Counterparty public and financial review
=======================================

Run examples/counterparty_public_review.py or source the native R example and
call run_counterparty_public_review(). The same synthetic2024 counterparty
position links source ingestion, routing, versioned assessment, approval,
performing monitoring, payment delay, an unaccepted restructuring proposal,
February1 default, May1 settlement and frozen replay. The source account value80
is the explicitly supplied claim exposure. One realized recovery32 gives loss48
and LGD0.6. Those contractual ledger observations are not generated by a simulated
benchmark path. The supplementary collateralized counterparty product example
executes signed netting, margin calls, default freeze and closeout mechanics,
with EAD80 and expected loss24; it remains a separate economic-contract scenario
and is never asserted to be the same observed2024 market-price history.

Local position and release CSVs are read through public APIs. Source lineage
checks one account, one selected historical release and independently supplied
balance80. The future release with x99 is unavailable to the January1 estimate.
The existing real governance helper executes route/applicability, model registry,
assessment, domain-specific policy, negative legal gate, freshness and historical
replay. No post-calibration is declared explicitly with a versioned identity
contract. Actual January15 monitoring uses only the same position's February1
default and coverage through May1, giving one case and Brier0.25 from PD0.5.
Additional helper diagnostic accounts remain clearly separate diagnostics; they
are not represented as this position's history or as a population validation.

An illustrative internal grade is obtained from prespecified PD cutoffs0.25
and0.75. There is no external rating supplied. benchmark_ratings executes its
full matching contract with an empty external history and reports NOT_APPLICABLE,
zero included matches and no invented agreement score. A declared potential
external mapping schema is not an external observation or evidence of accuracy.

support_transfer_backtest receives separately dated actual entity-default and
final-default stages. The transfer overlay is disabled by this domestic-EUR
scenario definition, not empirically estimated to have zero risk. Its zero
forecast contribution does not create an observed no-transfer label. Transfer,
joint-transfer and unsupported-parent counterfactual stages have no purported
observed outcomes and remain NOT_APPLICABLE. Actual entity and final default
stages each have Brier0.25; no support effect is identified. The validator rejects
adding a fabricated observed zero to an explicitly unobservable transfer stage.
A single claim cannot support a confidence interval or population performance
claim, so uncertainty is explicitly none.

The scenario does not apply bank/insurer/fund/sovereign financial ratios,
sector scorecards, operating-cashflow simulations, distributed macro lags or
parent-support/transfer overlays to a netting claim merely to fill an API list.
Those sector-specific assessment methods are not applicable here. The frozen
binary PD model and the separate signed-netting/CSA economic reference are the
chosen assessment mechanisms. A drawn-loan exposure schedule is also explicitly
not applicable to this derivative counterparty claim. Policy reasons are read
from the sealed original decision, not reconstructed from later outcomes.

Replay reruns the pinned source/model/policy and explicit identity calibration.
It respects the later-known January16 source correction, deduplicates the first
delivery and remains unchanged by a subsequently registered model replacement.
A deliberately altered checkpoint is BLOCKED. This source-correction scenario
is explicitly separate from the unchanged January15 forecast being backtested;
its revised input is not substituted into the original performance evaluation.

return_artifacts=True (TRUE in R) retains the original sealed lineage,
governance, monitoring inputs, workout, replay, benchmark and stage-validation
inputs, together with the separately labeled economic reference. Default tests
exercise the complete small fitted chain, local report reexecution, same-claim
checks and negative controls. Python and native R financial/report outputs are
compared without rounding or replacing original seals. Source verification does
not substitute for the installed-package acceptance gate.
