Loot & Baloot

v3.2.3 (2026-05-13)

Full Changelog Previous Releases

• docs: add v3.2.3 release notes
  Pos-3 Sun bot fix release notes for commit 7d32416 (relocate F5-3
  pos-3 Sun Takbeer/Tasgheer donate above Rule 1B). Focused bot-play
  bugfix framing — no UI / protocol / saved-variable / scoring /
  network changes. v3.1.x / v3.2.x clients remain addon-message
  compatible.
  The four safeguards covered in the entry:
  • doesn't override smother's led-suit point-card donation
  • doesn't override Tahreeb signal encoding
  • filtered with not-wouldWin so it can't steal partner's trick
  • highestByRank routes tied donor cards through pickRandomTied
    (closes the v3.2.2-deferred F5/D-1 tie-randomization site)
    F2 pos-3 Sun hold-back stays intentionally deferred to a future
    deception-play audit.
    Verification: 1,258 / 0 harness, 20 / 0 standalone smokes.
    No tag yet — release-prep only. Design + 4 rounds of Codex review
    live in .swarm_findings/v3_2_3_pos3_sun_relocation_design.md.
    Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com
• docs(v3.2.3): align design counts with BF.7 emitting 2 assertions
  Codex review of pos3-sun-relocation-v3.2.3 at 7d32416 approved
  the runtime implementation but flagged a count drift in the
  design doc: the v0.4 doc predicted 1,257 / 0 final harness with
  +12 BF.* checks, but BF.7 ships with two assertions in the same
  do-block (BF.7a assertEq(card, "9S") + BF.7b
  assertTrue(card ~= "KS")) for tighter diagnostics — so the
  real delta is +13 and the harness lands at 1,258 / 0.
  Doc-only amend on this branch (no runtime/test changes):
  • §0 executive summary F5-3 row: "8 behavioural + 4 source-pin =
    12 new harness checks" → "8 BF behavioural blocks emitting 9
    assertion checks (BF.7 emits 2 assertions for tighter
    diagnostics) + 4 source-pin sub-asserts = 13 new harness
    checks".
  • §5 test count summary: BF.7 row now labelled "(2 harness
    checks: 7a + 7b)"; total reworded to "9 behavioural assertion
    checks (BF.1, BF.2, BF.3, BF.4, BF.5, BF.6, BF.7a, BF.7b,
    BF.9) + 4 source-pin sub-asserts = 13 new harness checks";
    final harness 1,245 → 1,257 changed to 1,245 → 1,258.
  • §5 pre-runtime expected: 7 failing checks unchanged (BF.1,
    BF.2, BF.9, BF.8a-d); passing list now "6 harness checks (five
    BF blocks producing six passing checks: BF.3, BF.4, BF.5, BF.6,
    BF.7a, BF.7b)" with explanation that BF.7's two assertions
    both pass pre- and post-fix.
  • §8 implementation order step 5: target harness 1,257 / 0 →
    1,258 / 0 with delta +13.
    Branch full harness on 7d32416 already verified locally by
    Codex at 1,258 / 0; standalone smokes at 11/0 and 9/0. This
    amend brings the design doc into alignment with the actual
    shipped test count.
    Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com
• fix(Bot.lua): relocate F5-3 pos-3 Sun Takbeer donate above Rule 1B
  Per .swarm_findings/v3_2_3_pos3_sun_relocation_design.md (v0.4
  after 4 rounds of Codex review): F5-3 (v1.4.1 Takbeer/Tasgheer
  pos-3 Sun donate non-A/T) used to live in the opp-winning region
  at old Bot.lua:4444-4495 but was structurally unreachable
  (its gate required partnerWinning while the enclosing scope
  guaranteed partnerWinning=false after the early-return at
  Bot.lua:3886). v3.2.3 relocates the branch into the live
  partnerWinning block, between Tahreeb sender end (~L3737) and
  Rule 1B start (~L3739).
  Final ordering inside the partnerWinning block:
  Smother → Tahreeb sender → F5-3 → Rule 1B
  Three Codex-required design properties of the relocated block:
  1. Uses not wouldWin(c, trick, contract, seat) to filter each
     donate candidate, preventing F5-3 from stealing partner's
     current trick via a same-suit K/Q/J. Replaces the original
     #winners == 0 gate (semantically tied to the opp-winning
     context where this branch used to live).
  2. Builds a donate pool and calls
     Primitives.highestByRank(donate, contract) rather than a
     manual donateRank / cr > donateRank strict ranking loop.
     This closes the v3.2.2-deferred F5/D-1 tie-randomization site
     at F5-3 — without it, making the branch reachable would have
     re-introduced the v1.1.0 hand-order leak in newly-live code.
  3. Placed AFTER Tahreeb sender (signals beat point-feed on
     overlap) and BEFORE Rule 1B (so F5-3 fires before Rule 1B's
     "second-lowest" biggest-mistake mitigation).
     F2 hold-back («تخليه يمسك») remains deferred to v3.2.4+ per the
     design doc — its probabilistic 30%/40% fire-rate needs a separate
     deception-play test framework that doesn't fit alongside this
     deterministic batch.
     Behavioural tests (tests/test_state_bot.lua, section BF, +13
     checks):
  • BF.1: wire-proof — pos-3 Sun void-in-led, pos-4 void, hand
    {KH, TH, QH} → F5-3 returns KH (highest non-A/T).
    Pre-fix returned QH via fallback.
  • BF.2: F5-3 fires regardless of partner being bot or human
    (v1.4.5 removed Tahreeb's partner-bot gate; F5-3 doesn't
    gate on partner-bot either).
  • BF.3: Tahreeb T-1 Bargiya wins overlap regression — Tahreeb
    runs first, F5-3 doesn't steal the signal.
  • BF.4: Smother's A-donate stays canonical (#pointCards>=2
    gateOk) — Interpretation A invariant.
  • BF.5: F5-3 doesn't fire without pos-4 void (gate respected).
  • BF.6: F5-3 doesn't fire at non-pos-3 positions.
  • BF.7: not-wouldWin filter regression guard (hand {KS, 9S, 8S,
    7H}, partner JS leading: filter rejects KS as it would
    beat partner; bot returns 9S in both pre- and post-fix
    states).
  • BF.8: source-pin block (4 sub-asserts: marker, "relocated",
    not-wouldWin filter substring, old-dead-block-removed).
  • BF.9: tie-randomization wire-proof — hand {KH, QH, KC, QC}
    with stub math.random(2)=2 → F5-3 returns KC (tied
    second K via highestByRank+pickRandomTied). Pre-fix
    returned QC (fallback lowestByRank tied second);
    post-fix with strict ranking would return KH (BUG case).
    Pre-runtime test outcome: exactly 7 fails as designed — BF.1,
    BF.2, BF.9, BF.8a, BF.8b, BF.8c, BF.8d. After runtime fix:
    1,245 → 1,258 / 0 (+13 BF.* checks; design doc estimated +12
    but BF.7 ships with 2 assertions for tighter diagnostics).
    Standalone smokes: test_H1_pin_J9_trump 11/0;
    test_H7_sun_shortest_lead 9/0.
    Files: Bot.lua, tests/test_state_bot.lua only.
    Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com
• docs: add v3.2.3 pos-3 Sun relocation design
  Design pass + 4 rounds of Codex review for relocating F5-3
  (Takbeer/Tasgheer pos-3 Sun donate non-A/T) out of its currently
  unreachable opp-winning location into the live partnerWinning
  block at Bot.lua:~3738, between Tahreeb sender and Rule 1B.
  Approved scope:
  • F5-3 relocation only.
  • F2 hold-back remains deferred to v3.2.4+ (probabilistic
    deception play; needs its own audit + test framework).
  • Final ordering: Smother → Tahreeb → F5-3 → Rule 1B.
  • Required shape: donate-pool + Primitives.highestByRank (closes
    the v3.2.2-deferred F5/D-1 tie-randomization site at F5-3 as
    part of making it reachable).
  • Required filter: not wouldWin(c, trick, contract, seat) per
    card (prevents F5-3 from stealing partner's current trick
    via same-suit K/Q/J).
    Test plan: BF.1-BF.9, 12 new harness checks. Expected pre-runtime
    fails: exactly 7 (BF.1, BF.2, BF.9, BF.8a-d). Expected final
    harness: 1,245 → 1,257.
    Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com