Task: Provision the FOMC meeting schedule into every tenant

The feed synthetic.2026realistic.usdsofrfomc has published nothing in a provisioned tenant since at least 2026-08-12. Its template starts with the pillar SPOT to 1F, and 1F to 8F are SPECIAL tenors of the convention RATES_SPOT_FOMC. A resolution row of that convention names the schedule FOMC_MEETING and a step count, so 1F is the first meeting on or after spot.

On 2026-10-04, a database query found that the system tenant has these rows, but the ACME tenant has the rows without schedule_code and schedule_step_count. The provisioner's copy in projects/ores.sql/create/iam/iam_tenant_provisioner_create.sql does not list the two columns. A row with neither column falls back to the calendar axis, which is spot plus the row's offset. The offset is DAY 0, so in the tenant every FOMC tenor resolves to spot. The first pillar then has zero length. Before the market tick change, the curve pricer refused it with "year_fraction must be positive". After that change, make_pillar_quote_key refuses it first. The cause is the same.

The provisioner also copies no calendar events, and row security on ores_refdata_calendar_events_tbl shows a tenant only its own rows. So after the columns are fixed, the resolver finds no meeting dates and stops with "no event-lookup dates were supplied". Both gaps must close.

The events cannot simply be copied. The insert trigger of ores_refdata_calendar_events_tbl checks that the tenant has the event's calendar and its diary entry type. On 2026-10-04 the ACME tenant had 60 of the system tenant's 61 calendars, all but US.FOMC, because its calendars come from a dataset publish rather than the provisioner. It had no diary entry types, because nothing copies them. So the tenant also needs US.FOMC and central_bank_meeting before its meeting dates.

No pillar check stops a curve from publishing past data. A curve's dates are calculated from its own as-of date, and only a pillar of zero length from spot is refused.

Each step's insert trigger checks the step before it, in the same tenant:

1. countries, which a tenant gets only from a dataset publish, after provisioning;
2. the US.FOMC calendar, whose trigger checks its country;
3. the diary entry type central_bank_meeting, which nothing copies;
4. the tenor schedule FOMC_MEETING, which checks the calendar and the diary entry type;
5. the resolution rows' schedule_code, which checks the tenor schedule;
6. the meeting dates, which check the calendar and the diary entry type.

The provisioner runs before the publish, so it cannot create step 2. It copies no tenor schedules either, so ROLL_QUARTER is missing too, and the CREDIT_CDS_IMM resolutions lose their schedule in the same way as the FOMC ones.

• The provisioner copies the diary entry types.
• refdata.calendars carries US.FOMC.
• refdata.tenor_schedules and refdata.calendar_events are new datasets in the base bundle. Both depend on refdata.calendars. Their artefact tables are generated from the tenor_schedule and calendar_event models, and their seeds copy the system tenant's own rows, so the schedules and the meeting dates are listed once.
• The calendar event publish matches events on calendar, date and diary entry type. A tenant's event keeps its own id.
• The tenor schedule publish sets the schedule columns of the tenant's resolutions from the system tenant's rows. A second publish changes none of them.

Two inbox captures described this failure with a wrong cause, a meeting on today's date or a meeting before the valuation date. They were removed when this task was added.