AI isn't a feature of
this model. It's a capital type.

The standard experiment raises the nominal policy rate 100 basis points above baseline in year 2, holds it in year 3, and returns it to path in year 4, run under both live policy modes against the same baseline. All figures are deviations from baseline.

The economics reads directly. A credibly held 100 basis-point tightening buys roughly 0.21 percentage points of disinflation per year at its peak, at the cost of a contraction troughing at −1.89 percent of GDP in year 3. A Fed that follows its usual reaction function gives up 0.08 percentage points of that peak disinflation and buys 0.67 percentage points of trough depth in return, cutting about 69 basis points below the projected path at the reversal.

Read the timing, because it is the most useful thing the table says. The trough arrives one year after the rate rise lands, while the policy is still in place. About 30 percent of it is unwound within two years of the reversal, and there is no later re-deepening in the remaining twenty-three years. That is the timing the empirical literature on monetary policy reports, and the depth sits inside the range that same literature reports once the experiments are put on a comparable footing.

What does not come back is the capital stock. Firms work their existing plant less hard for about three years and are back to normal intensity by then. But the buildings and equipment they did not buy during the tightening are simply not there. Capital stays below its reference path in every one of the twenty-six years. GDP therefore approaches its reference path from below and never crosses it. The years of lost output are not recovered later. Nothing in the model repays them, and nothing was designed to.

The one cost that keeps growing is the government's interest bill, which compounds with the debt for as long as the run continues.

Policy questions run under one of three modes, all of which reproduce the identical baseline exactly.

1. The Fed holds its projected path. The nominal short rate follows its baseline path plus whatever deviation the scenario scripts. It is the classic scenario instrument, isolating a specified rate move without the central bank reacting to its own consequences.
2. The Fed responds as it usually does. A Taylor-type reaction function on deviations: 1.5 × the inflation deviation + 0.5 × the employment gap, with scripted policy shocks entering as departures from the rule. A control run reproduces the projected rate path endogenously to within a hundredth of a basis point.
3. The Fed responds to output. The same reaction function on the GDP gap, built and verified solvable and control-neutral, with full grading scheduled alongside the potential-output research program.

No mode requires a money market, an auxiliary expectations block, or a re-specification of the baseline.

The fiscal channel of monetary policy is live in the model rather than asserted. Each year, 0.32 of the outstanding un-repriced gap between the market government-bond rate and the government's average liability rate passes into the interest bill: the bill share plus the coupon roll at average maturity. Certification verifies the measured passthrough reproduces 0.320 × the market gap to three digits.

The channel is quantitatively material and runs in both directions. Under a 100 basis-point tightening, government interest payments rise by roughly $140 billion in the first year of the tightening and $261 billion at its peak in year 3, and stay elevated while the rate gap persists. It is a deficit-financed transfer to bondholders that acts as an automatic stabilizer and shallows the real contraction. Under a rule-based easing the same channel reprices interest cost downward and moderates the recovery. The net effect is to pull the two policy modes' long-run outcomes toward each other.

Money is reported, not imposed. The central bank sets the nominal short rate; deposits are created by the financial sector to meet demand; no monetary aggregate plays a causal role, matching the actual Federal Reserve operating regime. Money growth, real balances and velocity are demand-determined readouts.

Behavioral constants are set once and retained unless a registered measurement supersedes them. Elasticities, tax rates, bases, shares and maturity structures are measured directly from the database and official sources.

The carbon-tax base carried in the 2024 database is 4,982 million tonnes of carbon-dioxide equivalent, 92.5 percent of the EPA's published gross national total for the comparable year. A $50 per tonne price on that base raises about $249 billion, roughly 0.86 percent of GDP.

On the current model it cuts combustion emissions 11.3 percent on impact and 11.7 percent after full adjustment. The split is the informative part. Stationary emissions fall 16.0 percent on impact and 16.9 percent in the long run; transport falls only 2.8 and 2.1 percent. Stationary sources bear the cut because they can switch fuels. Transport, at this price, largely cannot.

Biogenic carbon dioxide and the criteria pollutants are held as a memo series outside both the headline and the tax base, as greenhouse-gas accounts require. The energy satellite behind these figures reconciles to the Energy Information Administration's published tables fuel by fuel; its final-energy total sits 1.9 percent above the EIA's end-use figure. The difference is electricity-system losses, which this satellite books as generation fuel.

The framework's identified limits as of this documentation vintage.

• Slow reconvergence. After an impulse decays, the GDP level gap closes at a measured rate rather than snapping shut, leaving a multi-year residual. Run to 2050, the residual deepens into the mid-2030s, then reverts, and is still narrowing at the end of the run, measured rather than extrapolated, and in the current version it closes. Measurement attributes what remains to the convergence structure of the framework rather than to any single channel. A potential-output specification is the next stage of the research program.
• The investment caveat. In the baseline, aggregate investment is data. It comes from the published official path, and the model's margins absorb the difference. In a scenario the model solves for it. The composition answer, which industries invest, is the stronger of the two readings because it rests on the industry structure; the level answer rests on a single calibrated response size and should be read alongside it.
• Export margins are baseline data outside scenario mode.
• Government asset-side rate passthrough awaits extension.
• Industry-level variation in the market-to-book ratio awaits external data.

Documentation vintage August 2026. Figures cited are from the certified experiment record. GSM8US and the Global Systems Model suite are proprietary to Phylleos, Inc.