gridoxide on the 2026 SV-IOP test cases

What gridoxide does for the grid calculations of the 2026 SV-IOP report (Network Code Profiles 2.5, ReliCapGrid), and with what result. The section numbers follow the report's, and only the use cases that need a calculation are here. Import, validation and export of the datasets are cimoxide's, in its own report, which this one builds on. Every statement here is backed by a test in this repository, but for the two marked as not. Each use case's Test line names what backs it; Finding the tests says how they fit together.

Versions

gridoxide0.4.0, commit 766c5a8 (branch cgmes-relicapgrid), Python bindings built with the cgmes feature
Datasetentsoe/relicapgrid cgmes-3.0_ncp-2.5_tc-2.0 @ f09bb34 (2026-09-11), the submodule tests/data/relicapgrid
CIM input and SAR exportcimoxide-model 0.7.0, a dependency of the cgmes feature
Test referencecimoxide 0.6.0 from PyPI, a test-only dependency

How gridoxide was tested

gridoxide did not take part in the event. These are its own runs, afterwards, on a later commit of the dataset, with the report's results known, so they do not stand beside the participants' on equal terms.

It was tested through its Python bindings and in Rust. The Python tests read the data through cimoxide, independently of gridoxide's own use of cimoxide-model, so that what is solved is compared with what another tool decodes. A converged answer is not taken as a result: each is held against something the data says.

CheckWhat gridoxide doesCompared withTest
Power flowPowerFlowModel.from_cgmes and solve on each casethe case's own SvVoltagetest_power_flow
FlowsPowerFlowModel.terminal_flowsthe SvPowerFlow of IGMs that agree with their SVtest_flows_agree_with_the_published_flows
An outagePowerFlowModel.with_out_of_servicethe same outage written into an SSH through cimoxidetest_the_outage_equals_one_written_into_the_ssh
The SARgridoxide.security_analysis, to_sarcimoxide's validation, beside the SAR ReliCapGrid suppliestest_cimoxide_finds_nothing_on_the_results_or_new_on_the_header

Finding the tests

FileWhat it holds
python/tests/test_relicapgrid_power_flow.pyPF_UC1, and BDS_UC1's merge
python/tests/test_relicapgrid_contingencies.pyCO_UC1's execution
python/tests/test_relicapgrid_security_analysis.pySAR_UC1
python/tests/_relicapgrid.pywhat they share: the file groups, reading the data through cimoxide
tests/cgmes_relicapgrid_test.rsthe importer's behaviour on this data, and the analysis's counts, in Rust
src/security_analysis.rsthe analysis, with its unit tests

Problems a power flow finds in the data are pinned in test_relicapgrid_power_flow.KNOWN, each with its cause. A new entry there is a new difference to look at.

git submodule update --init tests/data/relicapgrid
maturin develop --release --features python,cgmes
pip install 'cimoxide>=0.6' pytest
python -m pytest -q python/tests -k relicapgrid
cargo test --release --features cgmes relicapgrid

Overall results

Files or casesResultTest
Power flow17 cases (report Annex 2)13 solve with reactive limits, 6 IGMs agreeing with their SV; 4 stall on their data and converge without limitstest_power_flow
Boundary8 IGMsmerging the boundary changes no voltagetest_the_boundary_changes_nothing
Contingencies325, from 7 CO files312 solve on the CGM without limits, the reference taking the imbalance; 13 of Espheim's stall, unexplainedtest_every_contingency_runs_on_the_cgm
Security analysisthe CGM, 325 contingencies304 solve, the machines sharing the imbalance; 13 base-case and 4,003 post-contingency violationstest_the_counts
SAR1 dataset, 4,071 resultscimoxide reports nothing on its results, and nothing on its header the supplied SAR's does not havetest_cimoxide_finds_nothing_on_the_results_or_new_on_the_header

1. Power Flow on ReliCapGrid

1.1. PF_UC1: Power Flow ReliCapGrid model

Test. test_relicapgrid_power_flow.py: test_power_flow, on case_files, solve and compare, with each problem's cause in KNOWN; test_stalled_cases_converge_without_reactive_limits.

Report. Run power flow on each IGM, then on growing combinations of IGMs, then on the whole CGM (§1.1, Annex 2).

What gridoxide does.

  1. PowerFlowModel.from_cgmes on the case: an IGM's EQ, SSH, TP and SV with the common data; a combination's IGMs together; the CGM as ReliCapGrid's group cgmes-CGM-Jotunheim.
  2. solve(enforce_q_limits=True), Newton-Raphson with reactive limits. The four that stall are solved without them too.
  3. An IGM or the CGM has to agree with its own SvVoltage, within 1% and 0.5° per component. A combination has no state of its own, each IGM's SV describing that IGM alone, so it has to hold one angle reference per component and voltages within 0.8 to 1.2 p.u.

Result. solved, on the terms below. All 17 converge without reactive limits.

CaseWith reactive limitsLargest residual (p.u.)Largest difference from the SVCause
Belgoviaagrees1e-70.03%, 0.03°
Britheimagrees7e-70.14%, 0.21°
DC-Espheim-Svedalaagrees4e-100.06%, 0.12°
Galiaagrees2e-100.08%, 0.27°
Nordheimagrees0noneone bus, the angle reference: nothing to solve
Svedalaagrees2e-90.68%, 0.25°
DC-Nordheim-Galiasolves, differs6e-120.10%, 7°data: converter signs, below
Portheimsolves, out of range6e-12none, at 1.75 p.u.data: the SV has 700 kV on a 400 kV base
Espheimstallsdata: below; converges without limits (2e-7)
Nordheim + DC-Nordheim-Galiasolves6e-12
Svedala + DC-Espheim-Svedalasolves1e-7
Combination 1 (Belgovia + Svedala)solves1e-12
Combination 2 (+ Galia)solves1e-12
All except Espheimsolves, out of range1e-7Portheim, as alone
Espheim + DC-Espheim-Svedalastallsas Espheim; converges without limits (2e-7)
Combination 3 (+ Espheim + Britheim)stallsas Espheim; converges without limits (6e-7)
Jotunheim (CGM)stallsdata: below; converges without limits (3e-7)

Scaling the load in an SSH is noticed (test_notices_a_broken_hypothesis), and files exported by cimoxide solve the same (test_exported_files_solve_the_same).

Notes.

  • Espheim. Its SSH has TannrsCk SM and two ExternalNetworkInjections hold a node at 220 kV, where its SV has them hold 231 kV, within their reactive limits; the SV is otherwise consistent with the EQ and SSH. With the SSH's target the limits cannot all hold, and Newton stalls. With 231 kV it solves with limits, and without them to the SV's voltages (test_espheim_solves_with_the_target_its_sv_holds).
  • The CGM. Jotunheim's SV holds 38 of its 72 voltage controls off their SSH_2 targets, up to 1.32 times (ERIKSHAMN_G1: 20.9 kV, 27.7 in the SV), so it is no solution of that SSH_2, and with those targets the limits cannot all hold. It converges without them, at 0.77 p.u. at the lowest and up to 24% from that SV.
  • HVDC signs. A DC-Nordheim-Galia converter's targetPpcc is +70 where its ACDCConverter.p, and the SV, are -70.01; Britheim writes both with one sign. This is the convention the report's §4.1 asks to settle.
  • cimoxide checks that each IGM carries the data a power flow needs: see its PF_UC1.

Against the participants. Not a test: counted by hand from the report's Annex 2. Its "Yes" is not defined, and does not say whether reactive limits were on.

ToolSolvedReported as failedNot reported
VeraGrid1700
gridoxide, no reactive limits1700
gridoxide, with reactive limits1340
PowerFactory1232
ODMS1025
CIMdesk737

In the report only VeraGrid solves Espheim.

2. Boundary, Common and Reference data

2.9.1. BDS_UC1: Boundary Data Set Importing

Test. test_relicapgrid_power_flow.py: test_the_boundary_changes_nothing.

Report. Load the boundary dataset, merge it with the IGMs, check that matching nodes and terminals connect, with no duplication and no topology gap at the borders.

What gridoxide does. Solves each of the eight IGMs that solve with the six boundary files added.

Result. Merging the boundary changes no bus and no voltage.

Notes. The checks the report asks for are cimoxide's: see its BDS_UC1.

3. Contingencies

3.8.1. CO_UC1: Importing, executing, updating and exporting

Test. test_relicapgrid_contingencies.py. pytest python/tests/test_relicapgrid_contingencies.py.

Report. Import the contingencies, resolve them against the network model, execute them, update them, export the updated CO dataset in compliance with the profile. Execution does not permanently modify the base network state.

What gridoxide does. The execution. PowerFlowModel.with_out_of_service solves the network without each contingency's equipment, all outOfService, and leaves the base model as it was.

  1. For Table 3's equipment, of eight classes, test_an_outage_removes_its_equipment_and_nothing_else shows what each takes away: a branch only the admittances between its ends, a shunt only its bus's, a load or machine only its scheduled power, an HVDC converter or line the link's power at both ends.
  2. test_the_outage_equals_one_written_into_the_ssh reaches each by another road, Equipment.inService set in the SSH through cimoxide, with the same voltages.
  3. test_execution_leaves_the_base_network_alone checks the report's expected outcome.
  4. Every contingency of the CO files runs, 325 from 7 files.

Result. executed.

NetworkReactive limitsSolveStallTest
Belgovia, Britheim, DC-Espheim-Svedala, Svedala (108)yesalltest_every_contingency_of_an_igm_solves
Galia (12)yes11as above; CO_HVDC's equipment is another IGM's
Belgovia + Svedala, for Svedala-Belgovia_CO.xml (4)yesalltest_the_contingencies_between_two_igms_solve_on_both
Espheim (201)no, as its base stalls with them18813test_espheims_contingencies
Espheim at the 231 kV its SV holds (201)yes17229test_espheims_contingencies_with_the_voltage_its_sv_holds
Espheim at the 231 kV its SV holds (201)no19110as above
The CGM (325), the reference taking the imbalanceno, as its base stalls with them31213, Espheim'stest_every_contingency_runs_on_the_cgm

Notes.

  • The 13 trip heavily loaded lines of a base already at 0.77 p.u. Correcting Espheim's target removes only three of them, so that contradiction does not explain them. Newton not converging does not prove there is no solution: these are not explained.
  • Table 3 names two ties the wrong way round, and its N-2 differs from the data. The files say which equipment goes out, and they are what runs: see cimoxide's CO_UC1, which also has the import, update and export.
  • Which of these violate a limit is SAR_UC1's question. Remedial actions are not applied.

9. Security Analysis Results (SAR)

9.1.1. SAR_UC1

Test. test_relicapgrid_security_analysis.py; security_analysis_of_the_cgm in Rust.

Report. Export a SAR: the tool produces security analysis results and exports them; the export is checked against the profile, and for consistency.

What gridoxide does. gridoxide.security_analysis solves the CGM's base case and each contingency, without reactive limits as its base stalls with them, and holds each terminal's current against its limits; to_sar writes the violations, and the base case of each terminal a contingency violates. See Security Analysis and the SAR.

Result. produced, and validated by cimoxide.

Results
Contingencies solved / not converging304 / 21, Espheim's; the profile has no place for these, so they are not in the SAR
Base-case violations of PATL13
Contingency results4,003, on 303 contingencies
Of those, on a limit the base case does not violate214, on 81 contingencies and 55 terminals
Base cases of other terminals, given for comparison55, for 4,071 results in the dataset
Machines shared past their limitsnone, in the base case or any contingency solved

The largest of the 214 are 68-81 at 192% and 65-68 at 178% after CO190_Aaa_1_, CT11_T401 at 167% after CO31_CT112_T401, and GA-TR2_1 at 162% after CO3.

How far the numbers can be taken.

  • The flows. test_flows_agree_with_the_published_flows compares them with SvPowerFlow on the four IGMs that agree with their SV: a median of at most 0.5 MVA, at most 12 MVA. On the CGM they are up to 746 MVA off its SV, which is no solution of its SSH_2.

  • The state. The CGM's base is what its SSH_2 gives without reactive limits, as low as 0.75 p.u.

  • The imbalance. SSH_2 schedules some 530 to 650 MW more than the CGM takes, and who takes that changes the answer (test_the_answer_depends_on_who_takes_the_imbalance):

    sharingSolved / stalledBase-case violationsBO-TR2_3, behind the reference: contingencies, worst
    capability, by maxOperatingP (default)304 / 2113303, 127%
    reference, the CGM's one reference312 / 1317311, 596%
    equal, the same for every generator bus287 / 3821286, 127%

    None is a dispatch. reference is how the CGM's SV was made, which therefore fits it best, and cannot validate the others: there only two machines leave their schedule, and BO-G1, scheduled at 52.8 MW, runs at -481.8 MW (test_the_cgms_state_leaves_the_imbalance_on_one_machine). normalPF is scaled per IGM, so Belgovia's 2 units (700 MW) would take the share of Svedala's 38 (16,750 MW) (test_normal_pf_is_scaled_per_igm).

  • No outside check. The supplied SAR's one base case agrees with every choice: a dead-end stub under 2 MW, set by local data, so it says little for the solver (test_the_supplied_sars_base_case_cannot_tell_them_apart). Its one contingency result is on the line that contingency trips, so there is nothing to compare.

  • The limits. Only PATL. All 578 TATLs are 500 A and none is above its PATL, so they cannot be what they say (test_every_tatl_in_the_data_is_500_a_and_none_above_its_patl). Remedial actions are not applied.

Validation. test_cimoxide_finds_nothing_on_the_results_or_new_on_the_header: in the model with the CO files, cimoxide reports nothing on a result, and nothing on the header the supplied SAR's header does not have. A broken SAR is reported (test_a_result_without_its_value_is_reported).

Notes.

  • Not a test: triplets 0.2.0, ReliCapGrid's CI tool, was run once on this output in place of the supplied SAR, in cimoxide's harness, and reported nothing on its results and nothing new on its header. It would bring ReliCapGrid's whole validation into this repository.
  • What cimoxide does with the SAR ReliCapGrid supplies is in its SAR_UC1.

Annex 1: Test Use Cases Results

Only the use cases that need a calculation. pytest -k relicapgrid runs them.

Use casegridoxide 0.4.0Test
PF_UC1solved: 13 of 17 cases with reactive limits, all withouttest_relicapgrid_power_flow.py
BDS_UC1simulate: the boundary changes nothingtest_the_boundary_changes_nothing
CO_UC1execute: 312 of 325 on the CGMtest_relicapgrid_contingencies.py
SAR_UC1simulate, exporttest_relicapgrid_security_analysis.py

The rest of the report's Annex 1 is cimoxide's.

Annex 2: Issues found in the data

WhereIssue
ReliCapGrid dataEspheim's SSH holds a node at 220 kV through three controls; its SV holds it at 231 kV.
ReliCapGrid dataJotunheim's SV holds 38 of its 72 voltage controls off their SSH_2 targets, up to 1.32 times.
ReliCapGrid dataThe CGM's SSH_2 schedules some 530 to 650 MW too much; its SV leaves that on one machine, BO-G1 at -481.8 MW.
ReliCapGrid dataA DC-Nordheim-Galia converter's targetPpcc and ACDCConverter.p have opposite signs; Britheim's have one.
ReliCapGrid dataPortheim's SV has every node at 700 kV on a 400 kV base (1.75 p.u.).
ReliCapGrid dataAll 578 TATLs are 500 A and none is above its PATL; 558 are below it.
ReliCapGrid dataGeneratingUnit.normalPF is scaled per IGM (100 for most, 200 for Espheim, 1 for Britheim), so it cannot share across a CGM.

What is found in the files themselves, such as the supplied SAR's value and its result on a tripped line, or the report's Table 3 differing from the data, is in cimoxide's Annex 2.