# Fleet methodology and reproducibility

## What is being measured

This is a source-backed institutional ownership/control research ledger for an explicitly defined physical generating inventory. It is not a complete beneficial-owner register, regulatory market-power analysis, weather-performance panel, or accounting consolidation of every investor. The phrase ultimate_control_parent is an institutional endpoint: a public corporation, municipal/cooperative entity, named private platform, fund manager/control group, or explicitly identified joint group. Fund limited partners, public-company shareholders, tax-equity allocations and preferred claims often remain unresolved.

The main concentration finding is particularly well supported because the large thermal companies disclose their portfolios and the important co-owner fractions. Claims about the much broader renewable and battery universe require the accompanying evidence tiers and caveats.

## Source populations and dates

EIA annual Form 860 final 2021 and final 2025 provide plant, generator, utility and ownership schedules. September 2026 Form 860 M provides the latest monthly physical inventory used here. Original URLs, byte counts and SHA 256 hashes are in the raw source manifests and prepare_raw.py. The monthly inventory was released October 23, 2026; the annual 2025 data were released September 10, 2026.

Annual plant membership is balancing-authority code ERCO, not a Texas state filter. The actual EIA code is ERCO. One independently documented correction includes Jack A. Fusco/Brazos Valley, EIA 55357: annual 2025 reports MISO but DOJ and ERCOT identify the plant in ERCOT, and monthly 2026 reports ERCO. The raw BA and correction source are retained. Frontera 55098 is not included in 2021, when connected to Mexico; it is included in the current ERCOT inventory after returning in 2023.

The main current file enumerates every plant in this EIA population, including non-OP operable rows. It is not claimed to enumerate every ERCOT registration. EIA's reporting threshold is a plant total of at least 1 MW connected to the grid and able to draw or deliver power; smaller resources and some distinctions among distributed installations do not appear in the same way. A full ERCOT resource-code-to-EIA-generator crosswalk has not been completed.

ERCOT's May 2026 Generation Resource Forecast independently identifies Antelope 57865 and Elk 58835 as switchable resources although EIA assigns them to SWPP. These are retained in a separate exception ledger and appended to the expanded census with membership_class=ERCOT_switchable_non_ERCO. The same ERCOT source reports availability deductions for switchable capacity, including Tenaska Frontier, Gateway and Kiamichi. Physical capacity in this census must not be read as capacity committed to ERCOT every hour. No arbitrary availability fraction has been imposed.

The two annual baselines describe December 31, 2021 and December 31, 2025. The current physical snapshot describes September 30, 2026, with a transaction review through October 30. October plant additions are not comprehensively included. The 2021 ownership snapshot is after Winter Storm Uri: it cannot classify February 2021 owners without backdating transactions.

## Owner claims and parent mapping

Generator ownership code S means single ownership by the respondent; its respondent is therefore used as EIA's reported owner, not merely assumed from the operator name. Code J means joint ownership and W means wholly owned by an entity other than the respondent. Schedule 4 covers both J and W. The appropriate Schedule 4 fractions are expanded into separate rows and must sum to one for each generator. Thus an absent Schedule 4 row is not evidence that the operator is the ultimate owner.

The 2026 monthly workbook does not contain an owner schedule. Annual 2025 legal-entity claims are carried forward only for matching plant and generator keys. New monthly generators have unresolved reported owners unless an explicit cited ownership override is supplied. fraction_confirmed_from_annual distinguishes the two. Current primary sources can establish the sponsor/control of a new unit without proving all economic or tax-equity interests.

The parent crosswalk is a documented rule table. Rules have year bounds, optional plant/owner/operator/technology restrictions, source URLs, claim types and confidence. More specific project rules can override general legal-entity patterns. The supplemental battery and renewable tables retain evidence dates, relevant source entities, percentages and caveats. Audited 2025 subsidiary relationships carried forward into 2026 are labelled in those sources rather than presented as new 2026 equity filings.

Several corrections materially affect attribution. STP is allocated 44% Constellation,40% CPS and 16% Austin, rather than entirely assigned to the corporate operator. Hidalgo uses sworn PUCT shares 78.528%/21.472%, replacing EIA's rounded 79/21 but retaining the raw fraction. Existing Freestone shares remain separate from the new Pin Oak borrower; the old project's Rayburn interest is not transferred by inference. Gregory 2025 is split 43.3% Calpine/56.7% Cheniere, with Cheniere 100% after the 2026 redemption. Cedar Bayou 4 retains NRG's 50% and its separate Quantum/Cogentrix interest; Vistra's Altura acquisition does not imply it acquired Cedar Bayou 4.

Pasadena's legal reported owner is a statutory trust, while Calpine/Constellation holds the financing-lease economic interest. That 746 MW current summer attribution is explicitly marked financing_lease_economic_interest. It is not relabelled as ownership of legal title. Fund managers are described as managers/control groups, not assumed to own every fund dollar. BKV and Banpu Power US are grouped under their disclosed common control parent Banpu, while their separate project stakes remain a different level of the chain.

For renewables, a corporate portfolio mention establishes a relationship, not automatically ultimate control. The NextEra supplement is especially explicit: direct 100% reported portfolio interests may support corporate control, while minority interests and XPLR Opco interests remain separate unless further control evidence exists. DMI/tax-equity and the public partnership/Opco layers are preserved. ENGIE, Enel and Orsted audited consolidation classifications support control relationships even when the economic interest is below 100%. Those percentages remain in the source supplement; they are not mistaken for 100% beneficial ownership. Explicitly documented joint project stakes at Mockingbird, Sparta and Outpost are expanded into separate parent claims.

## Capacity and concentration calculations

OP-only is the default. All-operable sensitivity includes the other statuses in EIA's Operable/Operating inventory, such as OS and SB. Neither status is a guarantee of real-time availability. Physical summer, winter and nameplate MW are retained separately; they are not ELCC, accredited capacity, energy output, revenues, land ownership or firm gas supply.

Thermal excludes wind, solar, conventional hydro, batteries and flywheels from this inventory. The primary thermal denominator uses EIA sector names Electric Utility, IPP Non-CHP and IPP CHP; industrial and commercial sectors are included in a separate sensitivity. CHP output can serve industrial processes, so even the utility/IPP sector filter is not a direct measure of energy offered to the market. Thermal is not synonymous with all dispatchable resources: batteries and hydro can dispatch, and thermal units can fail.

For each generator, attributable MW is physical capability multiplied by the supported plant-interest fraction. These claims are grouped at the documented institutional endpoint. CR3 and CR4 divide the largest mapped groups by the full scope denominator, including unresolved capacity. The output explicitly records whether the unresolved total is smaller than the mapped third/fourth group. An unproven bound is not concealed. A complete ultimate-owner HHI is not reported because treating unresolved entities as independent firms would manufacture a false precision.

Constant-cohort analysis intersects generator keys classified OP at both dates and uses the earlier date's capacity weights for both ownership snapshots. The 2025 cohort keeps 73,022.3 MW fixed. The entire 6,390.822 MW increase assigned to the largest-three group is reconciled to specific transactions, including pre-existing Constellation assets joining Calpine, NRG's LS acquisitions, Vistra's Altura purchase and the Fusco/Gregory exits. It is not an artifact of newly resolving unnamed parents: the earlier LS portfolio is independently documented.

## Evidence tiers and interpretation

The main plant ledger includes primary institutional control claims; partially resolved primary claims; primary portfolio/sponsor relationships; secondary control attributions from Modo matched by normalized asset name, county and technology; fuzzy candidates; and unresolved endpoints. Secondary and fuzzy matches never enter the primary-source concentration series. Modo's publicly downloadable chart is an independent benchmark with a different whole-plant convention, including all STP under Constellation; this is a disclosed methodology difference, not automatically an error. October transactions also postdate itsSeptember 30 cutoff.

Confidence is attached to the claim actually supported. A source can strongly prove the immediate holding chain while leaving fund beneficiaries, voting percentages, or current transaction completion unresolved. Ownership_evidence_tier coverage is counted by whole plant; coverage_summary.csv instead measures the fraction-weighted capacity actually mapped. These two coverage measures must not be conflated.

## Files and reproduction

The principal deliverables are current_plant_ownership_ledger_expanded_boundary.csv, current_plant_ownership_ledger.csv, generators_all_snapshots.csv and ownership_claims_all_snapshots.csv. The first includes the two switchable exceptions. Supporting files include concentration_metrics.csv, parent_capacity_rankings.csv, constant_cohort_concentration.csv, fixed_cohort_top3_change_audit.csv, coverage_summary.csv, current_plant_evidence_tier_coverage.csv and validation_results.json. The crosswalks, primary portfolio supplement and source manifests are part of the audit trail.

Requires Python 3.10+, pandas and openpyxl. From the fleet directory, run these commands in order:

```bash
python prepare_raw.py
python make_parent_crosswalk.py
python build_fleet.py
python match_control_candidates.py
python analyze_concentration.py
python enrich_ledger_evidence.py
python audit_fixed_cohort.py
python validate_outputs.py
python add_boundary_exceptions.py
python write_findings.py
```

prepare_raw.py downloads the exact official EIA workbooks and verifies their hashes before constructing caches. The Modo matching script uses the supplied public-chart factual extract raw/modo_items.json. The primary source mapping CSVs are deliberately curated research inputs; rerunning code reproduces the calculations, not the human documentary research. If a source file changes, review the revision rather than silently accepting a different data vintage.

Validation checks unique generator keys, complete census plant keys, nonnegative and valid claim fractions, fractions summing to one, capacity reconciliation between generators and expanded claims, and source URLs for mapped parents. These checks establish internal integrity, not the truth or completeness of every external disclosure.

## Official source entry points

- EIA860: https://www.eia.gov/electricity/data/eia860/
- EIA860M: https://www.eia.gov/electricity/data/eia860m/
- ERCOT resource reports: https://www.ercot.com/gridinfo/resource
- ERCOT May2026 forecast: https://www.ercot.com/files/docs/2026/05/18/Generation_Resource_Forecast_May2026.xlsx
- PUCT2026 generating-capacity reports, control59141: https://interchange.puc.texas.gov/search/filings/?ControlNumber=59141&UtilityType=A&ItemMatch=Equal&DocumentType=ALL
- Modo independent benchmark: https://modoenergy.com/research/en/ercot-generation-buildout-queue-q3-2026

Every parent mapping's specific primary document is in parent_crosswalk.csv and the supplemental tables. The EIA instruction PDF supplied inside the annual 2025 archive documents the reporting threshold and Schedule 4 requirements.