Task: Delete the document families nothing reads

The story planned a table for every one of the twenty corpus root elements. Counting the consumers first showed that eight of them have none: no product code reads those documents, so there is no mapper to migrate and no feature that a table would serve. Building eight tables, and their harness cases, would prove a capability nothing uses. They are deleted instead, and a table is added if a feature ever needs one.

• Every corpus root element is classified as supported, migrating or unused, and a check fails when the tree stops agreeing with that.
• A family classified unused has no test that names it, so the deletion stays deleted rather than growing back as an unused test.
• The deleted families' mapper and its tests go together, so no dead code is left behind.
• What remains compiles and the XML suites pass.

Count consumers before mapping anything. The boundary between "migrate" and "delete" is whether the product reads the family, and that is a grep, not a judgement: domain.hpp – the generated binding – is included by six product files plus the mappers, so the families a mapper, a store or a service reaches are the ones that matter.

The census is a decision, and a check. build/scripts/check_xml_family_coverage.py names each family's state with its evidence and fails when the tree disagrees: a supported family without a table, an unused family that an XML family suite still names, or a corpus root element nobody classified. The check is what stops "seven families to delete" from becoming "seven families nobody got to".

Seven are deleted, six are not. Simulation, sensitivity analysis, reference data, script library, collateral balances, SIMM calibration and the cross asset model have no reader at all: nothing in a component's src, service or api touches them, so their tests and the family's xsdcpp-generated types go. The counterparty information and ibor fallback families already have tables and stay supported: a table that is written and proved costs nothing to carry.

Stress testing was reclassified by the review. The first reading of the tree said nothing in product code called stress_test_mapper, and the family was deleted with the rest. That was wrong: the analytics stress test store is modelled on stresstest.xml, its libraries, scenarios and shifts are live repository and service surface, and the mapper is the only importer that document has. The mapper, its two stress suites and the lexical-form suite, which proves the live writer's number rendering on a portfolio case, are restored, and the family is classified as migrating.

What is not deleted here is the xsdcpp-generated type surface for the seven families. It is generated from input.xsd as one unit, so it leaves when xsdcpp leaves, which is the last task.

Manual QA scenarios (scaffolded via compass add test_scenario) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Merged, with the review that followed it applied. Seven families are deleted, not eight: stress testing was restored because its store is live and its mapper is the only importer. The census check now defends the boundary, and the netting set mapper went the same way afterwards.