# Stage 125 — failure guard decomposition

Split the previously combined stop into separately recorded firm-cash, real-capital, negative-equity, and aggregate-equity flags, retaining the original stopping behavior. Multiple flags may be true. This is diagnostic instrumentation, not a financing intervention.

Re-run the original 1,620 matched v124 paths under the same parameters. Report category by retention rate, and distinguish cases where cash and equity fail together. Do not infer that a negative equity quote or redemption failure is a real resource shortage. Compare counts only after CI passes; scenario counts are not empirical probabilities.

Reproduce: `python scripts/retained_earnings_v125.py --output data/processed/retained_earnings_v125.json`

## Executed findings — GitHub Actions run 367

All 1,620 paths were replayed. The completion and qualified-success counts match v124 exactly (1,377 and 813, respectively), as expected for diagnostic-only instrumentation.

| Retention rate | Cash failures | Real-capital failures | Equity-redemption failures | Total failures |
|---|---:|---:|---:|---:|
| 0% | 27 | 0 | 0 | 27 |
| 25% | 54 | 0 | 0 | 54 |
| 50% | 54 | 0 | 0 | 54 |
| 75% | 54 | 0 | 0 | 54 |
| 100% | 54 | 0 | 0 | 54 |

Every stopped path reached the negative firm-cash condition; none had a simultaneous capital or equity-claims failure under the evaluated guards. These counts are scenario-grid outcomes, not probabilities. They establish the proximate accounting stop, not its causal economic source.

Next experiment: decompose firm cash into opening balance, withheld payouts, gross investment outlays and net household equity subscriptions, and isolate the contribution of changed consumption and stock transactions.