Pith. sign in

REVIEW 2 major objections 5 minor 34 references

The paper argues that simple, parameter-free encoders—fixed column-wise SQL aggregations paired with single-table foundation models—remain at least as good as pre-trained, parameterized relational encoders, and proves theoretical limits on

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 08:30 UTC pith:3I2M3PY7

load-bearing objection Solid, honest follow-up on parameter-free RDB encoders; the core empirical claim holds with caveats, but the 'no task-specific adjustments' line is overstrong because validation performance is used to pick the aggregation family. the 2 major comments →

arxiv 2607.05476 v2 pith:3I2M3PY7 submitted 2026-07-06 cs.LG

Parameter-Free Encoders Remain Viable for RDB Foundation Models

Linjie Xu , David Wipf This is my paper
classification cs.LG
keywords relational database foundation modelsparameter-free encodersin-context learningRDBLearntabular foundation modelslabel propagationSQL aggregationpre-training-free
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper is trying to establish that the added complexity of pre-trained, parameterized encoders for relational databases (RDBs) does not deliver a principled advantage over a much simpler, parameter-free design. It proves two information-theoretic limitations: a frozen encoder cannot reliably exploit label information in neighborhood subgraphs (Proposition 3.1), and labels cannot help an encoder figure out which feature columns matter (Proposition 4.1). Empirically, RDBLearn v1.1, using only fixed SQL aggregations plus a strong single-table foundation model, beats all open-source foundation models on all six benchmarks and beats closed-source ones on four of six. A sympathetic reader would care because it suggests RDB-specific pre-training may be unnecessary for many tasks, and that the frontier is in the single-table FM and aggregation design rather than the RDB encoder.

Core claim

The paper demonstrates through constructed distributions that any fixed encoder/prediction-head combination is limited when neighborhood labels are used as features: for certain deterministic labeling rules (e.g., parity checks), the encoder cannot do better than chance at test time, while task-specific supervised learning can fit the training labels perfectly (Proposition 3.1). Similarly, even when labels are provided as inputs, they carry no information about which feature columns constitute the true generative support, so labels cannot enable a trainable encoder to do feature screening (Proposition 4.1). Empirically, the authors show that RDBLearn v1.1, a parameter-free encoder with expan

What carries the argument

The central object is the parameter-free encoder of RDBLearn: it compresses relational context vertically within each column using a fixed set of SQL aggregations (mean, min/max, count, quantiles, etc.), avoiding any cross-column mixing. This 'vertical compression' design is paired with in-context learning via single-table foundation models (TabPFN-v3, TabICL-v2). The theoretical results hinge on adversarial constructions using parity (XOR) functions over labels or feature supports to show indistinguishability: no fixed encoder can extract the required signal from the given subgraph, and labels add no information about the support set.

Load-bearing premise

The parameter-free encoder relies on fixed aggregations computed within individual columns; if a prediction task requires combining information across columns or across tables in a way these aggregates cannot express, the parameter-free design will fail regardless of the downstream single-table model.

What would settle it

Find or construct an RDB benchmark where the target is an XOR or cross-column interaction of two or more feature columns (so that any per-column aggregation loses the signal), and show that a pre-trained parameterized encoder such as KumoRFM-2 solves it while RDBLearn v1.1 clearly fails, despite a supervised model on the raw rows succeeding.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • RDB-specific pre-training is not a prerequisite for strong performance; training-free pipelines remain competitive.
  • Adding neighbor labels to encoders does not, by itself, justify parameterized encoders; a fixed aggregation of labels may be just as useful.
  • Performance gains may come from improving single-table foundation models and aggregation sets rather than RDB encoder complexity.
  • Existing closed-source pre-trained RDB FMs may not offer a fundamental advantage on standard benchmarks, suggesting effort can be redirected to open approaches.
  • The theoretical results constrain the design space: any future encoder that uses labels must do so in a way that is task-adaptive (e.g., through ICL) rather than static.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The theoretical results suggest that adding label-derived aggregate features (e.g., label proportions in the neighborhood) to parameter-free encoders like RDBLearn could yield further gains without undermining their design — a direction the paper leaves open.
  • The empirical viability claim is limited to the benchmarks tested; on tasks requiring non-additive cross-column interactions (e.g., XOR of two columns), the parameter-free vertical aggregation approach may be insufficient, potentially leaving room for parameterized encoders.
  • The parity-based impossibility results may generalize: any frozen representation that is fixed across tasks cannot robustly exploit task-specific label relationships, a principle that extends beyond RDBs to other foundation-model settings.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper argues that parameter-free RDB encoders, when paired with single-table in-context-learning foundation models, remain competitive with parameterized, RDB-pretrained encoders. It contributes two propositions in the style of impossibility/existence results concerning the use of neighborhood labels as encoder inputs, and supports the thesis empirically with an updated RDBLearn v1.1 evaluated on six benchmark suites (SALT, RelBench-v1, RelBench-v2, 4DBInfer) against open-source and closed-source baselines. The central empirical claim is that RDBLearn outperforms all open-source foundation models on the considered benchmarks and exceeds closed-source models on four of six.

Significance. If the claim holds, the paper meaningfully simplifies RDB foundation-model design by showing that RDB-specific pretraining of encoder parameters is not necessary for strong performance. Strengths include explicit proof constructions in Appendix D, a publicly released implementation (RDBLearn v1.1), and evaluation over six diverse benchmark suites with many baselines. The paper also acknowledges the two RelBench-v1 benchmarks where closed-source models do better. The main weakness is that the empirical protocol is less clean than the headline 'no task-specific adjustments' claim suggests, because Appendix A discloses validation-based selection of the aggregation set. The theoretical propositions are correctly framed as existence results and are self-contained, though they are not strong enough to explain the empirical success by themselves.

major comments (2)
  1. [Section 5 / Appendix A] The claim that RDBLearn was run 'with no task-specific adjustments and zero exposure to RDB data of any kind at any stage' is contradicted by Appendix A: 'When available, the validation set performance is used to select between the original aggregations and the expanded group listed here.' Choosing between aggregation sets using a per-benchmark validation split is a task-specific adjustment to the core encoder, and validation labels are task-specific RDB signal. No ablation is provided to show that the rankings in Figure 1 and Tables 1-6 are invariant to this choice. Please add a fixed-aggregation comparison (e.g., always use the expanded set) and/or revise the claim; otherwise it is unclear whether the reported advantage reflects the parameter-free encoding principle or a validation-driven model selection.
  2. [Appendix A, 'Base Model Predictors' / 'Default Setting'] The paper does not specify which base single-table FM produced the results in Tables 1-6. Appendix A adds TabICL-v2 and TabPFN-v3, while the stated default for benchmarks without validation splits is TabPFNv2.5. Since the parameter-free encoder's output is consumed entirely by the base FM, the choice of base model is a substantial confound. A per-benchmark configuration table is needed, including how the two added base models were combined (if at all). Without this, the comparison to published KumoRFM-2 and TabPFN-REL results is under-specified.
minor comments (5)
  1. [Section 2.4] The sentence 'where relevance is largely indeterminate adequate label information' appears to be missing a word; it should probably read 'indeterminate without adequate label information.'
  2. [Appendix B.3] Typo: 'validaion-set' should be 'validation-set'.
  3. [Figure 1] The caption contains a stray '#' in 'lower is better #'.
  4. [Table 2] Formatting issue: '0.6210.16' should be separated into two numbers, and several entries appear as '1' instead of '1.00'.
  5. [Tables 1-6] No confidence intervals or repeated-seed results are reported. Many comparisons differ by less than one point in AUROC or MAE (e.g., Table 3, mimic-stay: 55.40 vs 56.33), so the rank-based conclusions should be interpreted as point estimates.

Circularity Check

0 steps flagged

No derivation-level circularity: the theory is self-contained and the empirical claims are benchmarked externally. Main caveat is a reporting inconsistency: Appendix A's validation-based aggregation selection contradicts Section 5's 'no task-specific adjustments' claim, but this does not make the central claim circular.

full rationale

The theoretical results (Propositions 3.1 and 4.1) are self-contained no-free-lunch/parity constructions; they do not assume the conclusion that parameter-free encoders are viable, and no equation in the paper reduces to a fitted value. The empirical claim is tested on external benchmarks (RelBench v1/v2, 4DBInfer, SALT) against published results of other models, so RDBLearn's performance is not obtained by predicting from its own fitted parameters. The paper does lean on same-author prior work (Xu et al. 2026; Zhang et al. 2026) for the parameter-free encoder design and RDBLearn toolbox, but the central viability claim has independent content: new theory plus externally benchmarked results. One issue worth flagging is not circularity but internal consistency: Section 5 says RDBLearn was run 'with no task-specific adjustments and zero exposure to RDB data of any kind,' while Appendix A states 'When available, the validation set performance is used to select between the original aggregations and the expanded group listed here.' This is a per-benchmark validation-based choice of the encoder's aggregation family, so the 'no adjustments' wording is inaccurate and an ablation isolating this selection is missing. However, selecting an aggregation set via validation is ordinary model selection, not a construction that forces the reported benchmark numbers; therefore it does not meet the threshold for a circularity step. Score 2 reflects the substantial but non-load-bearing self-citation, not a circular derivation.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical or mathematical entities; the parity constructions and support sets are proof devices, not invented model components. The main free parameters are model-selection choices (aggregation set, H) rather than fitted constants in a derivation.

free parameters (2)
  • Aggregation set per dataset = Selected via validation when available; otherwise the expanded default set
    Appendix A: validation performance is used to choose between the original and expanded aggregation groups. This is a per-task model-selection choice that affects the reported results.
  • Encoder depth H = H=4 (default for benchmarks without validation set)
    Appendix A default setting fixes H=4; no sensitivity analysis is reported.
axioms (4)
  • standard math Parity functions and ReLU networks with sufficient width/depth can represent and learn parity-based label functions.
    Invoked in Appendix D.1 (Proof of Proposition 3.1, Part 2), relying on Daniely & Malach (2020).
  • domain assumption Temporal RDBs exclude future rows; all labels in G*_H(x) have earlier timestamps than target x, and H>=2 for Prop 3.1.
    Section 3 defines G*_H and states labels within it have earlier time-stamps; the proof uses H>=2 to place extra neighbors.
  • domain assumption The six public benchmark suites (SALT, RelBench v1/v2, 4DBInfer) are representative of RDB foundation-model tasks.
    Section 5 and Appendix B.2 use these benchmarks to support the general claim 'parameter-free encoders remain viable'.
  • domain assumption Downstream single-table FMs (TabPFN-v2.5/TabPFN-v3, TabICL-v2) yield strong predictors on RDBLearn's aggregated features.
    Appendix A adds these as base predictors; if these FMs were weak, the parameter-free encoder would not be competitive.

pith-pipeline@v1.3.0-alltime-deepseek · 18022 in / 12506 out tokens · 120075 ms · 2026-08-02T08:30:03.557355+00:00 · methodology

0 comments
read the original abstract

Given a relational database (RDB) storing heterogeneous tabular information, how can we predict missing (or future) values in some target column of interest? As the space of potential targets is vast across enterprise settings, it is preferable to avoid learning a new model from scratch each time there is a new prediction task. Frozen foundation models based on RDB-specific encoders provide a viable solution, but ideal design remains an open question. On the one hand, it has recently been argued that certain parameter-free subgraph encoders combined with single-table foundation models can achieve near SOTA performance, with no RDB-specific pre-training required. Meanwhile, other contemporary studies advocate for parameterized encoders pre-trained to exploit observable labels for learning task-specific representations. To address this ambiguity, we analyze RDB encoder properties specifically when labels are present as inputs, proving limitations on the potential efficacy of trainable encoder parameters. As empirical validation, we demonstrate that considerably simpler parameter-free encoders are still capable of strong performance across many relevant benchmarking tasks.

Figures

Figures reproduced from arXiv: 2607.05476 by David Wipf, Linjie Xu.

Figure 1
Figure 1. Figure 1: RDBLearn outperforms all open-source FMs on all benchmarks, and even exceeds all closed-source FMs on 4 of 6 benchmarks. tiation generally possible even when GH(x) is switched to G ∗ H(x)? The answer is negative in the following sense: Proposition 4.1. Given any representative RDB ego￾network GH(xtest) constructed from non-empty tabular fea￾ture columns and |yH(xtest)| > 0, there will always exist distribu… view at source ↗
Figure 2
Figure 2. Figure 2: Taxonomy of existing RDB foundation models [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 19 linked inside Pith

  1. [1]

    RelGNN: Com- posite message passing for relational deep learning.arXiv preprint arXiv:2502.06784,

    Chen, T., Kanatsoulis, C., and Leskovec, J. RelGNN: Com- posite message passing for relational deep learning.arXiv preprint arXiv:2502.06784,

    Pith/arXiv arXiv

  2. [4]

    In fact, there may well be many instances where their inclusion directly benefits parameter-free encoder models like RDBLearn

    Note that even though we have argued in the main text that the presence of such labels need not necessarily favor a parameterized encoder over a parameter-free one, our analysis doesnotsuggest that such labels (when available) possess no advantage as extra encoder inputs. In fact, there may well be many instances where their inclusion directly benefits pa...

    2022

  3. [5]

    E., Ranjan, R., Robinson, J., Ying, R., You, J., and Leskovec, J

    Fey, M., Hu, W., Huang, K., Lenssen, J. E., Ranjan, R., Robinson, J., Ying, R., You, J., and Leskovec, J. Rela- tional deep learning: Graph representation learning on relational databases.arXiv preprint arXiv:2312.04615,

    Pith/arXiv arXiv

  4. [7]

    B., B ¨uhler, M., Garg, A., Sa- faric, D., Robertson, J., J¨ager, B., Alessi, S., Hayler, A., Moroshan, V ., Purucker, L., Singer, P., Arazi, A., Siems, J., Metzen, J

    Grinsztajn, L., Fl¨oge, K., Key, O., Birkel, F., Jund, P., Roof, B., Manium, M., Hoo, S. B., B ¨uhler, M., Garg, A., Sa- faric, D., Robertson, J., J¨ager, B., Alessi, S., Hayler, A., Moroshan, V ., Purucker, L., Singer, P., Arazi, A., Siems, J., Metzen, J. H., Grab, G., Erickson, N., Guo, S., Kalfon, E., Bing, S., Salinas, D., Cornu, C., Wehrhahn, L. C., ...

    Pith/arXiv arXiv

  5. [8]

    RelBench v2: A large-scale bench- mark and repository for relational data.arXiv preprint arXiv:2602.12606,

    Gu, J., Ranjan, R., Kanatsoulis, C., Tang, H., Jurkovic, M., Hudovernik, V ., Znidar, M., Chaturvedi, P., Shroff, P., Li, F., et al. RelBench v2: A large-scale bench- mark and repository for relational data.arXiv preprint arXiv:2602.12606,

    Pith/arXiv arXiv

  6. [9]

    O., and Pfister, T

    Han, S., Yoon, J., Arik, S. O., and Pfister, T. Large language models can automatically engineer features for few-shot tabular learning.arXiv preprint arXiv:2404.09491,

    Pith/arXiv arXiv

  7. [10]

    Huang, X., Tichelman, L., Kim, J., Olejniczak, K., and Ceylan, ˙I. ˙I. RelAgent: LLM agents as data scientists for relational learning.arXiv preprint arXiv:2605.07840,

    Pith/arXiv arXiv

  8. [11]

    E., Leskovec, J., and Fey, M

    Hudovernik, V ., L´opez, F., Kocijan, V ., Nitta, A., Lenssen, J. E., Leskovec, J., and Fey, M. KumoRFM-2: Scaling foundation models for relational learning.arXiv preprint arXiv:2604.12596,

    Pith/arXiv arXiv

  9. [12]

    SALT: Sales autocompletion linked business ta- bles dataset

    Klein, T., Biehl, C., Costa, M., Sres, A., Kolk, J., and Hof- fart, J. SALT: Sales autocompletion linked business ta- bles dataset. InNeurIPS 2024 Third Table Representation Learning Workshop,

    2024

  10. [14]

    and Schulte, O

    Motl, J. and Schulte, O. The CTU Prague relational learning repository.arXiv preprint arXiv:1511.03086,

    Pith/arXiv arXiv

  11. [17]

    Masked label prediction: Unified message passing model for semi-supervised classification.arXiv preprint arXiv:2009.03509,

    Shi, Y ., Huang, Z., Feng, S., Zhong, H., Wang, W., and Sun, Y . Masked label prediction: Unified message passing model for semi-supervised classification.arXiv preprint arXiv:2009.03509,

    Pith/arXiv arXiv 2009

  12. [19]

    Bag of tricks for node classification with graph neural networks.arXiv preprint arXiv:2103.13355,

    Wang, Y ., Jin, J., Zhang, W., Yu, Y ., Zhang, Z., and Wipf, D. Bag of tricks for node classification with graph neural networks.arXiv preprint arXiv:2103.13355,

    Pith/arXiv arXiv

  13. [20]

    Relational in- context learning via synthetic pre-training with structural prior.arXiv preprint arXiv:2603.03805,

    Wang, Y ., You, J., Shi, C., and Zhang, M. Relational in- context learning via synthetic pre-training with structural prior.arXiv preprint arXiv:2603.03805,

    Pith/arXiv arXiv

  14. [21]

    Tackling prediction tasks in relational databases with LLMs.arXiv preprint arXiv:2411.11829,

    Wydmuch, M., Borchmann, Ł., and Grali´nski, F. Tackling prediction tasks in relational databases with LLMs.arXiv preprint arXiv:2411.11829,

    Pith/arXiv arXiv

  15. [22]

    GFS: Graph- based feature synthesis for prediction over relational databases.arXiv preprint arXiv:2312.02037, 2023a

    Zhang, H., Gan, Q., Wipf, D., and Zhang, W. GFS: Graph- based feature synthesis for prediction over relational databases.arXiv preprint arXiv:2312.02037, 2023a. Zhang, X., Maddix, D. C., Yin, J., Erickson, N., Ansari, A. F., Han, B., Zhang, S., Akoglu, L., Faloutsos, C., Mahoney, M. W., et al. Mitra: Mixed synthetic priors for enhancing tabular foundation...

    Pith/arXiv arXiv

  16. [23]

    Rdbench: ML benchmark for relational databases.arXiv preprint arXiv:2310.16837, 2023b

    Zhang, Z., Yang, Y ., Zou, L., Wen, H., Feng, T., and You, J. Rdbench: ML benchmark for relational databases.arXiv preprint arXiv:2310.16837, 2023b. Zykov, R., Artem, N., and Alexander, A. Retailrocket rec- ommender system dataset,

    Pith/arXiv arXiv

  17. [25]

    • Feature Merging Efficiency:In RDBLearn, all relational features are generated by SQL executions

    and TabPFN-v3 (Grinsztajn et al., 2026). • Feature Merging Efficiency:In RDBLearn, all relational features are generated by SQL executions. The actual number of features can be large, varying according to RDB schema complexity, table sizes, and H. To accelerate SQL aggregations as needed to form these features we implement a chunk-based merging process. A...

    2026

  18. [27]

    that incorporates task-specific labels to the encoder as discussed in Section 4 of our main text. We also considerTabPFN-REL, which is likely based at least in part on the recently-released TabPFN-v3 (Grinsztajn et al., 2026), although no RDB-specific details of the architecture have been disclosed as of yet. Lastly, we compare withOpenRFM(Chen et al., 20...

    2026

  19. [28]

    On the open-source ICL-based side we include results fromRDB-PFN(Wang et al.,

    models, as well as an updated RT version enhanced withPluRelsynthetic pretraining data (Kothapalli et al., 2026). On the open-source ICL-based side we include results fromRDB-PFN(Wang et al.,

    2026

  20. [29]

    (2026) for details)

    and our own implemented baselineRandGNN, which combines a random-weight GNN encoder with a single-table FM (see Xu et al. (2026) for details). RandGNN represents a form of ablation w.r.t. RDBLearn, in that it explicitly mixes cross-column information in a manner at odds with the theoretical considerations upon which RDBLearn is based. And finally, for pur...

    2026

  21. [30]

    (2026)) andLLM-B(Wydmuch et al., 2024); both approaches are predicated on serialized RDB neighborhood representations and/or ICL samples

    (as tested in Ranjan et al. (2026)) andLLM-B(Wydmuch et al., 2024); both approaches are predicated on serialized RDB neighborhood representations and/or ICL samples. Although the underlying LLMs are not open-source, we nonetheless categorize these two approaches as open-source in that the prompting needed to reproduce results are publicly available and no...

    2026

  22. [31]

    Task-Specific Supervised Models.While our emphasis is on RDB foundation models, it is still useful to contextualize FM performance results w.r.t

    considers models both with and without real-world RDB data for pre-training, performance of the synthetic-only version is weaker and not included in our comparisons. Task-Specific Supervised Models.While our emphasis is on RDB foundation models, it is still useful to contextualize FM performance results w.r.t. methods benefitting from per-task supervised ...

    2026

  23. [32]

    Beyond these, we test over the 4DBInfer (Wang et al.,

    and RelBench-v2 (Gu et al., 2026). Beyond these, we test over the 4DBInfer (Wang et al.,

    2026

  24. [33]

    as adopted by Hudovernik et al. (2026). Note that for RelBench-v1, we omit only rel-f1 and rel-event datasets because of potential leakage issues mentioned elsewhere. B.3. Subtleties of RDB Benchmarking Language Model Agents.A growing body of evidence suggests that LLM agents can play a useful role in tabular model development (Han et al., 2024; Huang et ...

    arXiv 2026

  25. [34]

    Such functions can be directly represented by ReLU networks, and while not a requirement for this proof, are even efficiently learnable with additional assumptions (Daniely & Malach, 2020). D.2. Proof of Proposition 4.1 High-Level Intuition.The proof follows through modifications applied to the original parity setup from Appendix D.1. For illustrative sim...

    2020

  26. [2009]

    TabPFN-2.5: Advancing the state of the art in tabular foundation models.arXiv preprint arXiv:2511.08667,

    Grinsztajn, L., Fl¨oge, K., Key, O., Birkel, F., Jund, P., Roof, B., J ¨ager, B., Safaric, D., Alessi, S., Hayler, A., et al. TabPFN-2.5: Advancing the state of the art in tabular foundation models.arXiv preprint arXiv:2511.08667,

    Pith/arXiv arXiv

  27. [2014]

    Dwivedi, V

    URL https://kaggle.com/competitions/ acquire-valued-shoppers-challenge. Dwivedi, V . P., Jaladi, S., Shen, Y ., L´opez, F., Kanatsoulis, C. I., Puri, R., Fey, M., and Leskovec, J. Relational graph transformer.arXiv preprint arXiv:2505.10960,

    arXiv

  28. [2015]

    Justifying recommendations using distantly-labeled reviews and fine-grained aspects

    Ni, J., Li, J., and McAuley, J. Justifying recommendations using distantly-labeled reviews and fine-grained aspects. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Pro- cessing, pp. 188–197,

    2019

  29. [2019]

    Qu, J., Holzm¨uller, D., Varoquaux, G., and Morvan, M. L. TabICLv2: A better, faster, scalable, and open tabular foundation model.arXiv preprint arXiv:2602.11139,

    arXiv

  30. [2020]

    Gemma 3 Technical Report.arXiv preprint arXiv:2503.19786,

    Team, G., Kamath, A., Ferret, J., Pathak, S., Vieillard, N., Merhej, R., Perrin, S., Matejovicova, T., Ram ´e, A., Rivi`ere, M., et al. Gemma 3 Technical Report.arXiv preprint arXiv:2503.19786,

    Pith/arXiv arXiv

  31. [2022]

    kaggle.com/dsv/4471234

    URL https://www. kaggle.com/dsv/4471234. 6 Parameter-Free Encoders Remain Viable for RDB Foundation Models A. RDBLearn Updates We describe a few modest updates to the RDBLearn package, which collectively constitute what we henceforth refer to as RDBLearn version 1.1.2 These include the following: • Aggregations:Because the original RDBLearn package (Zhang et al.,

    arXiv

  32. [2024]

    P., Hoffart, J., Guestrin, C., and Leskovec, J

    Kothapalli, V ., Ranjan, R., Hudovernik, V ., Dwivedi, V . P., Hoffart, J., Guestrin, C., and Leskovec, J. PluRel: Syn- thetic data unlocks scaling laws for relational foundation models.arXiv preprint arXiv:2602.04029,

    Pith/arXiv arXiv

  33. [2025]

    OpenRFM: Dissecting relational in-context learning.arXiv preprint arXiv:2606.04320,

    Chen, Z., Yin, J., Gu, J., Xiong, S., Liu, X., Zhang, R., Zhou, K., and Guo, K. OpenRFM: Dissecting relational in-context learning.arXiv preprint arXiv:2606.04320,

    Pith/arXiv arXiv

  34. [2026]

    Supervised learning on relational databases with graph neural networks.arXiv preprint arXiv:2002.02046,

    Cvitkovic, M. Supervised learning on relational databases with graph neural networks.arXiv preprint arXiv:2002.02046,

    Pith/arXiv arXiv 2002