Pith. sign in

REVIEW 4 major objections 4 minor 21 references

Enabling Scalable Topology Inference in Distribution Systems via Constrained Multi-Source Inference

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A distribution-grid topology solver treats utilities' imperfect connectivity records as a noisy prior, corrects only suspicious assignments in local neighborhoods, and reports over 95% reconstruction accuracy on 8,000+ smart meters.

desk verdict A sensible engineering idea for localized topology correction, but the headline 95% accuracy claim is nowhere backed by numbers in the paper—no defined ground truth, no accuracy table, no baseline comparison. read the letter →

arxiv 2607.20480 v1 pith:VTCEBEA6 submitted 2026-05-30 cs.AI

Haoran Li , Lihao Mai , Muhao Guo , Jiaqi Wu , Yang Weng This is my paper
classification cs.AI
keywords topologyinferencedistributionsystemssmartmeterdataAMIconstrainedoptimizationlocalizedcorrectionreliabilityestimationfalsificationtesting
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to prove that distribution-grid topology—the map of which smart meter hangs off which transformer—can be recovered reliably from the messy, partially outdated records utilities keep, by treating those records as a noisy starting point rather than as truth to be rebuilt from scratch. It argues that most recorded connections are actually correct, and that the wrong ones come in small localized patches, so a solver only needs to detect suspicious nodes and rewire them within a small geographic neighborhood under physical constraints (transformer capacity, voltage range). On three operational feeders with over 8,000 smart meters, the paper reports over 95% reconstruction accuracy and runtime that scales roughly linearly with feeder size, plus a confidence score per connection so field crews know where to verify. If true, utilities would no longer need to re-derive the whole network from measurements, but could automatically correct record errors at scale and focus manual checks on the few ambiguous spots.

What carries the argument

The load-bearing mechanism is the detect-suspect-then-repair loop built on the 'errors are localized' premise. Suspicious nodes are flagged by a distance ratio (distance to assigned transformer vs. closest transformer, threshold τ) and by persistently low voltage correlation with their assigned transformer's other meters. Reassignment is restricted to a candidate set of K geographically nearest transformers, turning an exponential global assignment problem into roughly linear cost; the paper formalizes this as Lemma 1, giving O(|O|K) per iteration where |O|=γN with γ≪1. The final confidence score fuses a Davies–Bouldin cluster-separation ratio and a correlation-support ratio via CL = 0.7·Sco

What would settle it

Apply the solver to a feeder where a substantial fraction of connections (say 40–50%) are deliberately corrupted in a spatially spread, uniform way rather than in a few clusters, with the true topology independently known; the paper's claimed >95% accuracy and near-linear runtime advantage over global inference should fail if the localized-error premise is what carries the result.

Watch

Extended reading notes

Core claim

The central claim is that topology identification should be cast as constrained inference over a noisy prior: minimize a weighted objective of electrical inconsistency, geographic implausibility, and deviation from the utility's base record, subject to physical feasibility constraints, and solve it by local repair rather than global search. The paper's solver detects inconsistent assignments using a distance ratio and weak voltage correlation with neighbors, restricts reconnection candidates to the K nearest transformers, assigns nodes by clustering in a spatial–voltage feature space, and only accepts updates that keep transformer loading and voltage within limits. It then assigns each resul

Load-bearing premise

The method depends on the premise that most record errors are few, small, and geographically clustered, and that the detection heuristics (distance ratio and voltage-correlation weakness) catch exactly those wrong nodes; if errors are widespread or the detector misses the right nodes, local repair cannot restore topology and the linear-scaling guarantee collapses.

Editorial extensions

If this is right

  • Utilities can run periodic topology refreshes using only suspicious-node detection plus local rewiring, without full-network reconstruction, keeping runtime within operational refresh intervals.
  • Ambiguous connections come with a falsification-based confidence score, so field verification can be targeted at low-confidence assignments instead of blanket inspections.
  • In dense urban feeders where voltage correlations between different transformers exceed 0.95, combining spatial and physical constraints recovers assignments that correlation-only clustering cannot separate.
  • The approach is robust to moderate parameter choices: assignment results are stable across small and medium candidate-set sizes K.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the locality premise does not hold—say, a feeder's records are broadly corrupted by a systematic data-migration error—the suspicious detector could flag too many nodes or miss the wrong ones, and the O(NK) guarantee and the high accuracy would both degrade; the paper does not report performance under such non-local corruption.
  • The reliability score could be read as an implicit prioritization tool for verification budgets: a natural extension is to feed low-confidence assignments back into a data-collection plan (e.g., where to add sensors or field checks) rather than merely flagging them.
  • The same detect-repair-feasibility pattern likely transfers to phase identification or medium-voltage switching topology, since those problems share the structure of noisy records plus physical constraints.
  • The 95% figure's ground truth is not specified in the paper; an independent validation against a known-correct connectivity map on a feeder with a stated, controlled error rate would sharpen the claim.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a constrained multi-source inference framework for distribution-system topology refinement. It takes a utility-provided base topology as a noisy prior, detects suspicious node-to-transformer assignments via spatial and voltage-correlation criteria, performs localized reconnection among the K nearest transformers, enforces physical/operational feasibility constraints, and outputs a reliability score for each inferred connection. The authors claim over 95% topology reconstruction accuracy and significantly reduced computational effort relative to global inference, based on operational data from three feeders with more than 8,000 AMI meters. The manuscript also argues that correlation-only methods are ambiguous in dense urban feeders and that combining electrical, spatial, and operational constraints enables scalable, reliable topology recovery.

Significance. If the claimed results were properly substantiated, this work would be practically significant: it addresses a real utility problem, uses heterogeneous operational data, and proposes an actionable reliability metric for prioritizing field verification. The localized-refinement idea is reasonable and the paper is written around a plausible engineering workflow. However, the manuscript as submitted provides almost no quantitative evidence for its central claims: there is no definition of ground truth, no accuracy/precision/recall table, no baseline comparison with actual numbers, and no error analysis. The main contribution is therefore currently unverifiable, and the scalability result rests on an unexamined locality assumption. The paper would need a substantially reworked evaluation section before it could be considered publishable.

major comments (4)
  1. [Abstract; Section VI, especially VI.A and VI.C] The headline claim of 'over 95% topology reconstruction accuracy' is not supported by any quantitative evaluation in the manuscript. Section VI reports feeder statistics (Table III), capacity violations (Table IV), and qualitative figures, but no accuracy table, no precision/recall, no confusion matrix, no error bars, and no numerical comparison to the 'baseline methods' mentioned in the abstract. In addition, the term 'ground truth' is never defined. If the ground truth is the same utility record that was used to construct the base topology G_B, then the result is partly circular because the solver preserves most of G_B by design (J_prior in Eq. (1)). The authors must state exactly what the 95% is measured against, provide per-feeder and overall accuracy numbers, and include a quantitative baseline comparison.
  2. [Section II.C, Eq. (1); Section V] The central inference objective in Eq. (1) is never made concrete. The terms J_elec, J_geo, and J_prior are described only in words, the weights α and β are not specified, and the feasible set Ω(C) is not formalized. More importantly, the actual solver in Section V uses heuristics—distance ratios, correlation thresholds, K-means, mutual information, and DBI-based scoring—with no derivation showing that these implement or approximately minimize Eq. (1). This makes the 'constrained inference' formulation and its implementation hard to connect, and it prevents reproducibility.
  3. [Section V.F, Lemma 1] The scalability guarantee is a tautological restatement of the locality assumption. Lemma 1 assumes |O|=γN with γ≪1 and bounded K, and then concludes the cost is O(NK). No evidence is given that γ is small in the studied feeders, and no sensitivity analysis is provided for how accuracy degrades when errors are not localized or when the suspicious-node detector in Section V.A misses the incorrect assignments. The paper's own §III.D offers only an anecdotal statement that 'utility record inaccuracies rarely corrupt entire feeders.' If that assumption fails, the localized reconnection procedure cannot recover the correct topology, and the claimed linear scaling is irrelevant to accuracy. The authors should empirically measure the size and spatial concentration of errors and the recall of the detection step.
  4. [Section V.D and VI.D] The reliability metric CL = 0.7·Score_DBI + 0.3·Score_corr is presented as a falsification-driven measure, but both components are the same clustering-quality and correlation criteria used to make the assignment in the first place. It is therefore unsurprising that the chosen assignment scores higher than alternatives; the metric may not provide independent evidence of confidence. The weights 0.7/0.3 are arbitrary, and Appendix B tests sensitivity only to K, not to these weights or to α, β, and τ. No comparison against field-verified connections is reported, so the operational value of the reliability scores is unvalidated.
minor comments (4)
  1. [Section II.B, Table II] The notation V is used both for the set of nodes and for voltage magnitudes (e.g., V_A_it), which is confusing. A distinct symbol for voltage, such as U or v, would improve readability.
  2. [Table IV] The column headings and units are unclear: 'Rating 10', 'Peak 16.6', and '8' are not explained. Is the rating in kVA? What is the limit column? The transformer IDs are redacted but the table still needs a self-contained caption.
  3. [Appendix B] The appendix is titled 'Sensitivity and Robustness Analysis' but it only varies the candidate-set size K. The hyperparameters α, β, τ, and the reliability weights 0.7/0.3 are not tested, so the claim that results are insensitive to parameter choices is broader than the presented evidence.
  4. [Section VI.C, Figure 8] The comparison among database, correlation-only, and proposed methods is qualitative. A numerical table showing the number/percentage of nodes corrected, incorrectly changed, or left unchanged would be much more informative.

Circularity Check

1 steps flagged · score 4.0 of 10

Confidence metric is self-referential; headline 95% accuracy has no defined ground truth, but core localized topology refinement is not itself circular.

  1. self definitional [Section V-D (Reliability Estimation via Falsification Testing) and Section VI-D]
    "Reliability is therefore evaluated via falsification testing. Alternative candidate transformers are tested, and clustering quality is compared via Davies–Bouldin Index: ScoreDBI = σ(log DBIfalse/DBItrue) ... Correlation-based support is similarly evaluated: Scorecorr = σ(Corrours/Corralt). The final confidence score becomes CL = 0.7ScoreDBI + 0.3Scorecorr. ... Corrected assignments exhibit clear confidence increases, while already correct mappings remain stable, indicating the solver strengthens reliable assignments rather than introducing instability."

    The reconnection step (V-C) selects assignments using K-means on the joint spatial-electrical feature space and statistical dependence (MI/correlation) with the same candidate transformer groups, accepting only updates that reduce the inference objective. The reliability score CL is a rescaled combination of the same kind of clustering-quality (DBI) and correlation criteria. Therefore the reported 'confidence increase' after correction is not independent falsification: it compares chosen versus alternative assignments on the same family of scores that steered the reassignment. The statement that the solver 'strengthens reliable assignments rather than introducing instability' is thus a restatement of the selection procedure, not confirmation from external or field-verified connections.

full rationale

The central inference chain is not circular: the solver starts from a noisy base record GB, detects suspicious nodes via distance/correlation heuristics, restricts candidates to local transformer neighborhoods, and reconnects by clustering plus feasibility constraints. The objective (1) includes J_prior, but that is a legitimate prior term rather than a hidden ground-truth fit. However, the paper never defines the ground truth for the abstract's 'over 95% topology reconstruction accuracy,' and the only topology comparison shown (Fig. 8) is against the 'database topology' GB, which is also the prior GB. If accuracy were measured against GB, it would be partly forced by J_prior and by the localized-preservation design (Ê = EB \ ΔE ∪ ΔÊ); because the metric is undefined, this remains a reporting gap rather than a demonstrated circular reduction. The concrete self-referential step is the reliability metric, which reuses the selection criteria as confidence evidence. The paper also contains several self-citations, but none is load-bearing for the central algorithm. Overall, the core derivation is independent; the self-referential confidence claim and the undefined accuracy basis justify a moderate score, not a high one.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim depends on a handful of hand-chosen trade-off weights, candidate sizes, and thresholds, plus the empirical assumption that topology errors are localized. The most important unstated input is the definition of ground truth for the 95% accuracy claim.

free parameters (6)
  • α (J_geo weight) = not specified
    Eq. (1) defines the objective but gives no value or tuning procedure for the spatial penalty weight.
  • β (J_prior weight) = not specified
    Eq. (1) defines the prior-deviation penalty but no value or tuning procedure is provided.
  • distance-ratio threshold τ = not specified
    Section V-A flags suspicious nodes when r_i > τ; τ is never quantified.
  • K (candidate transformer count) = K=2 or K=3 in reported examples; changes appear at K=4
    Appendix B tests K=2,3,4 and finds assignment changes for two of four regions at K=4; no principled selection is given.
  • reliability weights 0.7/0.3 = 0.7, 0.3
    Section V-D sets CL = 0.7·Score_DBI + 0.3·Score_corr by hand with no optimization or sensitivity analysis.
  • sigmoid mapping parameters = not specified
    σ(·) is used in Score_DBI and Score_corr but its scaling is not defined.
assumptions (6)
  • domain assumption Each node is connected to exactly one secondary transformer
    Section II opening; excludes multi-service or multi-phase edge cases and medium-voltage switching topology.
  • ad hoc to paper Most base-topology assignments are correct and errors are localized (|O|=γN with γ≪1)
    Core to the detector and Lemma 1; introduced from utility observation in §III-D, not guaranteed in general feeders.
  • domain assumption K is small and independent of feeder size due to physical locality
    Needed for the O(NK) linear-scaling claim in Lemma 1.
  • domain assumption Voltage correlation and geodetic distance are reliable discriminative signals inside local neighborhoods
    Used in suspicious-node detection and reconnection in §V-A/B; the paper itself shows correlation fails globally in dense feeders.
  • domain assumption Transformer capacity and nominal-voltage bands are sufficient to certify physical feasibility
    Section II-C and V-E enforce aggregate apparent power ≤ rating and voltage within [V_nom,min,V_nom,max]; no power-flow model or phase-balance check is performed.
  • domain assumption Ground truth for the 95% accuracy is correct and independent of the base records used as prior
    Needed to interpret the headline accuracy; the paper never states what the reconstruction accuracy is measured against.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enabling Scalable Topology Inference in Distribution Systems via Constrained Multi-Source Inference." pith.science (2026). https://pith.science/paper/VTCEBEA6

@misc{pith2026260720480,
  author       = {Pith},
  title        = {Pith review of: Enabling Scalable Topology Inference in Distribution Systems via Constrained Multi-Source Inference},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VTCEBEA6}},
  note         = {Machine review of arXiv:2607.20480}
}
abstract

Accurate distribution system topology is essential for outage localization, voltage analytics, and operation of distribution grids, yet maintaining reliable connectivity records remains challenging in practice due to heterogeneous and imperfect utility data. Existing topology identification methods often rely primarily on electrical similarity or spatial records alone, which become unreliable in dense feeders and under inconsistent metadata conditions. This paper formulates distribution topology identification as a constrained inference problem that refines a utility-provided base topology using heterogeneous evidence while enforcing spatial feasibility and physical operational constraints. Instead of reconstructing connectivity from scratch, the proposed framework detects inconsistent assignments, performs localized reconnection within constrained neighborhoods to ensure scalability, and iteratively enforces physical feasibility to produce operationally consistent topology estimates. In addition, a falsification-driven reliability metric evaluates how strongly each inferred connection is supported relative to alternative feasible assignments, enabling utilities to prioritize verification efforts while preserving system-wide observability. The framework is validated using operational data from three feeders comprising more than $8{,}000$ AMI meters in collaboration with a large U.S. utility. Results demonstrate over $95\%$ topology reconstruction accuracy while significantly reducing computational effort compared with global inference approaches. The study further shows that correlation-based methods alone produce ambiguous assignments in dense urban feeders, whereas combining electrical measurements with spatial and operational constraints enables robust and scalable topology recovery under realistic deployment conditions.

Figures

Figures reproduced from arXiv: 2607.20480 by the authors.

Figure 1
Figure 1. Feeders in the U.S. provided by our utility partner. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Before and after geocoding-based spatial refinement. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Voltage measurements before and after data cleaning. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 6
Figure 6. Figure 6: Illustration of confidence evaluation via falsification [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 8
Figure 8. Figure 8: Topology correction comparison. C. Constraint-Guided Topology Recovery [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Confidence improvement after constrained inference. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Constraint-guided correction resolving overload. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Confidence evaluation across feeders with diverse configurations. [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: Sensitivity analysis across regions [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

21 extracted references · 1 linked inside Pith

  1. [1]

    Distribution grid impedance & topology estimation with limited or no micro-pmus,

    H. Li, Y . Weng, Y . Liao, B. Keel, and K. E. Brown, “Distribution grid impedance & topology estimation with limited or no micro-pmus,” International Journal of Electrical Power & Energy Systems, vol. 129, p. 106794, 2021

    2021

  2. [2]

    Hd-deep-em: Deep expectation maximization for dynamic hidden state recovery using heterogeneous data,

    Z. Ma, H. Li, Y . Weng, E. Blasch, and X. Zheng, “Hd-deep-em: Deep expectation maximization for dynamic hidden state recovery using heterogeneous data,”IEEE Transactions on Power Systems, vol. 39, no. 2, pp. 3575–3587, 2023

    2023

  3. [3]

    Distributed algorithms for convexified bad data and topology error detection and identification problems,

    Y . Weng, M. D. Ili ´c, Q. Li, and R. Negi, “Distributed algorithms for convexified bad data and topology error detection and identification problems,”International Journal of Electrical Power & Energy Systems, vol. 83, pp. 241–250, 2016

    2016

  4. [4]

    Machine learning-enabled distribution network phase identification,

    Z. S. Hosseini, A. Khodaei, and A. Paaso, “Machine learning-enabled distribution network phase identification,”IEEE Transactions on Power Systems, vol. 36, no. 2, pp. 842–850, 2020. 9 CL: 0.574 CL: 0.426 TransformerThe transformer connected to the outlierEndpoints in different clusters Our method (a) Case 1 CL: 0.589 CL: 0.411Our method TransformerThe tra...

    2020

  5. [5]

    Guaranteed con- version from static measurements into dynamic ones based on manifold feature interpolation,

    L. Mai, H. Li, Y . Weng, E. Blasch, and X. Zheng, “Guaranteed con- version from static measurements into dynamic ones based on manifold feature interpolation,”IEEE Transactions on Power Systems, 2025

    2025

  6. [6]

    Tajer, S

    A. Tajer, S. M. Perlaza, and H. V . Poor,Advanced data analytics for power systems. Cambridge University Press, 2021

    2021

  7. [7]

    Efficient manifold-constrained neural ode for high-dimensional datasets,

    M. Guo, H. Li, and Y . Weng, “Efficient manifold-constrained neural ode for high-dimensional datasets,” inInternational Joint Conference on Neural Networks (IJCNN). IEEE, 2025, pp. 1–8

    2025

  8. [8]

    Graph mining for classifying and localizing solar panels in distribution grids,

    M. Guo, Q. Cui, and Y . Weng, “Graph mining for classifying and localizing solar panels in distribution grids,” inPanda Forum on Power and Energy (PandaFPE). IEEE, 2023, pp. 1743–1747

    2023

Show all 21 references
  1. [9]

    Identifying errors in service transformer connections,

    L. Blakely and M. J. Reno, “Identifying errors in service transformer connections,” inIEEE Power & Energy Society General Meeting (PESGM), 2020, pp. 1–5

    2020

  2. [10]

    Core process representation in power system operational models: Gaps, challenges, and opportunities for multisector dynamics research,

    K. Oikonomou, B. Tarroja, J. Kern, and N. V oisin, “Core process representation in power system operational models: Gaps, challenges, and opportunities for multisector dynamics research,”Energy, vol. 238, p. 122049, 2022

    2022

  3. [11]

    Data quality challenges in existing distribution network datasets,

    F. Geth, M. Vanin, and D. Hertem, “Data quality challenges in existing distribution network datasets,”IET, 2023

    2023

  4. [12]

    An efficient approach to power system uncertainty analysis with high-dimensional dependencies,

    Y . Wang, N. Zhang, C. Kang, M. Miao, R. Shi, and Q. Xia, “An efficient approach to power system uncertainty analysis with high-dimensional dependencies,”IEEE Transactions on Power Systems, vol. 33, no. 3, pp. 2984–2994, 2017

    2017

  5. [13]

    Spatial-temporal deep learning for hosting capacity analysis in distribution grids,

    J. Wu, J. Yuan, Y . Weng, and R. Ayyanar, “Spatial-temporal deep learning for hosting capacity analysis in distribution grids,”IEEE Transactions on Smart Grid, vol. 14, no. 1, pp. 354–364, 2022

    2022

  6. [14]

    Solar photovoltaic assessment with large lan- guage model,

    M. Guo and Y . Weng, “Solar photovoltaic assessment with large lan- guage model,”Applied Energy, vol. 402, p. 126835, 2025

    2025

  7. [15]

    Topology identification and line parameter estimation for non-pmu distribution network: A numerical method,

    J. Zhang, Y . Wang, Y . Weng, and N. Zhang, “Topology identification and line parameter estimation for non-pmu distribution network: A numerical method,”IEEE Transactions on Smart Grid, vol. 11, no. 5, pp. 4440– 4453, 2020

    2020

  8. [16]

    Exarnn: An environment-driven adaptive rnn for learning non-stationary power dy- namics,

    H. Li, M. Guo, Y . Weng, M. Ilic, and G. Ruan, “Exarnn: An environment-driven adaptive rnn for learning non-stationary power dy- namics,”arXiv preprint arXiv:2505.17488, 2025

    2025 arXiv

  9. [17]

    Phase identification in electric power distribution systems by clustering of smart meter data,

    W. Wang, N. Yu, B. Foggo, J. Davis, and J. Li, “Phase identification in electric power distribution systems by clustering of smart meter data,” in IEEE International Conference on Machine Learning and Applications (ICMLA), 2016, pp. 259–265

    2016

  10. [18]

    An introduction to optimal power flow: Theory, formulation, and examples,

    S. Frank and S. Rebennack, “An introduction to optimal power flow: Theory, formulation, and examples,”IIE Transactions, vol. 48, no. 12, pp. 1172–1197, 2016

    2016

  11. [19]

    Physical equation discovery using physics- consistent neural network (pcnn) under incomplete observability,

    H. Li and Y . Weng, “Physical equation discovery using physics- consistent neural network (pcnn) under incomplete observability,” in Proceedings of ACM SIGKDD Conference on Knowledge Discovery & Data Mining, 2021, pp. 925–933

    2021

  12. [20]

    Adaptive data fusion for state estimation and control of power grids under attack,

    T. Mortlock and M. A. Al Faruque, “Adaptive data fusion for state estimation and control of power grids under attack,”IEEE Transactions on Industrial Informatics, 2024

    2024

  13. [21]

    A joint estimation method of distribution network topology and line parameters based on power flow graph convolutional networks,

    Y . Wang, X. Shen, X. Tang, and J. Liu, “A joint estimation method of distribution network topology and line parameters based on power flow graph convolutional networks,”Energies, vol. 17, no. 21, p. 5272, 2024. 10 TransformerThe transformer connected to the outlierEndpoints i...

    2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.