NIV.

Regenerationism — a new economic philosophy.

Regenerationism is the economic school this chapter proposes. Its foundational claim is that the leading indicator of macroeconomic regime health is the velocity of regenerative capital formation — capital flowing into areas of the economy with exponentially compounding margins of value generation (non-residential investment, residential investment, R&D, productive-capacity-feeding inventory) — measured against the cumulative friction that impedes its circulation (yield inversion, positive real rates, rate volatility, idle capacity). Regenerative capital is the liberator; friction is the drag; the velocity of the first net of the second is what the economy actually does.

This is a direct challenge to the three dominant frames in macroeconomics. The Fed yield curve is a sentiment signal — it measures what bond traders expect the Fed to do next, and becomes reflexive during QE and forward guidance. The Conference Board LEI aggregates ten coincident-or-lagging components into a linear arithmetic index and cannot detect non-linear regime shifts. Dynamic Stochastic General Equilibrium models assume an equilibrium return that the credit plumbing of 2008 showed does not hold. None of these schools measure where the money actually goes: into regenerative capital, or away from it. Regenerationism measures exactly that, and NIV is what happens when the philosophy is written as a scalar and handed 504 months of walk-forward history.

The repository is named regenerationism for exactly this reason — the school precedes the signal. NIV is its first emitted instrument; the next instruments will follow the same template (thrust · regenerative-capital kernel · slack · drag, validated on public data with walk-forward discipline) across monetary, energy-grid, and supply-chain regimes. The methodology is transferable because the philosophy is.

The empirical receipt for the thesis comes from the ensemble itself. Under Gini-impurity feature importance, the regenerative-capital term efficiency_sq scores 0.9328 — the highest in the entire framework. The Fed yield spread scores 0.0298, which is 31.2× smaller. The model was not told to prefer regenerative capital. It was asked which feature best predicts macroeconomic regime shifts, and it picked the school unprompted.

What a macro primitive looks like.

The Fed yield-curve spread is the canonical macro stress signal. It is also a single number the market has known about for decades. The interesting question is not whether you can match it, but whether you can add orthogonal information — a scalar that captures stress the yield curve misses, without being a black box.

NIV is built from four primitives that are already, on their own, understood by macroeconomists. The work was in composing them around the regenerative-capital thesis in a way that remains economically interpretable, differentiable, and reproducible from public data only — then validating the composition across 42 years and six stress tests with strict walk-forward discipline.

The formula.

NIV_t  =  ( u_t · P_t² )  /  ( X_t + F_t )^η         η ∈ [1.0, 2.5]

  u = tanh( +1.0 · ΔG  +  1.0 · ΔA  −  0.7 · Δr )          # Thrust
  P = ( Investment · 1.15 ) / GDP                          # Efficiency
  X = 1 − ( TCU / 100 )                                    # Slack
  F = 0.4 · s  +  0.4 · max(0, r − π)  +  0.2 · σ          # Drag

Thrust bounds investment YoY growth, M2 money growth, and Fed Funds rate change through a tanh. Bounded [−1, +1]. Positive = expansion impulse; negative = contraction impulse. Efficiency is regenerative capital as a share of GDP, with a 1.15 R&D/education multiplier, squared in the numerator to reward productive allocation. Slack is capacity headroom (1 − TCU/100). Drag is a weighted penalty over yield inversion magnitude, positive real rate, and 12-month rate volatility. Elasticity η controls how steeply friction punishes thrust. All six weights and η are adjustable in the live simulator — sliders, not secrets. No proprietary data, no hidden parameters.

Public data — 8 FRED series.

Every NIV input comes from the Federal Reserve Economic Data (FRED) API. You reproduce the entire pipeline with a free API key and eight series:

Walk-forward out-of-sample.

Protocol: expanding-window walk-forward, retrained every 5 months, warm-up through 1983. No lookahead bias — only data up to t is used to predict horizons at t + h. The ensemble is L2 Logistic Regression + AdaBoost (15 depth-1 stumps, learning rate 0.1) + a small tanh neural network. Probability thresholds: yellow 12 – 35%, red > 35%.

Multi-horizon performance.

Most economic indicators degrade as the forecast horizon lengthens. NIV does the opposite — it sharpens, because the destruction of regenerative capital takes roughly eighteen months to metastasize into headline GDP. That structural gap is exactly what the signal measures.

From 6 months to 18 months: AUC discrimination improves by +14.70%, F1 by +58.09%, Brier calibration error falls by −23.19%. The optimal classification threshold converges empirically to 31 – 33% — which is exactly where the red-alert threshold (> 35%) was set.

Six contraction cycles, forensic breakdown.

Every major systematic contraction of the modern era is flagged by NIV with measurable lead, and the dominant trigger is recoverable from the component decomposition:

In 5 out of 6 modern structural crises, the regime shift was triggered by thrust — the liquidity and investment plumbing fracturing long before lagging headline metrics register a drop. The 2020 COVID crash is the sole exogenous anomaly: NIV correctly diagnoses it as not a structural failure of regenerative capital formation, because the pre-shock plumbing was stable.

Ensemble false-alarm suppression.

A macro model lives or dies by its false-positive rate. No single ML layer is good enough: the neural net, starved of compute, flatlines into the baseline; logistic regression over-fires during healthy acceleration; boosted stumps are hypersensitive. The L2-regularized ensemble enforces cross-verification between the three, and the > 35% red threshold is only crossed when all three agree.

Seven critical alerts in 42 years. The seven months that cleared the threshold — Jan 1984 (post-Volcker inflation), Oct – Dec 1994 (the Bond Massacre and the 75 bp Fed hike), Jan – Feb 1995 (soft-landing friction), and Nov 2007 (GFC core onset) — are all genuine macro regime shifts, not noise.

NIV vs the Fed yield spread.

The orthogonality question is the most important one: is NIV just reverse-engineering the bond market? A correlation audit says no.

NIV shares the macroeconomic baseline priced in by the yield curve — that’s the 0.76 correlation — but 41.71% of its variance is independent, and that variance is not noise. Under Gini-impurity feature importance the ensemble ranks the NIV channels as follows:

Capital efficiency is 31.2× more important to systematic-risk detection than the yield spread inside a multivariate model. NIV front-runs the bond market by tracking where the money goes — into regenerative capital, or not — rather than the sentiment guiding it.

GDP forecast benchmark.

A 20-point grid search over lag (0, 3, 6, 12 months) and smoothing window (3 – 18 months) produces an honest head-to-head:

NIV dominates at zero lag — it is the better real-time snapshot. The yield spread dominates at 12 months, but its accuracy across the 20-config grid is 2.7× more volatile than NIV (RMSE std-dev 0.0061 vs 0.0023). NIV needs less smoothing to produce a clean signal (3 – 9 months) than the Fed spread needs (12 – 18 months). The two are complements, not substitutes.

Hybrid allocation.

An optimization over Fed + NIV weights lands exactly on 60% Fed / 40% NIV. The hybrid pays a marginal +0.0024 RMSE penalty relative to the pure Fed model but buys structural diversification — resistance to sentiment manipulation through QE and forward guidance, because the 40% NIV channel enforces real-world physical investment dynamics as a hedge against bond-market reflexivity.

Six out-of-sample validation tests.

1. Calibrated ensemble performance. L2 LogReg + AdaBoost + NN vote; disagreement widens the confidence interval. 98.5% false-alarm filter.
2. Multi-horizon analysis. 3, 6, 12, 18 months. NIV sharpens with horizon; peak AUC 0.8538 at 18 mo.
3. Expanding vs fixed window. The 15-year rolling window flatlines 78.2% of the time and missed the 2023 rate shock entirely (0% vs NIV’s 37.5%). Expanding window is the clear victor.
4. Tactical translation benchmark. NIV vs Fed spread RMSE grid search. NIV wins at 0 – 3 mo lag, Fed wins at 6 – 12 mo; NIV is 2.7× more stable.
5. Component analysis. Gini-ranked feature importance; efficiency_sq dominates at 0.9328. Black-box resolved.
6. Forensic orthogonality. 0.7635 correlation, 41.71% orthogonal variance, 60/40 hybrid allocation. Proves independence from the bond market.

Live dashboard — four-regime read-out.

The dashboard classifies the current NIV score into a four-state regime:

• Expansion — score > 5.
• Caution — score 2 – 5.
• Slowdown — score 0 – 2.
• Contraction — score ≤ 0.

Real-time updates from FRED, component breakdown (thrust / efficiency / slack / drag), confidence bounds derived from inter-model disagreement, and a CSV-export button. Everything is derived on the client from public data — there is no proprietary backend, no token-gated API.

Researcher workbench.

• Scenario simulator. Adjust every parameter — η, investment multiplier, thrust weights, drag weights. Project NIV 1 – 5 years forward.
• Monte Carlo analysis. Thousands of simulations with uncertainty bands. Quantify confidence intervals on the forecast.
• Sensitivity analysis. Tornado plots identifying which parameters drive output variance.
• Fork & customize. Clone any model configuration, build variants with custom weights.
• Reproducibility kit. Python notebooks, Docker containers, full API documentation — reproduce everything locally.
• Audit log. Full logging of every calculation for governance and peer review.

Decomposition into primitives.

NIV is already a composition of four simpler signals. That matters because each sub-primitive is independently useful to the Latent Ocean:

• Thrust (u) — growth and asset-formation net of real-rate drag. Reusable as a “direction” signal for any structural system.
• Efficiency (P²) — the regenerative-capital kernel. Investment-to-output ratio with a 1.15 R&D / education multiplier, squared to reward capital flowing into compounding-margin formation and punish hollow growth. This is where the regenerative-capital thesis lives inside the formula; the Gini-importance result (0.9328) is the empirical receipt.
• Slack (X) — complement of utilization; how much headroom the system has before idle capacity compounds into negative margins.
• Drag (F) — weighted penalty composed of yield inversion, positive real-rate, and rate volatility. Models the credit-circulation inhibitors a DSGE equilibrium assumption cannot.

Given BTUT for coordination and Crystara for structure discovery, NIV is the prototype of a class: small, transparent, composable scalar signals the Latent Ocean can emit to external systems. One is macro; the next will be monetary; the next will be energy-grid. The recipe — known primitives, economic interpretation, walk-forward validation, public data, published weights — transfers.