|
# Phase 2 update: flopscope v0.12.3, deadline extension, and final submission nomination
|
|
3
|
105
|
October 24, 2026
|
|
[Phase 2 write-up] A third-cumulant chain at its boundary: random projections, a skew law, and where the next order's accuracy lives (submission #333534)
|
|
3
|
47
|
October 24, 2026
|
|
[Phase 2 write-up] A moment chain with a learned cross-layer corrector and a wall-time guard (submission #337436)
|
|
0
|
5
|
October 24, 2026
|
|
[Phase 2 write-up] A learned corrector that reads the engine's cross-neuron terms, saturation-aware compute and a free fourth-cumulant fold on factored cumulant propagation at width 1024 (submission #336955)
|
|
3
|
2
|
October 24, 2026
|
|
[Phase 2 write-up] Sphere-frame cumulant propagation with age-structured κ₃ sources, and exact symmetry as a diagnostic of a tuned closure (1.50e-9, submission #337205)
|
|
3
|
12
|
October 24, 2026
|
|
Good game well played
|
|
13
|
60
|
October 24, 2026
|
|
[Phase 2 write-up] A bias/variance error budget for factored K=3 cumulant propagation: what the slices can tolerate, measured (submission #337475)
|
|
1
|
31
|
October 24, 2026
|
|
[Phase 2 write-up] How errors survive depth in cumulant propagation: a fit-free survival law, checked on ARC's reference chain, and four flopscope issues (submission #337350)
|
|
0
|
9
|
October 24, 2026
|
|
How the hidden-set evaluation invokes whest run
|
|
5
|
113
|
October 23, 2026
|
|
Notes on the Problem (Phase 1)
|
|
3
|
139
|
October 23, 2026
|
|
[Phase 2 write-up] A cost-engineered factorized K=3 chain + transferred LT2 corrector (3.82e-9) (Submission #337389)
|
|
1
|
23
|
October 23, 2026
|
|
[Phase 2 write-up] What noise-corrected oracles show about a cumulant-propagation estimator (submission #337578)
|
|
0
|
17
|
October 22, 2026
|
|
[Phase 2 write-up] The (3,1) slice of κ₄: where V29 loses accuracy, and a two-term closure confirmed in a pre-registered test on 77 held-out networks (submission #334821)
|
|
3
|
71
|
October 21, 2026
|
|
[Phase 2 write-up] Zero-Sample K=4 via Factored n^3-Memory Transport at 1024 x 16: 11.8x below augmented K=3, why memoryless closures stall, #1 write-up (#336537)
|
|
3
|
67
|
October 21, 2026
|
|
[Phase 2 write-up] A bias/variance error budget for factored K=3 cumulant propagation: what the slices can tolerate, measured (submission #337475)
|
|
0
|
19
|
October 21, 2026
|
|
[Phase 2 write-up] Removing the global scale mode makes a carried κ4 core pay: extraction on/off ×0.380 on the grader inside e30_a's engine (submission #336317)
|
|
1
|
52
|
October 20, 2026
|
|
[Phase 2 write-up] Cumulant propagation under a compute budget: state-matched corrections, verified fast arithmetic, and a learned terminal corrector (submission #335413)
|
|
0
|
35
|
October 19, 2026
|
|
[Phase 2 write-up] Where the per-source leg representation of factorized K=3 ends (adj 3.8e-9), and the measurements behind it (submission #336225)
|
|
0
|
30
|
October 18, 2026
|
|
Approximate cumulant propagation with exact compute reductions
|
|
0
|
45
|
October 16, 2026
|
|
Flopscope: fix pull requests, with tests, for the open billing issues
|
|
0
|
50
|
October 12, 2026
|
|
Eligibility of solutions built on another participant's MIT-licensed forum code (§4 / §6)
|
|
1
|
112
|
October 5, 2026
|
|
Everything we tried: a factorized K=3 cumulant-propagation estimator at 0.25 x B, where its FLOPs go, and 25 measured dead ends (team 504aldo, rank ~10)
|
|
5
|
358
|
October 4, 2026
|
|
Phase 1 final evaluation question: Overfitting to public instances
|
|
5
|
856
|
October 4, 2026
|
|
Leaderboard vs Overview leading scores
|
|
0
|
84
|
October 3, 2026
|
|
Phase 2 of the ARC White-Box Estimation Challenge is live 🚀
|
|
3
|
320
|
October 1, 2026
|
|
[Phase 1 write-up] Fixed grid of directions instead of random sampling (submission #320802) and five things that did not work
|
|
1
|
369
|
September 22, 2026
|
|
[Phase 2] The compute budget is not a lever, and the scoring rule is why: `adjusted = mse x max(0.1, util)`
|
|
0
|
119
|
September 17, 2026
|
|
Errors are leaking information
|
|
1
|
152
|
September 9, 2026
|
|
[Phase 2 write-up] Zero fitted constants, measured: what the 64 calibration numbers in a ReLU-MLP mean propagator are actually absorbing (submission #330001)
|
|
0
|
125
|
September 6, 2026
|
|
Phase 2 write-up: exact bivariate ReLU covariance plus per-layer calibration, 3x over the kit baseline, with negative results (submission 329951)
|
|
0
|
106
|
September 5, 2026
|