# When Can You Correct Distribution Drift in Temporal Graph Generation? A Sharpening--Drift Tension and an Impossibility for Observation-Based Correction

arXiv: https://arxiv.org/abs/2607.24662 · HTML: https://mlsanity.com/paper/2607.24662/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 13 references that are in the map.
- First posted on arXiv 27 Jul 2026.

## Builds on (most cited first, 12 of 12)

- [DiGress: Discrete Denoising diffusion for graph generation](https://mlsanity.com/paper/2209.14734/) (2022)
- [Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data](https://mlsanity.com/paper/2406.03736/) (2024)
- [Unlocking Guidance for Discrete State-Space Diffusion and Flow Models](https://mlsanity.com/paper/2406.01572/) (2024)
- [DeFoG: Discrete Flow Matching for Graph Generation](https://mlsanity.com/paper/2410.04263/) (2024)
- [Theoretical Benefit and Limitation of Diffusion Language Model](https://mlsanity.com/paper/2502.09622/) (2025)
- [Temporal Domain Generalization with Drift-Aware Dynamic Neural Networks](https://mlsanity.com/paper/2205.10664/) (2022)
- [Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees](https://mlsanity.com/paper/2509.16756/) (2025)
- [TIGGER: Scalable Generative Modelling for Temporal Interaction Graphs](https://mlsanity.com/paper/2203.03564/) (2022)
- [Error Bounds and Optimal Schedules for Masked Diffusions with Factorized Approximations](https://mlsanity.com/paper/2510.25544/) (2025)
- [DYMOND: DYnamic MOtif-NoDes Network Generative Model](https://mlsanity.com/paper/2308.00770/) (2023)
- [Primal-Dual Guided Decoding for Constrained Discrete Diffusion](https://mlsanity.com/paper/2605.09749/) (2026)
- [Evolving Graph Learning for Out-of-Distribution Generalization in Non-stationary Environments](https://mlsanity.com/paper/2511.02354/) (2025)

## Abstract (arXiv)

Generative models of temporal graphs are trained on one stretch of an evolving network and deployed on the next, and they degrade badly in the gap. We show this degradation is derivable, general, and not fixable from observations. The masked flow-matching loss decomposes exactly, with no independence assumption, into an irreducible entropy plus a divergence whose derivative along the training path is positive precisely for structures rare during training and common at deployment, diverging as their training probability goes to zero. Empirically the trade-off is a power law with exponent -0.605 (R2=0.9977), and drift raises the sampler's error floor without changing how many steps reach it: across seven well-powered conditions the drift-period marginal error varies by at most 6% over a 50× range of sampling budgets, while the floor sits 2.2× to 34.3× above the in-period floor. Because the deployment period is observed, correction looks like a matter of measurement. It is not. We prove that any corrector measurable with respect to past observations leaves at least the conditional variance of the statistic it tracks, and that trend extrapolation beats trusting the last observation only when μ2>v(1-2ρ). Both premises are measurable and both go the wrong way: the drift is trendless and mean-reverting, with a one-step innovation as large as the drift itself. An oracle removes 60% of the error, the best observation-based corrector recovers 5.7% of that, and extrapolation is strictly worse than doing nothing clever.