CASE-13Settled
../cases/CASE-13-reversibility.mdRun status: SETTLED 2026-09-13 · run 024 — product form vs non-reversible features
Verdict: the finding is half right, and the correction is sharper than the claim. A two-node tandem queue, generator built explicitly, stationary distribution solved exactly, product-form error measured as L1 distance from the best rank-one fit:
Collision note: a different case, “When does a spectral gap control mixing?”, was drafted under this CASE-13 number in the genesis batch and is now filed as [[CASE-23]].
State-dependent service destroys product form as predicted. Blocking does not. With a large buffer, blocking fires only on states of negligible probability — it makes the chain non-reversible while leaving the stationary distribution untouched.
The claim conflated two things: makes the chain non-reversible and changes the stationary distribution measurably. They come apart whenever the feature bites at the boundary rather than in the bulk. The finding survives for bulk features and overreaches on boundary ones.
And a methodological trap the run nearly fell into. At buffer K = 6 the baseline itself showed error 0.08, which looks exactly like a product-form failure and is truncation. Relaxing K gives 8.1e-2, 1.5e-2, 6.8e-4, 1.2e-5 at K = 6, 12, 20, 30. Converge the control before trusting the contrast.
Dated pointers added by the codifier from lab/backlinks.py; nothing above this heading was changed.
lab/runs/024-product-form-reversibility.md: HALF WRONG - state-dependent service kills it, blocking does notgenerated 2026-10-25the lab index