Skip to content

Reference: Feature Flag and Deployment

All 15-minute power volume models are gated by the time_series.15m feature flag.

What the flag controls

When time_series.15m: true, the following models build:

Numera-core layer:

  • core_endur_volume_15m, core_endur_volume_deal_daily*
  • core_endur_index_pricing_15m_daily
  • fct_volume_15m, fct_volume_15m_deal
  • fct_volume_pwr_15m_agg, fct_volume_pwr_floating_15m_agg
  • fct_volume_pwr_15m, fct_volume_pwr_15m_deal
  • fct_volume_pwr_floating_15m, fct_volume_pwr_floating_15m_deal

EWE wrapper layer:

  • core_ewe_volume_pwr_*__real intermediates
  • fct_ewe_volume_pwr_by_business_case[_deal]
  • fct_ewe_volume_pwr_floating_by_business_case[_deal]
  • fct_ewe_volume_pwr_by_business_case[_deal]_unfiltered, fct_ewe_volume_pwr_by_business_case_with_bu_unfiltered
  • fct_volume_pwr_quarter_fixed_price[_deal]

When the flag is off, none of the above build, and queries against them fail with "relation does not exist".

How to enable

Set the flag in dbt_project.yml:

vars:
  time_series.15m: true

Or set per-target in the same file's vars: block (under a specific target name) if you want it on for dev but off for some test stacks.

Then deploy as normal:

numera -c EWE -s <STACK> dbt -- build --select +fct_volume_pwr_15m

Semantic views and the flag

The semantic views (sv_power_volumes, sv_power_volumes_floating) ref() the EWE wrappers but are not themselves gated by time_series.15m. On stacks where the flag is off, you must exclude the semantic views explicitly or compilation fails on the disabled upstream:

numera -c EWE -s <STACK> dbt -- build --exclude sv_power_volumes sv_power_volumes_floating

See Semantic views for context.

Date-range vars

The view-form gap-fill respects two optional vars to bound the dim_period scan:

vars:
  time_series:
    15m:
      delivery_date_min: "2020-01-01"   # default
      delivery_date_max: "2199-12-31"   # default

These control the outer WHERE dp.start_date BETWEEN min AND max predicate in the view-form unroll. Tightening them narrows the dense grid the view materialises per query. In practice consumers should rely on their own WHERE delivery_date_utc ... clause in the query — that's what drives the actual partition pruning into sat_period_15m. The vars are a backstop, not a primary control.

Verifying flag state on a stack

To confirm whether the flag is on for a given stack:

numera -c EWE -s <STACK> dbt -- compile --select fct_volume_pwr_15m

If the model compiles, the flag is on. If you get "Found X resources, of which 0 selected", the flag is off (the model is disabled, so it doesn't appear in the graph).