A central bank's shadow-run calibration

This is an anonymized design-partner scenario; the institution is a central bank, the figures are illustrative, and no actor has agreed to anything. We are pre-revenue, building the POC. With that label on it, here is what a monetary-policy shadow run looks like when the substrate is a versioned graph and the calibration loop is the deliverable, not the cone.

The path the bank already lived

A monetary-policy path cone is a distribution over where the policy rate may go over the next quarters. The bank issued one; the quarters then happened; the realized path is now ground truth. The shadow run is the cone rerun against the NodeVersions that actually obtained, scored against the realized path via Brier and log-loss. The point of the shadow run is not to celebrate or mourn the cone — it is to measure whether the cone is getting more honest over time, which is the only question a central bank can answer in front of a legislature without overclaiming.

Why the loop is the deliverable

A central bank that says "our forecast was right" has made a claim it cannot defend once the path diverges. A central bank that says "we score every cone against the realized path, here is the Brier trend, here is the disclosed baseline we beat" has made a claim a legislature can read. The loop is the deliverable because the loop is what makes the next cone more honest. Gate 1b requires the shadow run to beat a disclosed baseline on one forecastable series before any forecast side is positioned as accurate; that gate has not passed here. The loop is in motion; the result is not.

A central bank does not need a forecast it trusts. It needs a forecast it can score against a path it already lived — and the scoring is the deliverable, not the cone.

The versioned substrate the loop reads

A shadow run is only meaningful if the cone and the realized path are scored against the same substrate. The GNL pins every NodeVersion — the L2 Government node for the policy decision, the L1 Instruments the market priced, the macro data the model read. The shadow run replays against those versions; the realized path is scored against the same ones. The gap is a calibration gap or a substrate gap, and the versioned graph is what lets the bank tell them apart — which is the difference between a loop that closes and a loop that hides.

• The shadow run is the cone scored against the realized path — a number, not a narrative.
• The loop is the deliverable: the Brier trend is what a legislature can read, not a single cone.
• Gate 1b is the gate: a shadow run must beat a disclosed baseline before any accuracy claim. Not passed here.
• The versioned substrate is what makes the gap a calibration gap, not a substrate gap.

What the central bank does not get

The central bank does not get a cone that is correct. It gets a cone that is scored, a loop that is honest about its own gap, and a manifest that a legislature can replay. Certified is not true; the manifest attests provenance and reproducibility, not that the policy path was forecast correctly. The honest version of "is our forecast accurate" is "we measure it against a disclosed baseline, and here is the loop that produced the number" — and that is the version a central bank can defend.

A central bank that says "our forecast was right" cannot defend it. A central bank that says "we score every cone, here is the Brier trend, here is the baseline" can. The loop is the deliverable; the cone is one iteration of it.

— design-partner calibration note