Methodology

Treasury clock

Federal deficit and total outlays are predicted and graded separately from G.19. The selected benchmark for each series minimizes point-in-time MAE among last value, seasonal naive, and AR(2). Empirical 80% and 95% bands use only prior out-of-sample errors, and live calibration is never blended across series. Historical MTS months without an exact official release timestamp remain omitted.

The 2018–2026 first-print backtest selected seasonal naive for deficit (MAE $130,896.7M; 83.3% / 100.0% interval coverage; 76.2% acceleration-sign accuracy) and AR(2) for outlays (MAE $90,750.6M; 86.2% / 100.0% coverage; 62.3% sign accuracy). Outlays improved only 12.5% over random walk and missed the inherited improvement, 80%-coverage, and sign-accuracy bars; it remains published unchanged rather than being replaced by an unproved model.

Credit flow

The view uses the registered current-period midpoint flow, Matrix A midpoint, and Matrix B midpoint for each lane. It groups only after those published mappings are applied, keeps every remainder group within one evidence tier, and does not enter the nowcast or change any company estimate.

Evidence Lift

Evidence Lift equals a company's estimated DFR% midpoint divided by its pure Tier 3 fungibility baseline for the same period. The baseline is a counterfactual built from the already-computed broad proportional lanes—general retail plus the two fungible categories—with company-specific financing and auto-category evidence removed.

A value of 1.00× is baseline-only: no company-specific financing evidence was found, so the estimate reflects proportional allocation. A higher multiple shows how much registered evidence lifts the midpoint above that baseline. It does not measure risk, credit quality, or investment merit.

Backtest coverage

In the point-in-time backtest, the nominal 80% band contained 71.3% of first prints and the nominal 95% band contained 93.1%. Both currently under-cover their nominal rates, especially the 80% interval, so these bands are empirical uncertainty ranges, not guarantees.

Assumptions

Low / midpoint / high values are triangular priors unless all three values are identical. Source files are public so a cold clone can reproduce every result.

Criticality is computed, not assigned: an assumption is critical when it supplies at least 5% of either the midpoint attributed numerator or covered-company denominator. This build has 11 critical assumptions and 0 critical assumptions without an independent fallback. A permitted but unavailable primary automatically selects its first permitted independent fallback and widens the registered band; if no fallback remains, the build emits BLOCKED rather than inventing a value.

A real fallback activation is a versioned source_fallback changelog event. It appears in the JSON and RSS event feeds with the reviewed activation date; the publisher does not manufacture a timestamp from a transient network failure.

Source resilience

The complete machine-readable strategy is versioned in the source registry. This compact table shows only critical assumptions; noncritical rows remain in the full registry and criticality report.

Matrix A and B

Matrix A has 16 product-to-category rows. Matrix B has 166 category-to-company rows. Tests reject negative weights, missing evidence, unregistered uncertainty, or row allocations over 100%.