TSPES scoreboard
Exact optimum already reached on wi29, dj38, qa194, uy734 and zi929. The active lu980 route progressed 11,367 → 11,366 → 11,365 → 11,364 → 11,358 → 11,355. Only 15 length units remain.
| Arena | Role | Current evidence | Gap / state | Assessment |
|---|---|---|---|---|
| TSPES v0.1.4.2 W6 | flagship product/solver | lu980 = 11,355 vs 11,340; 5/9 tracked instances exact; TEP gen1 epoch2 | 0.132275% | strongest current solver line |
| DEV2.3 | nu3496 hard-tail donor | 96,780 vs 96,132; cycle19 P5; last global best 5 Sep | 0.674073% | healthy, real plateau |
| DEV2.4 HF3 | hard-tail successor/control | 96,780; 318,291 proposals; strict GPU+Rust physically exercised | 0.674073% | tied donor; superiority not shown |
| BD-RPHS v0.1.2 W4 | sensor / representation testbed | active 79,374 vs 79,114; historical W6 79,288 won by N2_shuffled_reverse null control | 0.328640% active | saturated but scientifically useful |
| 8zTSP-R v0.1.1 R2/HF1 | pure relational native research | round26; 8/9 instances improved vs prior snapshot; coalitions non-stationary | absolute gaps still poor | research-alive, not product-competitive |
| 8zTSP-R v0.2.0 HF2 MVC | multi-view compression-native | wi29 / COMPRESSION_PROPOSES; gpu-strict requested; reused pre-commit snapshot | unranked | initializing; fresh committed evidence pending |
nu3496 hard tail
DEV2.3 and DEV2.4 both stand at 96,780. This is durable shared-project evidence, not independent algorithmic replication. DEV2.4 has not yet beaten the donor.
RPHS falsifier
The historical W6 result is numerically better than active W4, but its winner was a shuffled-reverse null control. The null result stays visible because it is more scientifically informative than a cherry-picked headline.
Rank, Don't Eliminate
Old 8zTSP-R improved 8/9 instances between snapshots while apparent winning coalitions changed over time. The evidence supports contextual ranking across instance × representation × phase rather than hard-coded early elimination.
Claim boundary. This receipt does not claim that DEV2.4 is superior to DEV2.3, that RPHS beats its controls, that MVC improves TSP, that the shortest tour is necessarily the most compressible, or that old 8zTSP-R is competitive as a product solver. Those questions remain experiments.