# MacroVision v99: corporate debt service and household lender exposure
29 October 2026. Audit of unchanged v98 simulations, not default estimation or a new restructuring model.

## Main finding
The v98 result that credit completes 18 additional actual paths needs a material qualification: **only 2 of those 18 borrowers pay all scheduled corporate interest and principal in cash throughout the observed path**. The other **16 capitalize interest and have at least one scheduled-principal shortfall**. Five have at least two consecutive accepted annual observations with interest capitalization.

| Terms | Actual complete | Completed actual borrowers | Full scheduled cash service | Any capitalized interest | Any principal shortfall | >=2 consecutive capitalization years |
|---|---:|---:|---:|---:|---:|---:|
| Disabled | 202 | 0 | 0 | 0 | 0 | 0 |
| Limit 20, 4% | 212 | 10 | 2 | 8 | 8 | 0 |
| Limit 60, 4% | 220 | 18 | 2 | 16 | 16 | 5 |
| Limit 60, 8% | 220 | 18 | 2 | 16 | 16 | 5 |

All 202 completed nonborrowers remain separate; absence of debt is not classified as successful borrowing. In this grid the completed actual borrowers coincide with the additional completing cases. The corresponding control variants each have 72 completed borrowers, all paying scheduled service, plus 72 completed nonborrowers.

No completed borrower has zero debt at the horizon. This is not automatically a default: the inherited rule amortizes 10% of each year's opening principal and sets no finite contractual maturity or terminal balloon requirement. Full scheduled cash service is therefore different from complete debt repayment.

These are counts on a selected hypothetical grid, not frequencies in a population or default probabilities. Matching counts at 4%/8% do not mean identical loan balances or outcomes.

## Definitions fixed in the audit code
- Interest due = scenario annual rate times opening corporate claims.
- Interest not paid in cash = due minus cash paid, matched to v98 capitalized interest.
- Scheduled principal = 10% of opening corporate claims; shortfall = max(0, scheduled minus cash principal paid).
- Full scheduled service requires actual borrowing and no interest capitalization or principal shortfall in any accepted annual observation.
- Persistent capitalization is a diagnostic of at least two consecutive annual observations, not a legal default trigger, covenant or estimated risk threshold.
- Completed classifications require year 20. Prefix exposure metrics on failed paths are explicitly observed-to-stop values; no truncated final balance is called terminal debt.

The sum of annual scheduled-principal gaps is a diagnostic sum, not an arrears ledger or recoverable contractual claim: unpaid principal remains part of outstanding debt and may recur in later schedules. Missing attempted failure-year transactions are excluded, as in v98.

Maximum terminal debt among completed actual paths is 18.4529 at limit 20/rate 4%, 42.2039 at limit 60/rate 4%, and 43.1141 at limit 60/rate 8%, in normalized nominal model units. No completed actual path has negative modeled residual equity, but this is not a solvency certification or a statement about truncated paths.

## Lender exposure
Loans use household cash buffers and remain illiquid par claims. For the limit-60/rate-4% actual grid, the maximum recorded claim/(cash+corporate claim) ratio is approximately 99.99% for every cohort. This narrow denominator excludes equity, housing and other assets/liabilities; it is **not** share of total wealth or probability of loss. The imposed pro-rata cash sweep can make narrow liquid portfolios extremely concentrated.

As a static diagnostic only, compute 25% and 50% of terminal claims in completed paths. Maximum gross 50%-haircut exposure by cohort in the limit-60/rate-4% actual grid:
| Cohort | Maximum gross terminal claim exposure under assumed 50% haircut |
|---|---:|
| Workers | 8.0475 |
| Displaced, no initial consumer debt | .5192 |
| Displaced, initial consumer debt | .4272 |
| Productive owners | 12.1081 |

Values are aggregate cohort claims in normalized nominal units, not per-person/currency amounts. Maxima can occur in different scenarios and must not be added. The same cohort may also own shares, so gross loan exposure is not net household loss.

This diagnostic does not apply a haircut to the running balance sheets, forgive borrower debt, change consumption or compute losses after equity/credit recovery. It is not expected loss (no PD/LGD), a realized lender loss, or a recovery waterfall. A dynamic haircut requires explicit borrower liability relief and/or proceeds, equity effects and counterparty accounting; none is silently introduced here.

## Interpretation for the thesis
Credit relaxes some first-stop cash constraints, but most additional completing actual paths rely on unpaid interest. Those paths are not evidence that lending makes the transition sustainable. The source's cash sweep also transfers household liquidity into risky illiquid corporate claims, so merely counting completed simulations hides lender-side fragility.

The two fully serviced additional paths are conditional model examples, not empirical proof of the automation thesis. Macro prices, benefits and production still use inherited hypothetical closures. US credit evidence or access enquiries do not calibrate these debt rules.

## Priority next
Specify corporate default/restructuring **before** another expansion: a trigger based on missed cash service, lender recovery resources, borrower liability treatment, equity losses, operating continuity and output feedback. Keep contractual cash service, accounting continuation and post-resolution recovery separate. Fix the rule and tests before comparing outcomes; do not choose it to maximize completions.

A cash-flow shortfall alone is not enough to infer economic insolvency or determine recovery. Capital and equity are not cash proceeds until an explicit transfer or sale counterparty exists. Avoid an unbacked lender reimbursement or debt deletion.

## Reproduction and checks
```bash
python scripts/debt_service_audit_v99.py
python -m unittest discover -s scripts -p 'test_*v9[6789].py' -v
```
Full local output: data/processed/debt_service_audit_v99.json. Checked-in aggregates: data/processed/debt_service_summary_v99.json. Source hashes pin the audit and inherited dependencies.

Eleven combined tests pass (two new tests distinguish nonborrowers, reconcile audited debt movements and validate exposure arithmetic). Every accepted observation in the 4,608-path rerun checks interest cash/capitalization and loan stock identities within 1e-8. All 4,608 completion classifications match v98. No model, loan terms, credit allocation, default rules or valuation policy was changed.

References: [v98 credit channel](CORPORATE_CREDIT_V98.md), [v97 financing boundary](FINANCING_DIAGNOSTIC_V97.md). This report changes interpretation, not empirical calibration.