Modeling decisions for 221100 sector disaggregation

[jvendries]

This discussion is meant to list the modeling decisions used in the implementation of the 221100 disaggregation.
There are 7 decision points discussed in independent comments under this discussion. My suggestion for feedback to this approach is to provide feedback in the order the decisions are posted and to resolve them in that order, as any changes to earlier decisions will affect later ones.

I would like to emphasize that the reason why disaggregation is (initially) built as described in this discussion is for ease of implementation to get something working in the code, rather than as a suggestion that they were the "right" implementation options to make for any of the decisions below. As a result I expect we will modify the code based on the consensus we reach in this discussion for all the decisions.

Below is a summary of the decisions implemented in this approach, and a report with the final status of the disaggregation after implementing these decisions.

Summary table

#	Decision	One-line takeaway	Resolution
1	Schema	405 taxonomy unchanged; 407 outputs only when flag is on	Finish disagg implementation with 407 as flag only, change to base schema in a follow up PR.
2	Make + weights	Split anchored to BEA UGO305 gross output (10 electricity sectors)	Use BEA UGO305 because A) it is other purposes within the code, B) possible further disaggregation of electric generation sectors for which we could leverage the per-generation output in this data, and C) uncertainty regarding how alternate data sources relate to G:T:D splits.
3	Use intersection	Diagonal-only 3×3 block; no cross electricity-for-electricity flows	Keep the values on the 3x3 disaggregated Use table diagonal, but use Purchased Power values instead of Production values from table 8.3 to derive the weights.
4	Use column	Fuels → generation; VA balances each column to gross output	Keep as is.
5	Use row + Y	Equal-price; same gen/T/D shares for all consumers	Keep the UGO305 weights as constraints on the row totals, but adjust the non-intersection row weights such that the UGO305 weights on the totals are kept despite the the Use intersection using other weights.
6	E	eGRID primary; gas-type fallback; distribution ≈ zero direct GHG	Remove fallback.
7	Scaling	Three children inherit parent Utilities summary ratio	Use UGO305 target year values to derive weights for scaling.

[jvendries]

1. Schema expansion — keep the 405-sector taxonomy, expand to 407 only when disaggregation is on

Implementation: bedrock/utils/schemas/cornerstone_schemas.py (CORNERSTONE_*_ELEC, active_cornerstone_commodities(), validate_cornerstone()); reindex helpers in bedrock/transform/eeio/electricity_disaggregation.py (reindex_v_to_elec_schema, etc.)

When electricity disaggregation is enabled, the model needs two more sector codes than the standard Cornerstone table. We chose not to change the canonical 405-sector taxonomy, type definitions, or BEA correspondences that the rest of the codebase relies on. Instead, disaggregation produces 407-sector Make/Use tables only on the flag-on path, with parallel 407-sector schema lists and validation that switches based on config. Downstream matrices (A, q, B, and related quantities) inherit the expanded size automatically once the underlying V/U bundle is 407-wide; a few paths that still assume 405 (validation, expanded gross-output series, emissions) are handled explicitly.

Modeling implications: Configs with the flag off behave exactly as before. Configs with the flag on expose three distinct electricity industries in IO and attribution results, without forcing every consumer of the 405-sector schema to become config-aware. The tradeoff is intentional duplication of schema logic (405 vs 407 variants) rather than a single taxonomy that morphs at runtime.

[bl-young]

i'm commenting on this now after reviewing the PR. My immediate reaction was to set the schema in one place and always draw on the same thing (so any gating could be done at the earliest point). But I realize that given the sequential nature of the disaggregations, first waste, and then electricity, that this becomes tricky.

[jvendries]

Right. I briefly considered what it would take to do both disaggregations simultaneously to get around the need to change the schema during the process but concluded that that approach would require a larger overhaul of the codebase and it seemed to be less generalizable than doing the disaggregations sequentially.

[MoLi7]

I agree with Ben on this one. If we're committed to implementing both waste and electricity disaggregation, they should be combined and turned on at the same time.

I understand the current setup is to do them sequentially and evaluate the impact of electricity in isolation. But I'd nudge us to tidy them up (i.e. generalization) after we're complete the work.

[jvendries]

I understand the current setup is to do them sequentially and evaluate the impact of electricity in isolation. But I'd nudge us to tidy them up (i.e. generalization) after we're complete the work.

Resolution: Initially implement the electricity disaggregation sequentially (i.e., after waste disagg). This means expanding the schema and related code during the electricity disaggregation step. After testing and validation of the electricity disaggregation, re-work its implementation in a follow up PR (or series of PRs) such that the waste and electricity disaggregation are part of the default cornerstone schema and always run when running the cornerstone pipeline.

[jvendries]

2. Make table disaggregation — BEA gross-output shares as the split weights

Implementation: build_electricity_disagg_go_weights() and disaggregate_make_intersection() in bedrock/transform/eeio/electricity_disaggregation.py; data source via load_go_detail() in bedrock/extract/iot/gdp.py (BEA table UGO305-A)

Because of the reallocation step performed prior to the disaggregation, the only value in the Make table for the electricity industries and commodity resides in the Make table intersection (i.e., diagonal) of 221100, which contains the total commodity (which is only electricity) produced by the entire electricity industry.

This intersection must be divided among generation, transmission, and distribution. We anchor that split to BEA detail gross output for the ten electricity-related sectors in UGO305-A: eight generation types plus transmission and distribution. Each sub-sector's weight is its gross output divided by the sum of those ten sectors only (not the full BEA table), so the three Cornerstone weights sum to 1. This step applies these weights to split the Make diagonal while preserving total Make row and column totals for electricity. These disaggregated industry and commodity totals are then used as constraints for the following steps in the disaggregation.

Modeling implications: The aggregate electricity output in the model stays anchored to the reallocated Make table. BEA UGO305-A (2017) supplies only the relative gross-output shares used to divide that total among generation, transmission, and distribution. The relative size of gen vs T&D in the disaggregated model therefore reflects BEA's 2017 output mix, not proportions implied by the single aggregate IO cell (which has no internal structure). All later disaggregation steps reuse the same three weights for consistency.

[bl-young]

If BEA's published output mix differs from the aggregate IO cell, that difference is resolved in favor of the external gross-output benchmark.

I don't follow this. Aren't we saying that the data in the Make table (after reallocating the off-diagonal production) is what we assume is the basis of the total 2211 output? And we just use the UGO305 data from 2017 as the basis for splitting across the 3 sectors? That seems to be the opposite of "that difference is resolved in favor of the external gross-output benchmark".

[jvendries]

Good point. That sentence was poorly worded and your reading is correct, and in line with what the code actually does. That is, the Make diagonal defines the total, the UGO305 BEA data from 2017 defines the split weights, and later steps in the disaggregation use the aggregate Make table values, not BEA Gross Output values.

I corrected the text in the original comment.

[jbws42]

On 2 (Make + weights), the question seems to be whether BEA's 2017 gross shares for the 10 electricity sub-sectors are a good proxy for the G/T/D value split. BEA classifies by primary activity, and vertically integrated utilities don't unbundle into G/T/D establishments. This means that UGO305-A shares may diverge from what you'd get from FERC Form 1 or EIA revenue decompositions (where supply/generation is typically ~55-60% of the retail dollar, transmission ~10-15%, distribution ~30%). What are the actual three weights that come out of UGO305-A, and have you checked them against a FERC Form 1 or EIA bundled-rate decomposition? If they diverge, which do we trust and why?

[jvendries]

@jbws42 Thank you for your thorough review! Some thoughts on your comments for this decision.

What are the actual three weights that come out of UGO305-A

The weights that come out of UGO305-A for 2017 are 34%, 4%, and 62% for generation, transmission, and distribution respectively. These weights remain relatively constant for the years following 2017.

Sector	2017	2018	2019	2020	2021	2022	2023	2024
All Gen	34%	36%	37%	35%	40%	39%	34%	31%
T	4%	4%	4%	4%	4%	4%	4%	4%
D	62%	60%	60%	61%	56%	58%	62%	65%

I should note that there are two assumptions for these weights. One is that the dollar values the weights are based on are nominal, not chained or indexed dollars. The second is that these weights are obtained by summing the individual power generation, transmission and distribution line items in UGO305-A, but do not include the Federal electric utilities or State and local government electric utilities line items. The assumption here is that the Gen:T:D split within the latter two are identical to the split calculated using the other line items, as the UGO data does not provide that split for those sectors.

Line	Description	Included in weight calculation
22	Hydroelectric power generation	Yes
23	Fossil fuel electric power generation	Yes
24	Nuclear electric power generation	Yes
25	Solar electric power generation	Yes
26	Wind electric power generation	Yes
27	Geothermal electric power generation	Yes
28	Biomass electric power generation	Yes
29	Other electric power generation	Yes
30	Electric bulk power transmission and control	Yes
407	Federal electric utilities	No
413	State and local government electric utilities	No

have you checked them against a FERC Form 1 or EIA bundled-rate decomposition?

I have not done a formal check against FERC Form 1, and I am not sure I see any explicit data for EIA bundled-rate split.

A quick search for analysis of FERC Form 1 shows that it supports functional T/D/admin O&M (Fares & King: ~16% / 39% / 45% of TD&A for IOUs, 1994–2014 average) but generation is reported on separate schedules.

From EIA, the only mention of dollar values associated with transmission and distribution I found are from Table 8.3, Revenue and Expense Statistics for Major U.S. Investor-Owned Electric Utilities, in the Electric Power Annual report. For 2017, a weight for production (closest to generation), transmission, and distribution expenses would be 87%, 9%, and 4% respectively, though A) there are more expenses than those 3 in the table, and B) dollar expense structure does not necessarily mirror dollar revenue or output structure.

Could expand on what you mean by "EIA bundled-rate decomposition"?

If they diverge, which do we trust and why?

Given that we are probably going to use using UGO305-A data for other purposes (e.g., inflation calculations and or estimating commodity totals), I feel inclined to use that same source for consistency. Additionally, if we move beyond a G:T:D split, and continue towards a split by generation technology, I believe UGO305-A data is probably useful to reconcile dollar and MWh data for specific technologies. Does this make sense?

[jvendries]

Resolution: Continue with the implementation of the Make table disaggregation as is.
The reasons for this are: A) the use Gross Output BEA data for other purposes within the code, B) possible further disaggregation of electric generation sectors for which we could leverage the per-generation output in this data, and C) uncertainty regarding how alternate data sources relate to G:T:D splits.

[jvendries]

3. Use table intersection — diagonal-only electricity block (high priority for revision)

Implementation: disaggregate_use_intersection() in bedrock/transform/eeio/electricity_disaggregation.py

The Use table has a matching "electricity into electricity" block (the intersection of electricity commodity rows and electricity industry columns). Currently, we split that block only on the diagonal: generation buys only the generation commodity, transmission only transmission, distribution only distribution. Off-diagonal cells in the 3×3 block are zero — there is no modeled purchase of transmission or distribution as electricity commodities by other electricity sub-sectors, and no cross-flow between T&D sub-sectors.

Modeling implications: This is a deliberate simplification of within-sector electricity flows. While this simplies the implementation and balancing of subsquent disaggregation steps, in reality generation may consume grid services while the T&D industries use electricity themselves; those relationships are not represented in the monetary IO structure after the current disaggregation. Indirect effects can still appear through other commodities (fuels, services, etc.), but direct electricity-for-electricity links between sub-sectors are absent.

Related discussion: #70

[jvendries]

@jbws42 Your understanding is right. The current structure is temporary, as I think we do want the final representation to be the vertical chain you mention. I also think it is expected that changing the implemention to match that vertical chain will impact downstream decisions (4, 5, 6, 7), but it is still better to build the chain structure now rather than later.

However, the question remains for how to distribute the original 221100 intersection values into the 3x3 disaggregated sector to achieve that vertical chain. For example:

• What data sources shouuld we use for determining the allocation splits?

  • For the US, an option we could use could be from EIA Table 8.3 from the Electric Power Annual for expenses I mentioned here. Are these numbers representing the vertical integration we are discussing here?

• Related to the data source, what allocation should we use for:

  • The generation commodity consumption by the generation industry (i.e., the generation intersection)?
  • The generation commodity consumption by T&D (i.e., power use by T&D)? Should they be the same or different allocations between T&D?
  • The T&D commodity consumption by the generation industry? An option we've explicitly discussed before for this is that it remains 0 (i.e., generation does not buy T&D, and only uses its own power) and document this assumption.

Once we have a better grasp of these allocations, I can modify the implementation to reflect these choices and re-work the table balance based on the resulting values.

[jvendries]

Further analysis required: Explore the diffrerence between the diagonal split and using Table 8.3 and see what effects that has.

Additionally, update the value used here to reflect Purchased Power, Transmission, and Distribution instead of Production, transmision and distribution.

[bl-young]

When you move to physical units, the natural representation is the vertical chain (consumers buy delivered electricity from D; D buys from T; T from G), and retrofitting that later will reopen 3, 4, and 5. Is the structure the intended end state, or a stepping stone to a vertical chain?

Given the intention to ultimately treat T & D akin to margins, and to avoid adding what would be additional pass-through nodes to a supply chain (e.g. consumer purchases D which purchases G), I believe we are ok simplifying here to a diagonal representation. I recognize that's a bit different from the physical flow of electricity but I think better aligns with most interpretations of use for these factors.

[jvendries]

I did some analysis on what the effects of using weights derived from Table 8.3 in the electric power annual instead of the GO-derived weights would impact this decision. The summary of the findings is as follows (full report can be found here):

• Using weights derived from table 8.3 along the disaggregated 3x3 use table intersection diagonal is feasible
  for industy balancing using VA as a residual, but the weights break commodity balancing by billions, so not feasible without rebalancing or additional methodological refinement.
• Using weights derived from table 8.3 with values placed on the off-diagonal cells of the 3x3 use table intersection will fail VA
  balancing (i.e., VA residual goes negative). Again, not feasible without reblancing or additional methodological refinement.
• Using weights derived from Purchased Power values instead of Production values from table 8.3 are less extreme alternative but do not resolve clearing or VA failures.
• Table 8.3 expense shares should not replace UGO305 wholesale; the weight gap is too large and structurally mismatched (IOU expenses vs national GO).

[jvendries]

Resolution: Keep the values on the 3x3 disaggregated Use table diagonal, but use Purchased Power values instead of Production values from table 8.3 to derive the weights.

The reasons are A) the more generation-heavy split seems to better reflect the consumption profile of the disaggregated 221100 sector, and B) simplification along the diagonal given the intention to ultimately treat T & D akin to margins, and to avoid adding what would be additional pass-through nodes to a supply chain (e.g. consumer purchases D which purchases G), as commented here

[jvendries]

4. Use table column disaggregation — fuel routing and value-added balancing

Implementation: _enforce_go_identity_precondition(), _split_aggregate_column_by_rule(), and disaggregate_use_industry_columns() in bedrock/transform/eeio/electricity_disaggregation.py

This step splits the 221100 industry column in the Use table and Value Added rows — everything the electricity sector buys as inputs. This cannot reuse the generic waste-disaggregation column logic because each disaggregated column must balance to its own gross-output target. Before splitting, we check that the aggregate electricity industry satisfies the standard gross-output identity (Make row total equals domestic + imported intermediate use + value added, using total use including imports). Small residuals are absorbed into aggregate value added with a warning; large mismatches fail loudly. Input rows are then assigned by rule: fuel commodities go entirely to generation (coal, oil & gas, refining, petroleum products, natural gas); all other non–value-added rows split by the same gross-output shares as in the Make table disaggregation; value added is computed as the residual so each column sums exactly to its target gross output, distributed across compensation/proprietors/taxes in proportion to the original aggregate VA mix. Any negative VA that results from this implementation is retained with a warning rather than clipped, though in reality this is unlikely to happen as VA is over 60% of the total 221100 industry output.

Modeling implications: Generation carries essentially all fuel purchases; T&D columns are dominated by services and VA rather than fuels — consistent with physical intuition but enforced by rule, not by BEA detail Use data at this granularity. VA acts as the balancing item, so generation can show small or even negative VA if fuel assignments exceed its gross-output share, while distribution VA is comparatively large. That skew affects input coefficients (A) for those columns and can interact with A scaling caps in non-base years (see the Cross-year scaling alignment decision).

[bl-young]

is one alternative to using value add as the residual to use the non-value add / non-fuel commodities as the residual? Since we by definition are giving generation a greater share of the commodity inputs this will inherently mean that its value added share will decrease.

[jvendries]

Yes this is an option we can pursue, the only caveat being that VA has a far greater value than the non-fuel intermediate rows. So using the latter for residual may result in larger changes to the U_norm (and consequently A) coefficients since the numerator for those has a greater change. We'd have to look into this more closely to really see the difference in effects.

Conceptually though I think your suggestion is valid.

[jbws42]

On 4 (Use column), splitting all non-fuel intermediate inputs by the same GO weights seems like a strong assumption. FERC Form 1 O&M expense categories (production, transmission, distribution, customer accounts) map directly onto this split and are public. Did you consider FERC Form 1 expense shares as row-specific split weights for the larger non-fuel input rows, rather than uniform GO shares?

[jvendries]

@jbws42 I did not look directly at FERC Form 1, but if I undertand correctly, Electric Power Annual Table 8.3 aggregates that data for Major Investor-Owned utitlies. This covers IOUs only (~75% of U.S. customers), not munis/co-ops/IPP-heavy regions. Does that provide adequate representation? If so, here are a few numbers from that table we could potentially use from electric utility operating expenses (operation, million 2017 dollars):

Category	$M (2017)	Share
Production O&M (excl. fuel & purchased power)	17,664	47.0%
Transmission	10,804	28.7%
Distribution	4,358	11.6%
Customer accounts	4,789	12.7%

This could be mapped to the IO columns in terms of commodity use as: 47% to generation column, 29% Transmission, and 24% Distribution (assuming we can combine Distribution with customer accounts).

Is this similar to what you had in mind?

[jvendries]

Resolution: Leave as is, and come back to it in PRs down the line if it becomes as problem.

[jvendries]

5. Use table row and final demand — equal-price split across all consumers (high priority for revision)

Implementation: disaggregate_use_commodity_rows() and disaggregate_electricity_commodity_row_in_y() in bedrock/transform/eeio/electricity_disaggregation.py; Y wiring in derive_disagg_Ytot_with_trade() in bedrock/transform/eeio/cornerstone_disagg_pipeline.py

This step splits the 221100 commodity row — electricity purchased by every other industry and by final demand. We assume consumers face a single blended retail price, so every non-intersection cell in that row (and the corresponding final-demand entries) receives the same gen/T/D proportions. Conceptually, each sub-commodity row should "market-clear" to its gross-output share of total electricity supply; because the Use table intersection decision already placed the intersection using the same weights, that condition reduces to applying the shared weights w everywhere on the row. Any change to that decision will affect this step, but currently the Use (domestic and imported) and final demand are split independently with the same weights.

Modeling implications: All downstream purchasers of electricity — households, other industries, exports — inherit the same gen/T/D mix in their direct electricity purchases. The model does not distinguish, for example, high-voltage transmission service sold to industrial customers vs. distribution to residences at this step. Totals are preserved: summed disaggregated electricity consumption equals the original aggregate row. Differentiated consumer-level pricing or T&D-specific sales requires additional rules beyond equal-price proportional splitting.

More broadly, while this decision is still done using monetary units, it is analogous to the decision of how many MWh of consumption to assign to each industry/FD when we introduce physical units. As a result, it is worth discussing and revising now if necessary in preparation for that step.

Related discussion: #70

[jbws42]

On 5 (Use row + Y), equal price overstates distribution purchases (and understates G purchases) for electricity-intensive industries. Since G carries the emissions, this biases sectoral electricity EFs for the heavy-industry sectors people scrutinize. EIA Form 861 sales and revenue by customer class would support at least a three-class differentiation (industrial/commercial/residential) mapped to industries vs. household final demand. Since this decision is explicitly the monetary analogue of the future MWh assignment, what's the plan for class-differentiated splits (i.e., do it now in dollars, or wait for physical units)? And can we get a sensitivity showing how much industrial-sector electricity EFs move under a Form 861 based split vs. equal-price?

[WesIngwersen]

An equal split does not achieve our goal of adding value by separating out physical use from marginal costs. I'm pasting here our internal thread where I made a recommendation on treating T&D as margins, because within that thread my case was based on the T&D costs NOT being proportional to physical use.

Making a case for modeling electricity transmission and distribution as margins.  It appears to be a significant source of cost differences in region and for different user types that is as of now muddled in the delivery of electricity at producer price. It would be more consistent with how delivery of other fuel sources are handed. This way we can actually have different prices as well for different users which is more realistic (for purchaser price but not for producer price), and also better reflects the physical delivery system and how its represented in other LCA databases. Other LCA databases have traditionally used delivery of different voltage types to different customers, where heavy industry has "high voltage" or "medium voltage" inputs and small users like households use "low voltage". In part this represents where more substations to step down the voltage are required for distribution to household type users.  Heavy industry will get electricity right from transmission lines, akin to getting it from wholesalers, and have their own internal substations to lower it for their own use.  We could use the physical system  information at least by industry type to partition the transmission and distribution margins to sectors.

[jvendries]

@jbws42 and @WesIngwersen I agree that we do not want an equal split long term.

Similar to my comment in Decision 3, the question is, how do we allocate?

Should we use Electric Power Annual Table 2.4, Average Price of Electricity to Ultimate Customers by End-Use Sectors 2014 through 2024 (Cents per kilowatthour), to weight the splits and map Industry Codes to one of the end-use sectors?

Total Electric Industry

Year	Residential	Commercial	Industrial	Transportation	Total
2014	12.52	10.74	7.10	10.45	10.44
2015	12.65	10.64	6.91	10.09	10.41
2016	12.55	10.43	6.76	9.63	10.27
2017	12.89	10.66	6.88	9.68	10.48
2018	12.87	10.67	6.92	9.70	10.53
2019	13.01	10.68	6.81	9.66	10.54
2020	13.15	10.59	6.67	9.90	10.59
2021	13.66	11.22	7.18	10.20	11.10
2022	15.04	12.41	8.32	11.59	12.36
2023	16.00	12.59	8.04	12.77	12.68
2024	16.48	12.75	8.13	12.75	12.94

[jvendries]

Resolution: Keep the UGO305 weights as constraints on the row totals, but adjust the non-intersection row weights such that the UGO305 weights on the totals are kept despite the the Use intersection using other weights.

In the follow up PR that converts electricity to physical units, apply the Electric Power Annual Table 2.4 prices. This will result in higher MWh consumption for those sectors with lower prices attributed to the mapping. Determine if further balancing is needed.

Consider an additional follow PR to include manual adjustments for sectors that don't purchase distribution, e.g., industrial sectors, to have the distribution value be 0 for that industry. For example, we could look at top consumers of electricity for this manual allocation.

[jvendries]

6. Emissions (E) attribution at 407 sectors and use for B

Implementation: E (numerator) — primary path: _apply_electricity_disagg_cornerstone_mapping() and eGRID FBS loading in bedrock/transform/allocation/derived.py; fallback: split_electricity_e_for_disaggregated_b() in bedrock/transform/eeio/electricity_disaggregation.py. x (denominator in E/x) — derive_cornerstone_B_via_vnorm() in bedrock/transform/eeio/derived_cornerstone.py; Make row shares applied only on the BEA time-series x path via distribute_electricity_aggregate_x_using_v_row_shares() (called from distribute_waste_parent_x_using_v_row_shares() in bedrock/transform/eeio/cornerstone_disagg_pipeline.py when building derive_cornerstone_x_after_redefinition()).

Emissions intensity (B) is B = (E / x) @ Vnorm, so both E and industry gross output x must sit on the 407-sector index. These are handled separately:

E (numerator). The preferred source is the eGRID-based GHG inventory, which already resolves plant-level combustion to generation NAICS and SF6 to transmission; with an updated activity-to-sector crosswalk, emissions land on 221110 and 221121 directly — no Make row shares involved. If eGRID is unavailable, a fallback routes aggregate 221100 emissions by gas type: SF6 to transmission (221121), other listed GHGs to generation (221110). Distribution (221122) receives negligible direct GHG under both paths (SF6 is attributed to transmission).

x (denominator). Make-table row shares are used only when B takes x from the BEA gross-output time series expanded to Cornerstone (derive_cornerstone_x_after_redefinition()). That path is active when use_E_data_year_for_x_in_B is True (typical Cornerstone full-model configs) or when deflate_x_to_detail_io_year_for_B is True. Because expand_vector still maps electricity to a single 221100 parent at 405, we split that parent total across 221110 / 221121 / 221122 using each child's share of aggregate electricity gross output from the disaggregated Make table (compute_x(V) row sums from PR3 V). When B instead uses Make-derived x directly (derive_cornerstone_x() = compute_x(V) with use_E_data_year_for_x_in_B False), V is already 407-wide and no row-share split is needed.

Modeling implications: Generation dominates direct power-sector GHG; transmission carries SF6; distribution is effectively zero direct emissions — a physically reasonable split for E, but not a full lifecycle or locational attribution. Indirect emissions from fuel supply chains remain in upstream sectors (e.g., fuels assigned to generation in the Use table column disaggregation decision). The Make row-share split affects E/x ratios (and therefore B) only on configs that align E to the GHG data year while using BEA time-series gross output as the denominator; it does not change how emissions are inventoried, only how the aggregate Utilities gross-output total is apportioned among the three electricity industries for normalization.

[bl-young]

If eGRID is unavailable, a fallback routes aggregate 221100 emissions by gas type: SF6 to transmission (221121), other listed GHGs to generation (221110).

I don't believe this is a case we need to prepare for. If the targeted FBS dataset is not available it should error.

I'm having a hard time seeing why we would need to make any distinction at this step. The denominator should be treated for these sectors in exactly the same way as all other sectors.

[jvendries]

If eGRID is unavailable, a fallback routes aggregate 221100 emissions by gas type: SF6 to transmission (221121), other listed GHGs to generation (221110).

I don't believe this is a case we need to prepare for. If the targeted FBS dataset is not available it should error.

Agreed, the fall back should be removed. Adding a possible resolution at the end of this comment to keep track of this.

I'm having a hard time seeing why we would need to make any distinction at this step. The denominator should be treated for these sectors in exactly the same way as all other sectors.

Re: x as denominator: I think the denominator is treated the same as other sectors except for the fact that we have to expand the output of derive_cornerstone_x_after_redefinition (which is used when use_E_data_year_for_x_in_B: true, as it is here). The output of that function is expanded from 405 to 407 sectors using the Make shares, which are the same as the BEA Detail Gross Output shares for electricity; this data set is in turn the source data for derive_cornerstone_x_after_redefinition, so it should be consistent.

Possible edit: remove fall back and fail with an error if the eGrid FBS dataset is not available.

[jvendries]

Resolution: Remove fall back and fail with an error if the eGrid FBS dataset is not available; otherwise keep as is.

[jvendries]

7. Cross-year scaling alignment

Implementation: load_bea_v2017_summary_to_cornerstone() in bedrock/utils/taxonomy/bea_v2017_to_ceda_v7_helpers.py; consumed by scale_cornerstone_A, scale_cornerstone_q, and scale_cornerstone_B via get_bea_v2017_summary_to_cornerstone_corresp_df() in bedrock/transform/eeio/derived_cornerstone.py and bedrock/transform/eeio/cornerstone_year_scaling.py

When scaling 2017 IO structure to the model year, BEA summary tables provide one price index per summary sector. Aggregate 221100 maps to BEA summary code "22" (Utilities). After disaggregation, 221100 no longer exists in the matrix; we replace it with the three child codes in the summary→Cornerstone map so all three inherit the same Utilities summary ratio as the parent would have. This is a replace, not an append — leaving 221100 in the map would error once the aggregate column is gone.

Modeling implications: Generation, transmission, and distribution scale identically across years — same relative structure as the pre-disaggregation aggregate sector had. The disaggregation introduces sector detail within 2017 dollars but does not, by itself, introduce differentiated year-over-year growth rates among the three electricity sub-sectors. For a 2017-base config this is a no-op; for model years like 2023 it prevents the three new codes from silently staying at 2017 levels while the rest of Utilities scales. If generation's VA went negative (decision 4), year scaling may further adjust that column through the A-matrix sum cap (0.98), which is a known coupling worth monitoring in diagnostics.

More broadly, this scaling adjustment for the disaggregated sectors is needed regarless of the intended replacement of monetary units for physical units for these sectors. That replacement is intended to be implemented at the IO model target year rather than during the 2017 disaggregation procedure (i.e., after scaling) to allow the scaling procedure to work as-is with full monetary values.

Note that this implementation is focused on the current bedrock scaling path; any changes to this path based on the broader scaling discussion will need to be accounted for here.

Whether it makes sense to deviate from the current scaling for the disaggregated electricity sectors or not (e.g., by using the BEA Gross Detailed Output data for target years instead of scale + inflate) before replacing them by physical units is an open question but one that was not implemented here.

[jbws42]

On 7 (Scaling), inheriting the parent deflator means G, T, and D scale identically 2017 through 2023, but this is a period where they diverged (T&D capex boom and rising delivery charges vs. gas-price-driven swings in G revenue). What would prevent using annual UGO305 shares to differentiate the three children's year-scaling?

[WesIngwersen]

Yep I agree with @jbws42. Again the advantage of splitting the sectors it to have the flexibility to treat each uniquely using more specific information.

[jvendries]

What would prevent using annual UGO305 shares to differentiate the three children's year-scaling?

I don't think there is a technical barrier from using the UGO305 shares for scaling, other than the fact that it would be a different approach from what it is currently done for all other sectors.

Possible edit: Use UGO305 to differentiate the three disaggregate sectors scaling values.

[jvendries]

Resolution: Use UGO305 to differentiate the three disaggregate sectors scaling values.