[guluchen]

Renames seven tier5 stems whose lower bounds were raised in single 6 h runs of GH-225 (#225) and GH-227 (#227, PR #229) and have now been reproduced. This follows the pattern of #223 / #216: same matrices, and only the Gx/Gz header comments change.

Rule used: an lb goes into a stem only when at least two runs with different search trees reached it, with the last level EMPTY on both halves (X half only under a verified XZ-dual map: TN_496 and the color codes). The trees differ by relabelled qubits and rows (-seed=1/-seed=2), a different row order (-cc-roworder=bfs), or a different binary and rule set. This is the same CC engine family, not an independent second method.

Reproduction runs: main 1239e9a built on yfclab2, 12 h each, one pinned core per run (CPUs 64-95), 4 GB cap. "E" = CC default plus erasure (-cc-rules=excl,pack,orbf,parity,leaf,overlap,charge,tiebreak,sparse,sparseov,sparsemrv,erasure, ungated on this binary); "A" = CC default without erasure.

Old	New	Original single runs (6 h, seed 0)	Reproduction (12 h; time to lb)	Without erasure
TN_324_4_22to25	TN_324_4_23to25	GH-225 E: 23 at 20 985 s	E -seed=1 30 309 s; E -seed=2 31 369 s	22 (seed 0, and -seed=1 in 12 h)
TN_496_2_27to30	TN_496_2_28to30	GH-225 E 17 996 s; GH-227 C 19 683 s	E -seed=1 20 699 s; E -seed=2 19 095 s	27 at seed 0; 28 with -seed=1 (24 919 s)
TN_540_4_23to34	TN_540_4_24to34	GH-225 E 19 168 s; GH-227 C 18 399 s	E -seed=1 35 026 s; E -seed=2 32 893 s	23 at seed 0; 24 with -seed=1 (42 322 s)
TN_648_14_21to28	TN_648_14_22to28	GH-225 E 20 292 s; GH-227 C 19 331 s	E -seed=1 35 385 s; E -seed=2 36 573 s	21 at seed 0; 22 with -seed=1 (36 164 s)
COLOR_1027_1_28to37	COLOR_1027_1_31to37	GH-225 E 16 090 s; GH-227 C 18 190 s	E -cc-roworder=bfs 33 206 s	30 at seed 0; 28 with -seed=1
TN_684_2_22to29	TN_684_2_24to29	lb 23 in all four GH-225/GH-227 arms (16 367-21 196 s)	E -seed=1: 23 at 13 945 s, 24 at 39 232 s; E -seed=2: 23 at 14 375 s, 24 at 39 063 s	23 with -seed=1 (17 570 s)
COLOR_2791_1_26to61	COLOR_2791_1_28to61	GH-225 A (42d7d1d) 10 312 s and GH-227 M (1239e9a) 11 706 s, one tree; GH-227 C (755e0c2, gated erasure) 9 612 s, a different tree	(from the existing runs)	28

Notes:

• TN_684 goes to 24, one above the [Mac agent] CC engine: erasure bound as a default rule (on top of GH-217 candidate A) #225 claim: both relabelled erasure runs completed level 23 EMPTY on both halves, with different node counts.
• Not renamed (still PLAUSIBLE, single tree):
  • COLOR_1027 lb 32: an E run with -cc-roworder=index searched the same tree as the [Mac agent] CC engine: erasure bound as a default rule (on top of GH-217 candidate A) #225 erasure run and reached lb 32 at 37 190 s.
  • COLOR_2791 lb 29: [Mac agent] CC engine: cost-aware erasure gate (GH-225 follow-up) #227 C only, at 20 721 s. Two 12 h runs on main 755e0c2 (-cc-roworder=bfs / index) are checking it; they are not part of this PR.
• On COLOR_1027, the relabelled runs are about 7x slower per level and reached only lb 29 in 12 h. The bfs order carries the reproduction there.
• E.bfs on TN_496 searched the same tree as the [Mac agent] CC engine: erasure bound as a default rule (on top of GH-217 candidate A) #225 erasure run, so it is not counted; the seeded runs are.
• Upper bounds are unchanged.
  • TN: the [Mac agent] Stern/Dumer collision step in the verified upper-bound finder (large open codes) #146 witnesses are copied under the new stems in optimization/experiments/GH-146/witnesses/renamed-2026-11-06/, and VERIFY.txt reports 5/5 VALID.
  • COLOR: the family formula (d = 37 / 61).

Records: optimization/experiments/GH-225/lb-repro/ holds:

• REPRODUCTION.md;
• analysis.md, with per-run times and a tree comparison against the original runs by node counts per half and level;
• results, jobs, the runner and full logs.

All 25 runs passed the runner's science check, apart from the COLOR_1027 index run: I stopped that run by hand at 42 721 s (rc -15), after it had reached lb 32.

Changes:

• git mv of the Hx/Hz/Gx/Gz files;
• rename-history rows and a witness note in data/matrices/README.md;
• dated paragraphs in NOTICE (sections 5 and 9);
• stem names in optimization/baselines/SERVER.md, main_server_reference.json, and in data/matrices/untiered/MANIFEST.csv, validation.json and README.md (placed-stem column of the two color codes).

QDistSAT mirror: guluchen/QDistSAT PR (same branch name, data/gh225-lb-stems); all 80 matrix files are byte-identical to this branch.

Hub: #15.

🤖 Generated with Claude Code