Case · Software whose output gets acted on · Public

A power-flow solver reported convergence for transformer types it does not support

pandapower, an open-source power-system analysis library, and its unbalanced three-phase power flow.

The system said: converged=True.
The evidence showed: 994,686 per unit on a 0.4 kV bus. Four unsupported transformer types skip the gate that rejects the other ten.

§ 1

The claim tested

Transformer vector groups the three-phase solver does not support are rejected rather than solved.

§ 2

What Omega did

Classified all 18 two-winding vector-group strings found in the source, the documentation and the shipped standard types as documented-supported or unsupported before running anything, and wrote a predicted outcome per cell into a pre-registration hashed before execution.
Ran each string once, in its own process, on the network from a closed 2020 issue where the same class of input had failed to converge.
Applied five authored sanity bands to any result that reported convergence, and diffed the source tree against the pinned archive afterwards.

§ 3

What was found

Ten of fourteen unsupported groups raised the intended error. Four (Dy, Yy, Yd, Dd) returned converged=True with no warning and results identical to the last digit: about 994,686 per unit on the 0.4 kV buses, 714% transformer loading, and a 10.9 MW balance residual against a 0.1 MW load.
The cause is visible in the source: zero-sequence modelling is skipped for exactly those four strings before the gate that rejects the others, so they never reach it.
One group, Yyn, is documented as supported in two places and rejected by the code. Reported as an inconsistency, not adjudicated.
All 18 cells matched their pre-registered prediction.

§ 4

Verdict

CONTRADICTED: Unsupported vector groups are rejected. 4 of 14 are solved silently. The four map to one model, so this is one observation, not four.
INCONCLUSIVE: Supported groups give a correct three-phase result. Plausible under five authored bands on one network. No independent oracle was run, so plausible is the ceiling.

§ 5

What this does not establish

One network, one load, zero phase shift, one commit. With a non-zero phase shift the four strings might no longer collapse to one model; not run. Whether the skip is intended for another calculation path is upstream's call.

§ 6

Upstream

pandapower OPEN ISSUE issueWhether transformer vector groups the three-phase power flow does not support are rejected rather than solved. Issue #3122, 6 September 2026. Open issue. No maintainer response as of 20 September 2026.

Released means the fix is in a tagged release. Merged means it is on the default branch but not verified in a tagged release. Open pull request means an Omega-authored change awaits upstream action. Open issue means a report remains open without a merged fix. Maintainer acceptance is review of one change, not endorsement of Omega.

§ 7

The records

Every field of each finding this case rests on, as it appears on the ledger.

OBSERVED pandapower-unsupported-vector-groups-return-convergedverify · coverage illusion A three-phase power-flow solver reported convergence for four transformer types it does not support, returning bus voltages near a million per unit with no warning. Ten other unsupported types were correctly rejected.

Claim: In pandapower develop at fd7346f, runpp_3ph on the network from closed issue #1041 returned converged=True with no warning for the vector groups Dy, Yy, Yd and Dd, with identical non-physical results (about 994,686 per unit on the 0.4 kV buses, 714% transformer loading, a 10.9 MW power-balance residual against a 0.1 MW load), because zero-sequence modelling is skipped for those four strings before the gate that rejects the other ten unsupported groups. All 18 cells matched their pre-registered prediction.

Status: OBSERVED Witnessed in one instance. No claim about how often.

Question: Passed is not correct

Subject: pandapower, an open-source power-system analysis library (third-party system); the classifier and the sanity bands are Omega's. third-party subject

Frame: The #1041 network verbatim; 18 two-winding vector-group strings from the source, the docs and the shipped standard types; shift_degree 0; each cell in its own process.

Method: Each string classified as documented-supported or unsupported before the run, with a predicted outcome per cell frozen in a pre-registration; the run compared against the prediction; the source tree diffed against the pinned archive after the run.

Oracle: Five authored sanity bands on converged results: finite values, low-voltage magnitudes in 0.9 to 1.1 per unit, external-grid power in band, power balance under 1e-3 MW, transformer loading under 100%.

Negative control: Present The supported Dyn group as a positive control and the known-silent Dy group as a negative control both reproduced earlier runs the same day to the digit; no environment drift.

Denominator: 18 of 18 cells matched their frozen prediction; 4 of 14 unsupported groups returned converged=True; 10 of 14 raised; 0 failed to converge.

Preregistration: PREREG-VECTOR-GROUP-BOUNDARY, hashed before execution and unchanged after; no deviations.

Limitation: One network, one load, zero phase shift, one commit. The four silent strings map to one model, so they are one observation, not four. No independent oracle was used for the supported groups, so plausible is the ceiling of what is said about them. Whether the zero-sequence skip is intended for another calculation path is upstream's call. The issue is open and unanswered; nothing has been fixed.

Source: artifact not public

Reproduce: not reproducible from a public clone

Independent reproduction: None known.

Upstream: REPORTED UPSTREAM Issue #3122, opened 6 September 2026; no maintainer response as of 20 September 2026. Link

This finding on its own page

All cases Bring us a problem