Independent numerical audit of arXiv 2605.31172 (OpenReview Iww9TICvKj), ICML 2026.
The paper proves stability + almost-sure convergence of two-timescale stochastic approximation under
Markovian noise (prior work required i.i.d.), with no projection operator, and applies it to
TDC(λ) — off-policy temporal-difference learning with gradient correction and eligibility traces,
linear function… See the full description on the dataset page:
https://huggingface.co/datasets/MarxistLeninist/repro-tdc-2timescale-markovian-rl-bundle.
