Two preregistered datasets on production-type labelling in the ENTSO-E Transparency Platform: multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency

arXiv:2609.16152 · physics.soc-ph · Submitted 2026-09-14 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Two preregistered datasets on production-type labelling in the ENTSO-E Transparency Platform".

Mira: Here's the gist of the paper. The paper investigates two preregistered datasets measuring production-type labelling in the ENTSO-E Transparency Platform to assess multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency Two preregistered datasets on production-type labelling in the ENTSO-E Transparency Platform: multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency The ENTSO-E Transparency Platform is the single mandated publication point for European electricity market data Three of its data items carry a production-type label, including Installed Capacity per Production Unit (14.1.B), Actual Generation Output per Generation Unit (16.1.A), and Aggregated Generation per Production Type (16.1.B) The production-type axis is weak because Article 5 requires TSOs to develop a manual specifying appropriate classification of production types The two datasets described here measure what that classification does in practice across four axes Dataset Structure and Axes Each dataset has its own preregistration, locked criteria, and checksum, carrying two axes: psrtype-samfyring for generation data and kapregister for the capacity register The psrtype-samfyring axis includes simultaneous multi-fuel reporting (Axis 1) and cross-product consistency (Axis 4) The kapregister axis includes type change between register vintages (Axis 2) and persistent misclassification without type change (Axis 3) Axis 4 does not sit inside the locked preregistrations Measurement Criteria Details The measurement criteria are fixed before adjudication rather than before measurement, and additions are recorded in dated supplement files For the multi-fuel axis, a hit is defined as two TimeSeries in the same GL MarketDocument with different psrType Axis 4 requires three criteria to be met for cross-product consistency, including persistence and unique attribution Axis 3 is added because the locked criteria explicitly exclude correctness, measuring change rather than correctness Prior Art and Contribution Neither the observation nor the method is new, as prior art exists regarding stale labels in ENTSO-E The contribution here is the application and measurement of this known observation Prior work has documented discrepancies such as coal units being converted to biomass but reported generation remaining low This study adds measurement through dated primary-source evidence per zone and criteria fixed before the data were read Technical Validation and Controls Technical validation checks include reproducing an independent published method for annual emission factors Control categories such as nuclear return 0.99–1.00× and wind 0.94–0.98× against national sources Mass conservation is conserved even when the split between coal and biomass is not, with ratios around 1.00–1.12× across twelve zone-years The sampling design was fixed before measurement, establishing that twelve days recover what a full year recovers in that zone Data Records and Provenance All records are archived on Zenodo under the Apache-2.0 licence for code and CC BY 4.0 for data and documentation Version DOIs exist for both Record group A (psrtype-samfyring) and Record group B (kapregister) Raw XML is never committed but is re-fetched by retrieval scripts to rebuild it The paper cites ten DOIs in the Appendix, including those for the specific version DOIs and concept DOIs Concept DOI always resolves to the newest version Concept 10.5281/zenodo.22733391 resolves to the newest version The paper cites various external sources including Commission Regulation (EU) No 543/2013 and relevant GitHub issue threads Usage Notes and Findings Coverage limits show that four zones publish no A73 in any of the three years probed, returning Acknowledgement MarketDocument with code 999 Great Britain published A73 through 31 May 2021 and not after, with uneven coverage inside that period The platform separates fuels precisely when each unit burns one fuel, identifying the gap as fuel mixed inside a single unit Users combining A73 and A75 should expect A73 ≤ A75 and should not treat a shortfall as an error The design is strong against a persistent phenomenon and weak against a rare one, which was chosen against the persistent one The regulatory failure is attributed to the shortfall sitting with one data provider rather than systemic deficiency The Spanish A75 nuclear column, when expanded correctly, shows a ratio of 1.000 tracking each other hour by hour within 0.1 % The Solina illustration no longer reproduces after the A03 correction, returning zero false attributions in the Polish B12 column The paper concludes that compliance is not systemically deficient on the register axis The final conclusion is that the regulatory failure presupposed hits in many countries Data Records Summary Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself Key Takeaways on Data Integrity The paper emphasizes that a unit carrying one type in one period and another later is a register reclassification propagating forward A zone without documented co-firing is not counted as zero, guarding against the central fallacy of data absence The operational profile is a supporting test, never proof, and the source decides whether a label should be checked against an independent source The discrepancy that prompted Axis 4 was recorded in a dated supplement outside the locked hit definition The paper concludes that users comparing either series should not treat a shortfall as an error but rather as the threshold The study validates that the design is strong against a persistent phenomenon and weak against a rare one, which was chosen against the persistent one The paper confirms that mass conservation is conserved even when the split between coal and biomass is not, validating duration weighting and column selection The overall finding is that compliance with Regulation 543/2013 does not mandate publication without an obligation to revise The paper provides empirical support for the need for better classification methods by quantifying the effect of existing classification issues The paper is a measurement of what that classification does in practice across four axes Final Synthesis The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself Data Records Summary Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself Data Records Summary Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself Final Synthesis The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.

Paper summary: Kai: So, to wrap up this paper, "Two preregistered datasets on production-type labelling in the ENTSO-E Transparency Platform: multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency," the authors are providing a measurement of what that classification does in practice across four axes.

Mira: They’re applying measurement to a known observation about stale labels by fixing their criteria before they measure the data and using dated primary-source evidence.

Lev: It really highlights that you have to be careful about what you're measuring; the operational profile is a supporting test, not proof, and the source decides whether a label should be checked against an independent source.

Kai: They’ve shown that mass conservation holds even when the coal and biomass split isn't perfect, which validates duration weighting and column selection in their analysis.

Mira: The overall finding is that compliance with Regulation five hundred forty-three/two thousand thirteen doesn't mandate publication without an obligation to revise, which provides empirical support for the need for better classification methods by quantifying the effect of existing issues <ref:2609.16152#pg1>.

Lev: So, this isn't about a systemic deficiency in the platform itself; it’s about where regulatory failure is happening in practice across different countries.

Kai: It means users comparing either series should not treat a shortfall as an error but rather as the threshold, and they shouldn't assume that because of one data provider, there's a universal problem.

Conclusion: Kai: So, we've looked at how these two specific datasets measure production type labeling in that European electricity platform for multi-fuel reporting and consistency issues across register vintages and misclassification.

Mira: It really boils down to testing how well the existing classification system works in practice by fixing the rules before you even look at the numbers.

Kai: The authors, they’re applying this measurement framework to known discrepancies, like coal units being called biomass but still reporting low generation.

Mira: That's what I mean. It adds real, dated primary source evidence per zone and fixes the criteria before adjudication so it measures change rather than just correctness on that persistent misclassification axis.

Lev: From a hardware side, this tells us if the classification system is actually breaking down or if it's just a slow propagation of errors where one type stays one type for too long.

Kai: The conclusion they draw is that the platform itself isn't systemically deficient in how it classifies things; the failure lies with individual data providers in many countries.

Mira: That means users shouldn't treat a shortfall as an error but rather as a threshold, especially when comparing those two different series of data they have.

Kai: It’s about understanding what that classification actually does on the ground, not just trusting the system blindly.

Mira: So it sets up a question for us now: what does this practical measurement tell us about how we should actually design better classification methods?

EcoDeco AS

physics.soc-ph

Submitted: 2026-09-14

Updated: 2026-11-06

Comments: 11 pages. Datasets: https://doi.org/10.5281/zenodo.22726238 and https://doi.org/10.5281/zenodo.22722721. Supplementary working documents: https://doi.org/10.5281/zenodo.22734405. v2 (6 Nov 2026): adds a cross-reference to doi:10.5281/zenodo.24014891, an AI-use statement and a note on A75 and CC BY 4.0; corrects the GB 2021 file-glob description (30 content files, not 30 days); no figures changed

Code: https://github.com/electricitymaps/electricitymapscontrib

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 89/100

The gist: The paper investigates two preregistered datasets measuring production-type labelling in the ENTSO-E Transparency Platform to assess multi-fuel reporting, register vintages…

Key concepts

psrtype-samfyring
This axis measures simultaneous multi-fuel reporting (Axis 1) and cross-product consistency (Axis 4). It tracks whether different fuel types are reported together in the same document and how consistent that reporting is across different products, which is crucial for understanding how the system handles mixed fuels.
kapregister
This axis focuses on type changes between register vintages (Axis 2) and persistent misclassification without a vintage change (Axis 3). It tests whether a unit's classification changes when looking at older data or if it remains incorrectly labeled over time, helping identify classification drift.
persistent misclassification
This measurement specifically looks for errors where a unit is consistently labeled incorrectly without changing its type. The study uses this to measure change rather than correctness, focusing on the stability of the classification error across different data points and time periods.
regulatory failure vs. systemic deficiency
The research concludes that misclassification is caused by regulatory failures in many countries, not a fundamental flaw in the ENTSO-E platform itself. The platform design is robust against persistent problems, meaning the issue lies with how data providers or regulations enforce labeling standards.

Terminology

Summary

The paper investigates two preregistered datasets measuring production-type labelling in the ENTSO-E Transparency Platform to assess multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency

Two preregistered datasets on production-type labelling in the ENTSO-E Transparency Platform: multi-fuel reporting, register vintages, persistent misclassification, and cross-product consistency

The ENTSO-E Transparency Platform is the single mandated publication point for European electricity market data Three of its data items carry a production-type label, including Installed Capacity per Production Unit (14.1.B), Actual Generation Output per Generation Unit (16. p. 0

Where the paper says it

One row per production unit of 100 MW or more, with an EIC code, a name, a capacity in MW and one Production Type. A (A73), Actual Generation Output per Generation Unit.

1.A), and Aggregated Generation per Production Type (16. p. 0
Where the paper says it

Hourly output per generation unit of 100 MW or more, delivered as GL_MarketDocument with one psrType per TimeSeries. B (A75), Aggregated Generation per Production Type.

1.B) The production-type axis is weak because Article 5 requires TSOs to develop a manual specifying appropriate classification of production types The two datasets described here measure what that classification does in practice across four axes

Dataset Structure and Axes

Each dataset has its own preregistration, locked criteria, and checksum, carrying two axes: psrtype-samfyring for generation data and kapregister for the capacity register The psrtype-samfyring axis includes simultaneous multi-fuel reporting (Axis 1) and cross-product consistency (Axis 4) The kapregister axis includes type change between register vintages (Axis 2) and persistent misclassification without type change (Axis 3) Axis 4 does not sit inside the locked preregistrations

Measurement Criteria Details

The measurement criteria are fixed before adjudication rather than before measurement, and additions are recorded in dated supplement files For the multi-fuel axis, a hit is defined as two TimeSeries in the same GL MarketDocument with different psrType Axis 4 requires three criteria to be met for cross-product consistency, including persistence and unique attribution Axis 3 is added because the locked criteria explicitly exclude correctness, measuring change rather than correctness

Prior Art and Contribution

Neither the observation nor the method is new, as prior art exists regarding stale labels in ENTSO-E The contribution here is the application and measurement of this known observation Prior work has documented discrepancies such as coal units being converted to biomass but reported generation remaining low This study adds measurement through dated primary-source evidence per zone and criteria fixed before the data were read

Technical Validation and Controls

Technical validation checks include reproducing an independent published method for annual emission factors Control categories such as nuclear return 0.99–1.00× and wind 0. p. 5

Where the paper says it

Nuclear returns 0. 00× and wind 0.

94–0.98× against national sources Mass conservation is conserved even when the split between coal and biomass is not, with ratios around 1. p. 5
Where the paper says it

Mass conservation The sum of the coal and biomass columns is conserved even when the split between them is not. Against Energinet’s own declaration data the ratio is 1.

00–1.12× across twelve zone-years The sampling design was fixed before measurement, establishing that twelve days recover what a full year recovers in that zone

Data Records and Provenance

All records are archived on Zenodo under the Apache-2.0 licence for code and CC BY 4. p. 4

Where the paper says it

Data Records All records are archived on Zenodo under the Apache-2. 0 licence for code and CC BY 4.

0 for data and documentation Version DOIs exist for both Record group A (psrtype-samfyring) and Record group B (kapregister) Raw XML is never committed but is re-fetched by retrieval scripts to rebuild it The paper cites ten DOIs in the Appendix, including those for the specific version DOIs and concept DOIs Concept DOI always resolves to the newest version Concept 10. p. 4
Where the paper says it

Ten DOIs exist: eight version DOIs and two concept DOIs. A concept DOI always resolves to the newest version; a version DOI never changes.

5281/zenodo. p. 4
Where the paper says it

Record group A — psrtype-samfyring (axes 1 and 4) Version DOI Content v1. 5281/zenodo.

22733391 resolves to the newest version The paper cites various external sources including Commission Regulation (EU) No 543/2013 and relevant GitHub issue threads

Usage Notes and Findings

Coverage limits show that four zones publish no A73 in any of the three years probed, returning Acknowledgement MarketDocument with code 999 Great Britain published A73 through 31 May 2021 and not after, with uneven coverage inside that period The platform separates fuels precisely when each unit burns one fuel, identifying the gap as fuel mixed inside a single unit Users combining A73 and A75 should expect A73 ≤ A75 and should not treat a shortfall as an error The design is strong against a persistent phenomenon and weak against a rare one, which was chosen against the persistent one The regulatory failure is attributed to the shortfall sitting with one data provider rather than systemic deficiency The Spanish A75 nuclear column, when expanded correctly, shows a ratio of 1.000 tracking each other hour by hour within 0. p. 7

Where the paper says it

6 GWh, ratio 1. 000, tracking each other hour by hour within 0.

1 % The Solina illustration no longer reproduces after the A03 correction, returning zero false attributions in the Polish B12 column The paper concludes that compliance is not systemically deficient on the register axis The final conclusion is that the regulatory failure presupposed hits in many countries

Data Records Summary

Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself

Key Takeaways on Data Integrity

The paper emphasizes that a unit carrying one type in one period and another later is a register reclassification propagating forward A zone without documented co-firing is not counted as zero, guarding against the central fallacy of data absence The operational profile is a supporting test, never proof, and the source decides whether a label should be checked against an independent source The discrepancy that prompted Axis 4 was recorded in a dated supplement outside the locked hit definition The paper concludes that users comparing either series should not treat a shortfall as an error but rather as the threshold

Conclusion on Data Quality

The study validates that the design is strong against a persistent phenomenon and weak against a rare one, which was chosen against the persistent one The paper confirms that mass conservation is conserved even when the split between coal and biomass is not, validating duration weighting and column selection The overall finding is that compliance with Regulation 543/2013 does not mandate publication without an obligation to revise The paper provides empirical support for the need for better classification methods by quantifying the effect of existing classification issues The paper is a measurement of what that classification does in practice across four axes

Final Synthesis

The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself

Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself

Record group A tracks psrtype-samfyring, with miss rates of 96.8 / 79.3 / 35.7 / 3.17 % Record group B tracks kapregister, showing that the K requirement exists to prevent undetermined classification The data records show that the A03 correction moved what the A03 correction moved and the Spanish retraction The paper notes that parallel series showed all direction splits with zero psrType differences The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.00× It confirms that the platform separates fuels precisely when each unit burns one fuel The paper is a measurement of what that classification does in practice across four axes The final finding is that regulatory failure as the cause of misclassification is not systemically deficient in the platform itself

The paper provides measurement to a known observation by fixing the criteria before the data were read and using dated primary-source evidence The paper's contribution is the application and measurement of this known observation It adds measurement to a known observation through fixed criteria and public records of what did not survive The paper notes that the control categories behave as expected, with nuclear returning 0.99–1.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved classification accuracy for production types by applying measurement to known observations, specifically addressing persistent misclassification without type change by requiring a link to an independent, dated primary source for any mislabeling.

  2. Enhanced detection of simultaneous multi-fuel reporting by identifying a hit is two TimeSeries in the same GL MarketDocument with the same MktPSRType/PowerSystemResources/mRID and different psrType.

  3. Better handling of capacity register changes by defining clear criteria for revision, such as revised C ≥ 1 when units have a documented conversion evidenced by an independent, dated source stored on disk and cited by URL.

  4. Increased robustness against data gaps in time series through the application of the A03 step curve, ensuring that A03 curve Type A03 means a missing position inside a Period does not mean missing data; it means the previous value holds until the next stated position.

  5. More accurate reconciliation of aggregated versus per-unit generation data by implementing cross-product consistency checks, specifically looking for cases where the A73 sum exceeds the A75 column over the same hours.

  6. Automated identification of stale labels by distinguishing between type change and persistent misclassification, using Condition 4 is the whole distinction from unrevised: an error that postdates a conversion is a stale label and belongs on axis 2.

Abstract

The ENTSO-E Transparency Platform is the single mandated publication point for European electricity market data, and three of its data items carry a production-type label. Commission Regulation (EU) No 543/2013 Article 16(1)(a) requires actual generation output per generation unit but does not mention production type; the label on that item is ENTSO-E's own addition through the schema, and Article 5(d) delegates the classification of production types to a Manual of Procedures. This data descriptor documents two preregistered datasets that measure what that classification does in practice, along four axes: simultaneous multi-fuel reporting, type change between capacity-register vintages, persistent misclassification without type change, and cross-product consistency between the per-unit and the aggregated series. Criteria were locked and checksummed before measurement, thresholds are derived from measured noise rather than chosen, and what did not survive is recorded: six claims are withdrawn, three of them from published versions of the datasets themselves. Across 2191 documents covering 199 generation units in three bidding zones where biomass co-firing is documented in a dated primary source, no unit carries two production types in the same document. One practical warning applies to every user of the hourly series: all of it carries curveType A03, a step curve in which a missing position means the previous value holds rather than that data are absent. A parser that sums only the points listed under-reports energy unevenly, by 22.6 percent for one Finnish unit in 2021.

Sources

Related papers