TSPES asks whether the mature family can solve standard instances extremely well.
Why six?
Because one leaderboard would flatten different scientific roles. A strong product solver, a plateau donor, a hard-tail successor, a falsification arena and two native research systems should not be judged by one scalar.
DEV2.3 and DEV2.4 ask whether architectural changes actually beat the proven donor.
RPHS and both 8zTSP-R branches test mechanisms that may matter only at particular scales, phases or representations.
The six lines
Current numbers below are evidence-bound to the verified 12 September Home/Work extracts. Historical numbers are explicitly marked as such.
1 · TSPES v0.1.4.2 W6 RUNNING · FLAGSHIP
Mature product / solver research
The strongest current solver line. Exact optimum has already been reached on wi29, dj38, qa194, uy734 and zi929. Active lu980 = 11,355 versus optimum 11,340: only 15 units remain. Current stage is TEP generation 1 / epoch 2.
5 / 9 exacttracked benchmark ladder
0.132275%current lu980 gap
W6workers
CPU + Rust + GPUactual hybrid evidence
2 · DEV2.3 RUNNING · PLATEAU
Proven nu3496 hard-tail donor
The continuous donor remains at 96,780 on nu3496, 648 above the 96,132 optimum. The current extract shows cycle 19 / P7 Rank-Don't-Eliminate alternative 3opt45 basin. The campaign incumbent remains 96,780; P7 itself was at 96,908 at the snapshot, so the incumbent is not attributed to P7.
96,780best length
0.674073%gap
P7current branch
9.38%current variant progress
3 · DEV2.4 HF3 RUNNING · TIED
Hard-tail successor / control
DEV2.4 has physically exercised the strict hybrid GPU+Rust stack and reached the same 96,780 as DEV2.3 after 374,483 global proposals. That tie is useful evidence, but it is not superiority.
96,780best length
374,483global proposals
PASSindependent validation
RTX 3080 + Rustactual backend path
4 · BD-RPHS v0.1.2 W4 RUNNING · SATURATED
Representation / sensor falsification line
Active uy734 W4 is 79,374 / 79,114 = 0.328640%, with the best dating back to about cycle 5 despite the campaign now reaching cycle 120 / 966 completed chunks. A historical W6 snapshot reached a better 79,288 / 0.219936%, but the winner was the N2_shuffled_reverse null control—a result we keep visible rather than hide.
79,374active W4 best
cycle 120966 completed chunks
79,288historical W6 best
null controlhistorical winner
5 · 8zTSP-R v0.1.1 R2/HF1 RUNNING · RESEARCH
Pure relational / native discovery
This line is not a competitive product solver. Its value is mechanistic: all 9 of 9 instances now have committed snapshots; the latest completed visit was mu1979 round 37 and the campaign has advanced to nu3496. The apparent useful representation families remain contextual rather than universal. That is direct evidence against hard-coding one universal winner too early.
9 / 9committed instance snapshots
round 37latest completed mu1979 visit
nu3496current campaign instance
Rank, Don't Eliminatesupported research lesson
6 · 8zTSP-R v0.2.0 HF2 MVC RUNNING · EARLY
Multi-View Compression Native
The newest line asks the founding compression question most directly: can multiple description-complexity views become useful search coordinates? It has eight matched lanes including legacy compression, CMVC, combined views, compression-blind and native-distance controls. The fresh committed snapshot is at generation 2 with 8/64 proposals and best length 43,487. No full visit has completed yet, so no performance-advantage claim is made.
8 lanesmatched compression/control design
wi29startup instance
gpu-strictrequested backend
gen 2 · 8/64committed proposals
What is strongest today? TSPES is the strongest demonstrated solver line. DEV2.3/DEV2.4 carry the strongest nu3496 hard-tail evidence. The old 8zTSP-R line carries the strongest current methodological signal about late/contextual value. The new MVC line carries the highest uncertainty—and therefore the highest near-term information value.
Claim discipline. A shared 96,780 incumbent does not make DEV2.4 an independent algorithmic replication of DEV2.3. RPHS's historical 79,288 must be shown with its null-control winner. The old 8zTSP-R line must not be marketed as competitive TSP solving. MVC now has fresh committed run behavior, but remains unranked until completed visits support a fair comparison.