risk-xva

Risk Xva

Use risk_xva with an explicit local project directory, relative config path, new output directory outside the project, and data_mode of synthetic, user-licensed or upstream-example. Read input contracts for the supported ORE project boundary. The CLI equivalent is riskdesk xva --project PROJECT --config Input/ore.xml --output NEW_OUTPUT --data-mode synthetic. It needs the optional extra, riskdesk[xva] (risk-setup skill).

Run only a user-provided or inspected local project. The ORE configuration determines instruments, curves, counterparties, CSA/netting, simulation and requested adjustments. Do not invent these inputs. Both simulation and XVA must be active; an old cube is not reused.

Report first

1. Whether the run succeeded. Any ORE error, a failed worker or a missing XVA report is a refusal, not a result. Quote nothing from a failed run, and do not read partial files from its output directory.
2. currency (ORE's baseCurrency), as_of and requested_xva_flags.
3. Then figures, each with the report level (netting set or trade) it comes from.

Sign conventions

Preserve ORE column names and signs. Disabled metrics, null values and zero results are distinct: null is ORE's unavailable value, and a zero in a column whose adjustment was not requested is a disabled calculation. Netting-set and trade rows overlap and must not be summed together.

What not to claim

Positive exposure is not signed mark-to-market, and risk-neutral XVA is not a real-world VaR forecast. The included upstream example validates execution and report translation only. It does not validate all models, regulatory capital, wrong-way risk, SIMM rights or a user's calibration. The process boundary is not a sandbox for hostile XML.

Going deeper

• reference.md: what the adapter checks, runs, keeps and writes, with code lines, and how the result feeds risk_exposure_profile.
• workflows/run-ore-project.md: inspect, run, check and report one project.
• scripts/check_artifact.py OUTPUT/riskdesk-xva.json: one line verdict on a saved result: no ORE errors, XVA rows present, every row matches its report's columns, input hashes recorded, and which XVA columns hold nulls. Standard library only.