Loot & Baloot

v3.1.14 (2026-05-11)

Full Changelog Previous Releases

• v3.1.14: Codex delta review — real Local* paths for SWA deny + escalation retries
  Single tag bundling two batches of audit work, plus two Codex delta-
  review fixes for the same bug class hitting both batches.
  Batch 1 (v3.1.13 documented section): round-2 wire-retry sweep
  adding 250ms re-broadcast to 10 additional message types — MSG_MELD,
  MSG_AKA, MSG_BID, MSG_CONTRACT, MSG_BELOTE, MSG_BIDCARD, MSG_DOUBLE,
  MSG_TRIPLE, MSG_FOUR, MSG_GAHWA. Closes nearly all remaining items
  from the v3.1.12 Codex audit's deferred P2 list.
  Batch 2 (v3.1.14 documented section): two Codex delta reviews caught
  the same bug class twice — "helper-level retry test misses the
  Local* integration path":
  Bug 1 — SWA deny: N.LocalSWAResp(false) cleared S.s.swaRequest
  immediately after calling N.SendSWAResp, killing the v3.1.12 retry
  guard before the 0.25s timer fired. Fix: keep swaRequest alive
  through the retry window via responses[localSeat]=false, then clear
  at 0.35s with identity guards. The 3s swaDenied toast is preserved
  independently.
  Bug 2 — Escalation chain: SendDouble/Triple/Four/Gahwa retry guards
  checked pre-apply state (not contract.doubled, phase == PHASE_DOUBLE),
  but LocalDouble/Triple/Four/Gahwa and bot dispatch call S.Apply*
  BEFORE N.Send* — by the time the 0.25s retry fires, the flag is set
  and phase has advanced. Pre-apply guard always failed in production.
  Fix: switch to post-apply identity guards capturing the contract
  table reference at send time, plus verifying flag set and (for
  SendDouble) doublerSeat == seat. Phase guards dropped because
  Apply* intentionally advances the phase.
  Optional cleanup: SendContract retry guard now includes trump
  equality for consistency with the (bidder, type, trump) keying
  claim.
  Tests: 1058/1058 pass (was 982, +76 new behavioral pins in section
  AZ). AZ.23-25 replaced with integration tests starting from
  N.LocalDouble/Triple/Four/Gahwa, not from N.Send* directly.
  Process lesson codified: the AZ section now consistently starts from
  the user-facing entry point (N.Local*) and asserts wire effects
  downstream — never the inverse. Both Codex review rounds applied the
  same north star ("helper-level tests miss Local* bypasses"); v3.1.14
  finally applies it uniformly across the wire-retry test surface.
  Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com