# Stage 119 — Annual coupling of redistribution, investment and dividends

## Result

All 54 annual policy paths complete twenty years; 18 pass the annual collection, supplement-payment and recipient-consumption gates. All 18 are in the capital-feedback mode with 10% or 20% dividend collection ceilings.

**The capital-feedback control already meets the declared consumption floors.** These 18 passes therefore do not demonstrate that supplements are necessary or responsible for preservation. The fixed schedules were inherited rather than recalculated from the new control.

At the 90% floor and 10% dividend ceiling, terminal changes against the matched capital-feedback control are:

| Owner tax response q | Real capital | Nominal output | Total dividend payout | Firm cash |
| --- | ---: | ---: | ---: | ---: |
| 0 | -7.64% | -2.24% | -3.58% | +0.31% |
| 0.5 | -3.17% | -0.92% | -1.47% | +0.22% |
| 1 | +1.29% | +0.35% | +0.56% | +0.16% |

These are year-20 differences, not estimated causal effects or policy rankings. Firm cash is a separate stock from productive capital and need not move in the same direction.

## What is now recomputed

An isolated variant of the existing annual v100 engine introduces an owner levy and recipient supplements before household consumption and portfolio decisions. Dividend-only collection is capped by the declared percentage of current owner dividends and available cash. Taxes and supplements have matching cash debits and credits; unpaid assessments and entitlements carry forward.

The engine then recomputes consumption, personal debt repayment, equity issuance/ownership, corporate cash and investment. The capital-feedback mode uses beginning-of-year capital to recompute future output, profits, public revenue and total dividends, following the inherited v74 capital-only elasticity of 0.3. Wages, employment and price schedules remain external.

Each feedback mode has its own freshly simulated no-policy control. Owner equity ownership changes also affect the allocation of dividends. Existing government benefit allocation and budget settlement are recomputed inside the annual engine rather than frozen from the previous monthly tape.

No original simulator files were modified. The zero-policy variant recovers the original annual model exactly in both feedback modes.

## Declared grid

Two modes: external production and capital-only feedback. Three inherited schedules: 90%, 95% and 100% consumption floors. Three owner responses: 0, 0.5 and 1. Three dividend ceilings: 5%, 10% and 20%. There are 54 policy paths and two controls.

The nominal supplement schedules are held at their stage-115 values. The new annual floor denominator is each mode's matched annual-control year-1 cohort consumption, not the prior monthly reference or pre-automation consumption. This change of periodization is explicit; case counts are not directly comparable with stage 118.

A 5% ceiling does not clear the assessed supplement obligations at every annual step. Under external production, even fully collected schedules do not quite preserve the exact annual floors after household and equity paths are recomputed. For example, at the 90% floor, q=0.5 and a 10% ceiling, minima are 89.93% and 89.91%. These small failures are retained; the acceptance tolerance is not widened to call them passes.

With capital feedback, the control itself has higher subsequent consumption and already meets all declared floors. Applying an inherited positive schedule in that setting is an intervention comparison, not a minimally necessary transfer.

## Interpretation of the feedback

Investment remains the residual of output less household consumption, housing upkeep and existing government goods. Under that closure, higher aggregate consumption reduces contemporaneous investment unless output changes sufficiently. q determines how much owner spending offsets recipient spending, so it can reverse the sign of the investment and future-dividend response.

This is a consequence of the model's accounting closure and declared elasticities, not empirical evidence that transfers generally crowd out investment or improve productivity. The dividend base is now responsive in the capital-feedback experiment, but the experiment is still not a full general equilibrium.

## Annual scope and validation

The model settles at annual steps. An annual collection/payment pass cannot establish punctual monthly transfers. It does not replace the monthly timing tests in stages 116–118, and it must not be presented as their integrated extension.

Six tests verify exact zero-policy recovery, tax/payment/arrears and capital identities, responsive output/dividends only in the feedback mode, explicit control-floor attribution, exclusion of a deliberately halted path from success, and archive reproduction. Maximum consolidated cash-identity error in the declared paths is below 1e-8 model units.

```sh
python scripts/annual_funded_feedback_v119.py --output data/processed/annual_funded_feedback_v119.json
python scripts/test_annual_funded_feedback_v119.py
```

[Annual corporate, fiscal and cohort results](data/processed/annual_funded_feedback_v119.json) retain every case, separate gates, controls and source hashes. Full detailed records are reproducible with build(). There is no empirical calibration or claim of full macro integration.

## Next step

Compute supplements from the actual consumption gap in each recomputed annual state, rather than imposing a fixed schedule that can pay even when the matched control already meets the floor. Keep a no-supplement control, explicit funding caps and annual/monthly scope labels. Then test whether an endogenous targeted schedule preserves the floor while changing investment and future income.