https://gblayer.github.io/forward-pass-newsletter/Your weekly digest of the top papers in tabular AI.https://gblayer.github.io/forward-pass-newsletter/issues/issue-11.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-11.htmlThu, 13 Aug 2026 21:45:38 +000010 papers, 1 industry update — Two papers probe TFM reliability from opposite ends: TabPFN/TabICL/TabDPT all fail basic Bayesian self-consistency checks, while frontier LLMs uniquely lose accuracy as tabular input dimensionality grows.https://gblayer.github.io/forward-pass-newsletter/issues/issue-10.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-10.htmlMon, 03 Aug 2026 07:48:22 +000010 papers, 1 industry update — A 9-model benchmark (TabPFN v2/v2.5/v2.6/v3, TabFM, Mitra, TabICL/v2, LimiX) finds all tabular FMs degrade under distribution shift, with real-data pretraining helping in-distribution fit more than true OOD robustness.https://gblayer.github.io/forward-pass-newsletter/issues/issue-9.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-9.htmlMon, 27 Jul 2026 08:08:54 +00009 papers — TabPFN's internal representation topology (persistent homology) tracks when its in-context predictions become unreliable.https://gblayer.github.io/forward-pass-newsletter/issues/issue-8.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-8.htmlMon, 20 Jul 2026 07:49:36 +00008 papers, 1 industry update — Two papers push TabPFN into new territory: as an in-context head for discrete-choice/marketing panels (2607.13314) and as a calibration-boosting classifier head on frozen multimodal embeddings (2607.11007).https://gblayer.github.io/forward-pass-newsletter/issues/issue-7.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-7.htmlThu, 16 Jul 2026 07:43:14 +00001 paper — Quiet day for tabular ML: 49 candidates screened, only one cleared the relevance bar.https://gblayer.github.io/forward-pass-newsletter/issues/issue-6.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-6.htmlWed, 15 Jul 2026 07:43:20 +00002 papers — GRAFT reframes table retrieval in data lakes as graph matching, gaining ~8-11% over prior retrievers on Spider/BIRD.https://gblayer.github.io/forward-pass-newsletter/issues/issue-5.htmlhttps://gblayer.github.io/forward-pass-newsletter/issues/issue-5.htmlTue, 14 Jul 2026 07:44:15 +00004 papers — Graph foundation models are borrowing tabular ICL backbones: GTAlign aligns graph structure to a TabPFNv2.5/LimiX-16M input format for in-context node/graph classification.