`apply_inflation_to_V: True` on full model yaml introduces dollar-year inconsistencies in Cornerstone pipeline

[jvendries]

Context

#460 adds apply_inflation_to_V: True to 2025_usa_cornerstone_full_model.yaml.

The flag inflates the Make table V from usa_base_io_data_year (2017) to model_base_year (2023) via industry price ratios. A separate config (2025_usa_cornerstone_full_model_A_commodity_price_index.yaml) already bundles this flag with scale_a_matrix_with_commodity_price_index: True, which leaves unclear how V inflation as implemented in #412 and refactored in #453 should interact with the default full-model scaling path alone.

Problem

With waste disaggregation on and default A scaling (summary ratio 2017→2022 + industry PI 2022→2023), turning on apply_inflation_to_V creates multiple dollar-year mismatches:

1. Disagg base A (_derive_cornerstone_Aq_from_disaggregated)
   • q and x use uninflated 2017 V
   • Vnorm uses inflated 2023 V
   • A = Unorm(2017) @ Vnorm(2023) mixes years in the structural A matrix
   • scaled_q is 2017 while public derive_cornerstone_q() returns 2023 q
2. Scaled A (derive_cornerstone_Aq_scaled default branch)
   • Scaling assumes base A is entirely 2017 detail dollars
   • Base A already embeds 2023 Vnorm when the flag is on
   • Risk of overlapping price effects (V row inflation + summary scale + industry PI on A)
3. B (derive_cornerstone_B_via_vnorm)
   • A separate discussion topic (reconciling q used for A and B, but relevant here as well since this flag affects the values of A as the inflated V is used via Vnrom

See diagram below for a graphical representation of points 1-3 above in the pipeline:

Issues in pipeline figure

Unaffected paths

U, VA, Y, default derive_cornerstone_x(), and published uninflated V remain 2017 chain dollars.

Questions for discussion

• Should apply_inflation_to_V be enabled on the production full model yaml at all, or only on analysis / commodity-PI configs?
• If yes:
  • Should we fix disagg Aq to use inflated q (and optionally inflated x for Unorm), or keep structural 2017 x and only inflate Vnorm?
  • How should we address the failure of test_ita_q_at_2017_matches_derive_cornerstone_q in Add V inflation flag to full model yaml #460?
• How does this interact with the discussion in Refine q estimate #457?

References

• derived_cornerstone.py: derive_cornerstone_q, _derive_cornerstone_Aq_from_disaggregated, derive_cornerstone_Aq_scaled, derive_cornerstone_B_via_vnorm
• Comment at _derive_cornerstone_Aq_from_disaggregated (mixed dollar-year, intentional for now)
• Bundled config: 2025_usa_cornerstone_full_model_A_commodity_price_index.yaml