# Stage 128 — matched common-year cash attribution

This experiment compares 324 identical scenario parameter tuples at each retention rate (25%, 50%, 75%, 100%) against 0%, for 1,296 pairs.

Only committed annual rows common to both variants enter the analysis. At each year, the exact change in corporate cash before credit is decomposed into the difference in opening cash after debt service, retained dividends, net household equity subscriptions and total investment outlays. A strict identity test protects the decomposition. Failure-year candidate plans are **not** treated as realized transactions.

The primary descriptive cohort comprises the 27 paths per retention rate that complete under 0% but fail under retention. We summarize each pair's *last common committed year*, never pooling different calendar years as if simultaneous. This is an accounting attribution conditional on scenario and survival, **not a causal identification of behavioral response**. No empirical calibration.

Reproduce: `python scripts/matched_common_cash_v128.py --output data/processed/matched_common_cash_v128.json`

Results pending verified CI run.

## Verified results — GitHub Actions run 401

All 1,296 matched comparisons passed the research, web and frontend CI checks. Each retention rate has 324 pairs; 270 complete under both policies, 27 switch from completed to failed, and 27 fail under both. No failed-to-completed transitions occur.

For the 27 newly failing pairs per retention rate, the **mean delta in the last common committed year** (retention policy minus 0% baseline; model units) is:

| Retention | Opening cash after debt service | Retained payout | Net equity subscriptions | Investment outflow | Closing cash before credit |
|---|---:|---:|---:|---:|---:|
| 25% | -3.97745 | +18.90500 | -4.49774 | +25.72653 | -15.29672 |
| 50% | -1.47910 | +40.81849 | -4.08573 | +49.46284 | -14.20918 |
| 75% | -7.37737 | +66.89624 | -5.20487 | +76.31682 | -22.00282 |
| 100% | -5.55118 | +94.53379 | -6.59684 | +103.49592 | -21.11016 |

The sign convention is closing delta = opening delta + retained payout delta + equity-subscription delta - investment-outflow delta. The maximum absolute matched accounting residual is 2.14e-14.

**Interpretation:** In the last shared committed year of newly failing pairs, extra retained payouts are outweighed by increased investment outflows and lower opening cash/equity subscriptions. This is a path-conditional accounting decomposition, not a causal identification of why investment or subscriptions changed. Last-common-year dates differ across pairs and policies. The grid is uncalibrated and cannot support population probabilities or real-world forecasts.

Next research should investigate the behavioral source of investment changes and test sensitivity to price/equity valuation assumptions.