# Obligor cohorts and claim loss monitoring

`structured_monitoring` executes portfolio preparation, concentration, monitoring cohort construction and default-count/LGD/EAD validation for the four structured products. It accompanies the product recipes, governance, lifecycle and workout examples. All observations are explicit synthetic records, all forecasts are frozen descriptive reference assumptions, and no fitted performance claim follows from this tiny sample.

Each portfolio contains four equal fully funded claims against three legal obligors; two claims share one obligor. Position HHI is 1/4, while legal-obligor HHI is (1/2)^2+(1/4)^2+(1/4)^2=3/8. These are declared legal identities, not an inferred connected-client classification. The sealed portfolio records each claim once and groups common default units. Exposure concentration does not imply a particular loss dependence or regulatory capital charge.

A separate January 16 obligor forecast cohort has three 30-day physical PDs of 0.5 and two subsequently observed defaults. The four position-level rows contain one repeated obligor, which the count validator deduplicates. Expected defaults are 1.5, observed defaults 2 and their ratio 4/3. A conflicting forecast on the duplicate is rejected. Immature windows contribute neither realized nor expected counts, even when a default is already observed. The issuer/obligor target is explicit: issue/tranche default forecasts from the product recipes are not relabeled as borrower PDs.

The held-out claim ledger has one closed and one unfinished workout per product. A supplied January 16 predefault LGD forecast of 0.5 is compared with the actual losses established in `structured_workout`: 55/80, 32.5/100, 30/100 and 48/80. Forecast-minus-observed biases are -0.1875, 0.175, 0.2 and -0.1. The March 15 open-workout model is deliberately not substituted for this earlier frozen forecast. Unfinished workouts remain excluded, not zero-loss observations. This closed-only estimand may differ from the full default population; no completion independence or causal interpretation is assumed, and no confidence interval is manufactured from a single closed claim.

For securitization, covered bonds and receivables, the supplied fully funded contractual claim balances equal observed balances at the dated default, so monetary EAD bias is zero. Their undrawn amounts are zero: CCF is undefined rather than zero. Counterparty closeout exposure is individually outside the loan-balance EAD backtest because signed netting, collateral and the margin period have different timing. `counterparty_credit_products` instead exercises the dedicated validator, including separate observed ordinary exposure zero and default closeout exposure 80. This exception does not exempt counterparty credit from exposure validation or portfolio monitoring.

Dates, prediction versions, default definitions, claim currencies and observation knowledge dates are retained in packaged records. Changing a forecast knowledge date to after its origin is rejected. Using these examples on real data requires replacing the reference forecasts, legal identities and independently recorded outcomes while retaining those contracts.
