainable loop that converts public Treasury data into operational constraints. There are no opaque ML models—every score and classification follows testable rules.

​
The Core Loop

1

Fetch Treasury Data

Server fetches US Treasury yield curve data from the US Treasury Fiscal Data API (/v2/accounting/od/daily_treasury_yield_curve).Required fields:
  • 1-month yield (with 3-month fallback)
  • 2-year yield
  • 10-year yield
Caching: 24-hour revalidate (Treasury updates end-of-day)
2

Normalize to TreasuryData

All outputs include explicit metadata:
  • source - API endpoint URL
  • record_date - Date of Treasury data
  • fetched_at - Timestamp of retrieval
If the API fails, Whether serves snapshot data with OFFLINE labeling.
3

Run Regime Engine

The Regime Engine scores two dimensions:
  • Tightness (0-100): How expensive is capital?
  • Risk Appetite (0-100): How willing are investors to take risk?
These scores are then classified into one of four regimes.
4

Render Regime + Playbook

The classified regime produces:
  • Regime label and description
  • Operating constraints (pace, hiring, roadmap bias)
  • Decision Shield verdicts
  • Export-ready briefs with full source metadata

​
How the Regime Engine Scores Tightness

Tightness measures how expensive capital is right now. It blends two components:

​
1. Base Rate Points

The base rate (1-month or 3-month Treasury yield) represents the cost of short-term money. 
// From regimeEngine.ts:108-109
export const BASE_RATE_TIGHTNESS_THRESHOLD = 5;
export const TIGHTNESS_BASE_RATE_POINTS = 90;

// From regimeEngine.ts:300-304
const baseRatePoints = clamp(
  Math.round((baseRate - baseRateThreshold) * BASE_RATE_TIGHTNESS_MULTIPLIER),
  SCORE_MIN,
  TIGHTNESS_BASE_RATE_POINTS
);
Formula:
  • Base rate points ramp from 0 to 90 as rates move above 5%
  • Multiplier: 180 (each 1% increase above 5% adds 180 raw points, capped at 90)

​
2. Inversion Points

When the yield curve inverts (10Y < 2Y), it signals stress and caution. 
// From regimeEngine.ts:110
export const TIGHTNESS_INVERSION_POINTS = 25;

// From regimeEngine.ts:306-312
const inversionPoints =
  curveSlope < SCORE_MIN
    ? clamp(
        Math.round(Math.abs(curveSlope) * CURVE_INVERSION_TIGHTNESS_MULTIPLIER),
        SCORE_MIN,
        TIGHTNESS_INVERSION_POINTS
      )
    : SCORE_MIN;
Formula:
  • Inversion points ramp from 0 to 25 as slope moves below 0%
  • Multiplier: 50 (each 1% inversion adds 50 raw points, capped at 25)

​
Combined Tightness Score

// From regimeEngine.ts:313-315
const score = baseRatePoints + inversionPoints;
return clamp(score, SCORE_MIN, TIGHTNESS_CAP);
Total cap: 100
Example: If base rate is 5.5% and curve slope is -0.3%:
  • Base rate points: (5.5 - 5.0) × 180 = 90 (capped at 90)
  • Inversion points: 0.3 × 50 = 15
  • Tightness = 90 + 15 = 100 (capped at 100)

​
How the Regime Engine Scores Risk Appetite

Risk appetite measures investor confidence based on the 10Y-2Y yield curve slope. 
// From regimeEngine.ts:112-113
export const RISK_APPETITE_MIN_SLOPE = -1.0;
export const RISK_APPETITE_MAX_SLOPE = 1.5;

// From regimeEngine.ts:318-322
export const computeRiskAppetiteScore = (curveSlope: number) => {
  const normalized =
    (curveSlope - RISK_APPETITE_MIN_SLOPE) / (RISK_APPETITE_MAX_SLOPE - RISK_APPETITE_MIN_SLOPE);
  return clamp(Math.round(normalized * SCORE_MAX), SCORE_MIN, SCORE_MAX);
};
Formula:
  • Maps slope from -1.0% (very cautious) to +1.5% (confident)
  • Scaled linearly to 0-100
Example: If curve slope is +0.5%:
  • Normalized: (0.5 - (-1.0)) / (1.5 - (-1.0)) = 1.5 / 2.5 = 0.6
  • Risk Appetite = 0.6 × 100 = 60

​
How Regimes Are Classified

Once tightness and risk appetite are scored, Whether classifies the regime using fixed thresholds. 
// From regimeEngine.ts:114-115
export const TIGHTNESS_REGIME_THRESHOLD = 70;
export const RISK_APPETITE_REGIME_THRESHOLD = 50;

// From regimeEngine.ts:324-339
export const classifyRegime = (
  tightness: number,
  riskAppetite: number,
  thresholds: RegimeThresholds
): RegimeKey => {
  if (tightness > thresholds.tightnessRegime && riskAppetite <= thresholds.riskAppetiteRegime) {
    return "SCARCITY";
  }
  if (tightness > thresholds.tightnessRegime && riskAppetite > thresholds.riskAppetiteRegime) {
    return "DEFENSIVE";
  }
  if (tightness <= thresholds.tightnessRegime && riskAppetite <= thresholds.riskAppetiteRegime) {
    return "VOLATILE";
  }
  return "EXPANSION";
};

​
Classification Matrix

TightnessRisk AppetiteRegime
> 70≤ 50SCARCITY
> 70> 50DEFENSIVE
≤ 70≤ 50VOLATILE
≤ 70> 50EXPANSION
Operators can override these thresholds via URL query params. Overrides are tracked in an on-screen audit trail.

​
How This Translates to Operational Constraints

Each regime maps to concrete operating guidance.
Regime: Tightness > 70, Risk Appetite ≤ 50Description: Capital is expensive and risk appetite is low. Prioritize survival over growth.Constraints:
  • Shorten payback windows and preserve cash
  • Delay speculative hiring or large platform rewrites
  • Route roadmap bets through revenue certainty
Decision Shield Bias: Most initiatives marked Dangerous or Risky

​
Data Provenance

Every output includes explicit source metadata: 
// From regimeEngine.ts:229-249
return [
  {
    id: "base-rate",
    label: baseRateLabel,
    value: baseRate.used === "MISSING" ? null : baseRate.value,
    unit: "%",
    sourceLabel: "US Treasury Fiscal Data API",
    sourceUrl: treasury.source,
    recordDate: treasury.record_date,
    fetchedAt: treasury.fetched_at,
    notes: baseRateNotes,
  },
  // ... more inputs
];
If Treasury data is missing, Whether defaults to conservative assumptions (base rate = threshold, curve slope = cautious floor) and adds data warnings to the output.

​
Offline Mode

If the Treasury API fails:
  1. Whether serves snapshot data from data/ cache
  2. UI displays OFFLINE / SIMULATED badge
  3. All outputs tagged with fallback_reason and fallback_at metadata
Operators should pause irreversible decisions until the live feed returns or they confirm the cached data is acceptable.

​
Next Steps

Quickstart

Get started with your first regime reading

Decision Shield

Learn how to use verdict analysis for planning decisions

Build docs developers (and LLMs) love

Get started for freeTalk to us