feat(hydra-gates): a schema can declare that it carries no demo data

[rubenvdlinde]

What this changes

A schema can now say that it carries no demo data:

"Application": { "x-openregister-demo-data": false, ... }

A string is accepted in place of false and is printed as the reason. true and absence both mean generate, so nothing changes for any app that adds no marker.

Why

buildiq's register holds the schemas the app keeps about itself: the apps it has built, their versions, the slug route index, the template store, the export jobs it has run. ADR-111 rule 1 asks for three demo objects per schema, so the generator wrote three of each.

On the demo instance that produced three apps in the Apps list and on the dashboard that cannot be opened. Their applicationVersion rows carry the placeholder a format: uuid property gets, 00000000-0000-4000-8000-000000000000, and a manifest with no pages, so the detail page renders empty and /apps/buildiq/builder/<slug>/ never resolves. Three more sat in the template store beside the four real built-in templates and would clone into an equally empty app. Three export-job rows read as finished exports nobody ran. Measured 2026-09-18 on app c5a2e155-432b-4ea8-a965-56661b991463, slug ccdc.

Every one of those objects satisfies its schema. That is why --check was green on the dataset that broke the demo: conformance is about an object's shape, and nothing in a schema says whether its rows are content somebody authors or bookkeeping the app writes. So the app states it, the way a catalogue already does.

Three places honour the marker

• build() generates nothing for the schema, and it does so before --keep carries objects forward. An app adopting the marker regenerates an existing descriptor, and the objects it wants gone are exactly the ones --keep preserves.
• check() prints SKIP instead of demanding three, so gate 101 cannot ask for the rows the app just removed.
• check() prints FAIL when a mock descriptor still carries objects for the schema. Skipping alone would leave the broken rows in place for ever, and those rows are the defect.

How this was verified

python3 hydra-gates/scripts/lib/test_generate_mock_register.py: 25 tests, exit 0.

Six of them are new. Reverting generate_mock_register.py and rerunning turns five of the six red:

FAILED (failures=5)
  test_a_schema_declaring_false_gets_no_objects
  test_a_string_is_accepted_in_place_of_false
  test_check_skips_an_excluded_schema_instead_of_demanding_three
  test_check_fails_when_the_excluded_schema_still_carries_objects
  test_keep_does_not_carry_forward_the_objects_being_removed

The sixth, test_true_and_absence_both_still_generate, is the control. It is green with and without the change on purpose: it is what separates "the exclusion works" from "the generator stopped generating".

Blast radius

The fleet's apps run gate 101 from ConductionNL/.github@main, so this lands everywhere the minute it merges. No app declares the marker today, so the SKIP arm and the FAIL arm are both unreachable until one does. The only behaviour change for an app that adds no marker is none.

What merges after this

ConductionNL/buildiq marks its five control-plane schemas and regenerates its dataset. That PR's gate 101 needs this one on main first, otherwise the gate asks buildiq for the three built-app rows the PR removes.

🤖 Generated with Claude Code