wpr-verify 2.0.0 (revised manuscript)

2.0.0 — 2026-09-28 (revised manuscript)

Added

• wpr_faults.py: failure tables holding node and link failures; the recording
  criteria of Algorithms 5–6 (including the bookkeeping that turns a fully isolated
  node into a node record); the notification of Algorithms 7–8 simulated in
  synchronous rounds, so that several failures can be notified concurrently
  (concurrent model) or one after the other (sequential model); the two
  forwarding rules — original (Algorithms 11–12 as originally submitted) and
  trace (the revised rule: the header carries the set of failures the current path
  avoids, and a node recomputes only when its table is not contained in it) — and a
  greedy failure-oblivious baseline reproducing the behaviour of RPS and GRWMS.
• wpr_experiments.py: the experiments of the revised paper — two-failures
  (exhaustive, sequential and concurrent, node or node+link), k-failures (random
  concurrent node or link failures), mitigation (affected set, blind-spot radius
  ρ*, probe radius and courtesy hops), large (10^4 to 3.6×10^5 nodes), borel (the
  RCRR scenario on B(47,23) with 5–35 % of the links failed), greedy, and
  self-test (recomputes the quick tables and compares them with results/).
• wda_cost.g and run_wda.py: size of the shortLex automatic structure (word
  acceptor, word-difference automata, general multiplier) of every family, computed
  with GAP/kbmag, with the computation time.
• results/: CSV files behind every table of the paper (link_failures/ and
  extra_families/ hold the runs mentioned in the text but not tabulated), plus
  wda_cost.csv / wda_cost.txt.
• LICENSE (MIT, as declared on Zenodo), CITATION.cff, .zenodo.json.

Changed

• README.md: documents the new scripts, the two forwarding rules, the meaning of
  the criterion set vs. the notified set, and corrects the claim about --radius 2 --courtesy-hops 1 (the probe radius must be at least the blind-spot radius ρ*,
  which is 2 for Bubble-sort graphs and m/2−2 for the m×m torus).

Unchanged

• wpr_verify.py (single node failure; Tables 5, 8 and 9) and
  edge_transitivity.g. Under a single failure the two forwarding rules visit the
  same nodes, so the published single-failure figures are unaffected.

Known limitations

• The original rule is kept only to reproduce the comparison; it is not loop-free
  with two or more failures (see Remark 1 of the paper).
• kbmag's gpaxioms overflows its file-name buffers with long relators and long
  temporary paths; run_wda.py uses a short temporary directory and, for the
  100×100 torus, a gpaxioms rebuilt with larger buffers (see results/wda_cost.txt).
  The Star and Pancake graphs of Sym_8 did not complete within a ten-minute cap.