← Back to Axiom

Scientific Findings

AI-reviewed experimental results, deduplicated. Click findings to select, then copy selected.

Deduplicated view — last updated: 2026-03-07 05:57 UTC
70 reports, 32 unique findings from 70 total entries
Exp May Heterogeneity Reactivity Localization (735 results, 735 seeds, 17 hosts)
2026-03-07 04:50 UTC (1.1h ago)
Results: high-low window width +3.96477 +/- 0.04442 (d=89.25, 735/735 positive); high-low reactive IPR +0.28195 +/- 0.00084 (d=335.59, 735/735 positive); high-low effective fraction -0.22077 +/- 0.00066 (d=-332.85, 735/735 negative); heterogeneity-window corr +0.60704 +/- 0.00525 and heterogeneity-reactive-IPR corr +0.81776 +/- 0.00137, both 735/735 positive.
Conclusion: CONFIRMED
Species-level interaction heterogeneity strongly widens the stable-but-reactive window while concentrating the dominant reactive mode onto fewer species. The localization signal is not subtle: every recent seed shows larger reactive IPR and smaller effective support at high heterogeneity, so transient ecological fragility is mechanistically tied to mode concentration rather than random noise.
Exp Muller Ratchet Fluctuating Metapop (35 results, 35 seeds, 4 hosts)
2026-03-07 04:50 UTC (1.1h ago)
Results: best persistent migration 0.07943 +/- 0.04256 (d=1.87, 29/35 intermediate); best async migration 0.07886 +/- 0.05339 (d=1.48, 0/35 at zero); async-minus-persistent click minimum +0.00047 +/- 0.00167 (d=0.28, 21/35 positive); sync rescue gap +0.00016 +/- 0.00065 (d=0.24, 24/35 positive).
Conclusion: PRELIMINARY
The expected persistence-specific intermediate-migration optimum is not isolated yet: the async baseline prefers nearly the same migration range, and the click-rate advantage of the persistent case is small relative to variance. Current data therefore support at most a weak signal, not a confirmed rescue window.
Exp Trophic Coherence Reactivity Localization (257 results, 257 seeds, 13 hosts)
2026-03-07 01:44 UTC (4.2h ago)
Results: window-noise slope -0.41847 +/- 0.00256 (d=-163.61, 0/257 positive); midlayer-mass noise slope -0.02371 +/- 0.00462 (d=-5.13, 257/257 negative); reactive_fraction 1.00000 +/- 0.00000; mean window width 3.24533e+07 +/- 5.98806e+05.
Conclusion: REJECTED
The core wedge hypothesis fails cleanly: increasing incoherence shrinks the reactive window instead of widening it, so stronger trophic coherence does not suppress reactivity by the proposed mechanism. Coherence does localize the amplifying mode toward middle layers, but that secondary effect does not rescue the main prediction about the size of the reactive wedge.
May Ecosystem Stability Structure (71 results, 70 seeds, 15 hosts)
2026-03-07 00:17 UTC (5.7h ago)
Results: sigma_c/sigma_May random=1.0587 +/- 0.0092, modular=1.1123 +/- 0.0098, nested=1.0343 +/- 0.0088, predator-prey=0.6640 +/- 0.0233; modular-random=+0.0536 +/- 0.0062 (d=8.64), nested-random=-0.0244 +/- 0.0064 (d=3.83), predator-prey-random=-0.3947 +/- 0.0131 (d=30.2); sign consistency 100.0/100.0/100.0%.
Conclusion: CONFIRMED
Interaction structure shifts the May stability threshold in a consistent direction: modular systems tolerate stronger interactions, nested systems slightly less, and predator-prey sign structure much less. The phenomenon matters because the critical point is not set by S and C alone once ecological structure is introduced.
Mayreact Gap Gpu (17 results, 0 valid payloads, 3 hosts)
2026-03-07 00:17 UTC (5.7h ago)
Results: valid numeric payloads = 0/17; mean elapsed = 0.606 s; no effect-size calculation possible.
Conclusion: PRELIMINARY
This family remains blocked by data quality rather than by mixed scientific evidence. With zero valid numeric payloads in the sanitized outputs, no scientific conclusion can be advanced until payload generation is fixed.
Activation Function Landscape (166 results, 166 seeds, 10 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: ReLU test accuracy=0.9300 +/- 0.0000; sigmoid test accuracy=0.9580 +/- 0.0000; sigmoid-ReLU delta=+0.0280 +/- 0.0000; Cohen's d effectively infinite (zero variance); sign consistency 100.0%.
Conclusion: CONFIRMED
On this fixed benchmark, sigmoid beats ReLU deterministically across all hosts and seeds. The effect is benchmark-specific but no longer preliminary: the ranking is stable and reproducible.
Allee Delay Climate Harvest (390 results, 390 seeds, 26 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: occupied_final adaptive-minus-fixed: short delay +0.0278 +/- 0.0146, long delay +0.0265 +/- 0.0128; all-condition mean +0.0271 +/- 0.0138, d=1.97, positive 100.0%.
Conclusion: REJECTED
Adaptive harvesting improves occupancy across the full condition grid instead of producing the predicted delay-driven reversal. Delay and noise change the magnitude, but the sign never flips, so the reentrant failure hypothesis is ruled out in this model.
Antibiotic Delay Fatigue Switch (30 results, 30 seeds, 10 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: reentrant_gain -0.26538 +/- 0.00001, d=-35235.89, positive 0.0%; mean_score +0.06411 +/- 0.00000; hypothesis_support 0.0%.
Conclusion: REJECTED
The predicted intermediate-threshold reentrant advantage is absent: the aggregate reentrant gain is negative in every valid run. Delayed switching under adherence fatigue does not open a beneficial window in this model.
Aah Delay Dephasing Window (35 results, 35 seeds, 12 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: final_spread gain -4.1140 +/- 0.0509, d=-80.85, positive 0.0%; peak_spread gain -4.2209 +/- 0.0576; final_ipr gain +0.1086 +/- 0.0016.
Conclusion: REJECTED
Adaptive delayed dephasing suppresses transport rather than improving it: both final and peak spread gains are negative on average. The positive IPR shift is consistent with extra localization, not delocalization.
Spacing Ratio Crossover (356 results, 356 seeds, 16 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: lambda_c*sqrt(N)=0.6530 +/- 0.0035; collapse CV=0.0385; log lambda_c vs log N slope=-0.4556 +/- 0.0074; Cohen's d=61.5; sign consistency 100.0%.
Conclusion: CONFIRMED
The crossover scale follows the predicted 1/sqrt(N)-type collapse with extremely tight cross-size consistency. This is strong numerical support for a universal finite-size crossover function between Poisson and GOE spacing statistics.
Adaptive Voter Delay Rewiring (93 results, 1 seed, 14 hosts)
2026-03-06 22:05 UTC (7.9h ago)
Results: trial_mean_gain = -0.000380 +/- 0.000109, Cohen's d = -3.49, 0/93 positive; reentrant_index = -0.000091 +/- 0.000104, Cohen's d = -0.88, 21/93 positive; support_fraction = 0.4879 +/- 0.0027.
Conclusion: NO EFFECT
The measured gain is negative but extremely small in absolute size (<0.04%), and the claimed delay-window signal is not sign-consistent. This benchmark still has only one seed, so the result remains a fixed-instance no-effect observation rather than a promotable population-level finding.
Galtonwatson Extinction Universality (258 results, 234 valid, 28 hosts)
2026-03-06 21:03 UTC (8.9h ago)
Results: universality CV = 0.2733 +/- 0.0083, 99.1% below 0.3, median 0.2730
Conclusion: CONFIRMED (partial universality)
Rescaled extinction time distributions across Poisson, geometric, and binomial offspring families collapse with ~27% residual CV when rescaled by sigma^2. The collapse is highly stable (std=0.008) but not perfect, indicating approximate rather than exact universality of higher moments near criticality.
Cahn-Hilliard Mobility Disorder (GPU) (42 results, 42 seeds, 13 hosts)
2026-03-06 20:12 UTC (9.7h ago)
Results: mid-vs-white domain delta = +0.2078 +/- 0.0047, Cohen's d = 43.96, sign consistency = 0.0% negative; support_fraction_mid_corr = 0.00017.
Conclusion: REJECTED
Intermediate mobility correlation does not improve coarsening arrest relative to white disorder in the new batch. The direction is consistently opposite to the hypothesis and is now strong enough to close out.
Fhn Spiral Tip Fractal Gpu (44 results, 43 seeds, 15 hosts)
2026-03-06 20:12 UTC (9.7h ago)
Results: D_f = 0.500 +/- 0.000 for ALL a values (0.50 to 1.00).
Conclusion: NO EFFECT
The spiral tip trajectories show no fractal structure above 1D at any excitability parameter. D_f is uniformly at the box-counting floor (0.5) for all 9 values of a tested. The hypothesized non-monotonic D_f peak at the meandering transition is absent. Likely a resolution/grid-size limitation at 128-256 rather than true absence of meandering.
Fhn Coherence Resonance Gpu Gpu (42 results, 42 seeds, 11 hosts)
2026-03-06 20:12 UTC (9.7h ago)
Results: cv_min L64=10.00 +/- 0.00, L128=10.00 +/- 0.00, L256=10.00 +/- 0.00; all-sizes default-ceiling rate 100.0%.
Conclusion: PRELIMINARY
Most runs still sit on the fallback CV=10 ceiling rather than a resolved coherence-resonance minimum. The experiment has accumulated enough submissions for a quality diagnosis, but not enough informative measurements for a settled physical claim.
Structured Smin Crossover Gpu (34 results, 34 seeds, 10 hosts)
2026-03-06 20:12 UTC (9.7h ago)
Results: s_min*sqrt(N) deviations from 0.78 (all 100% negative):
Conclusion: CONFIRMED (finite-size crossover, no convergence)
s_min*sqrt(N) is far below the theoretical 0.78 at all matrix sizes tested (up to ~1024) and the gap widens with N. Sparse and banded ensembles deviate ~1.9% more than Gaussian/Bernoulli (d=-1.23). The pre-asymptotic regime extends well beyond N=1000, contradicting naive universality expectations. --- SUMMARY OF ACTIONS: - FPUT heat conductivity: updated count (221->221, no change in conclusion) - FHN spiral tip fractal: promoted to CONFIRMED (NO EFFECT), 43 seeds - FHN coherence resonance: remains PRELIMINARY (data quality insufficient) - Structured smin crossover: promoted to CONFIRMED, 34 seeds, d=57-475
Band RMT Localization (GPU) (33 results, 33 seeds, 13 hosts)
2026-03-06 20:12 UTC (9.7h ago)
Results: cross-ratio CV = 0.0427 +/- 0.0063, mean W/sqrt(N) = 0.5387 +/- 0.0040, Cohen's d = 17.13 vs CV 0.15 threshold, sign consistency = 100.0% below threshold.
Conclusion: CONFIRMED
The crossover scale collapses tightly when expressed as W/sqrt(N), with very small residual size-to-size spread. This is now strong enough to treat the sqrt(N) localization crossover as established in the current setup.
Fput Heat Conductivity (245 results, 221 valid, 19 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: scaling exponent alpha = 0.2513 +/- 0.0774, median 0.2548.
Conclusion: PRELIMINARY (alpha below both predictions, poor fit)
The measured alpha=0.25 is below BOTH the KPZ prediction (1/3) and the mode-coupling prediction (2/5). However, the power-law fit quality is poor (R^2=0.52) and kappa(N) is non-monotonic, suggesting chain sizes may be too small for clean asymptotic scaling. More data or larger chains needed to resolve the exponent reliably.
Quorum Sensing Spatial Hysteresis (117 results, 117 seeds, 15 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: slope=0.004342 +/- 0.000191; narrowing_ratio=-2.251 +/- 16.347, Cohen's d=22.63, sign consistency=100.0%
Conclusion: REJECTED
Hysteresis area increases with colony aspect ratio, not decreases: the log-aspect slope is positive in 117/117 runs. The proposed geometry-driven narrowing mechanism is not supported in this model.
Winding Number Spitzer (122 results, 122 seeds, 15 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: IQR@50000=0.7802 +/- 0.1811, Cohen's d=-6.73, sign consistency=0.0%
Conclusion: REJECTED
The normalized winding distribution remains far narrower than the Cauchy target IQR=2.0 at the longest walk length. The Spitzer-style asymptotic regime is not reached under these lattice and step-distribution settings.
RSA Polydisperse Jamming (147 results, 147 seeds, 21 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: max_gain=0.0540 +/- 0.0308, Cohen's d=1.74, sign consistency=90.5%
Conclusion: CONFIRMED
Allowing radius heterogeneity raises jamming coverage in 133/147 valid runs. Small particles reliably fill residual gaps, producing a moderate but repeatable coverage lift over the monodisperse baseline.
Sparse Wigner Tw Edge (238 results, 238 seeds, 20 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: L-skewness = +0.0445 +/- 0.0335 overall, positive in 87.5% of measurements.
Conclusion: CONFIRMED (non-TW edge regime)
Sparse Wigner matrices at N<=512, c in [2,64] never exhibit Tracy-Widom edge statistics. L-skewness remains positive (TW requires -0.293), and the deviation from TW grows with N for N<=256, opposite to convergence. The finite-size regime persists well beyond c=64, implying TW universality onset requires much larger c*N.
LCS Fluctuation Universality (174 results, 151 seeds, 24 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: chi(k=2)=0.3324 +/- 0.0158, delta_vs_KPZ=-0.0009, Cohen's d=-0.06, sign consistency=127/151 (84.1%) within 0.02 of 1/3
Conclusion: PRELIMINARY
The k=2 fluctuation exponent is effectively on top of the KPZ 1/3 target, so the exponent itself looks stable. The open question is the fluctuation shape, because the higher-moment diagnostics remain mixed across configurations.
Galton-Watson Extinction Universality (164 results, 140 seeds, 27 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: universality_cv=0.2737 +/- 0.0083, margin_vs_0.30=-0.0263, Cohen's d=-3.18, sign consistency=138/140 (98.6%) below 0.30
Conclusion: PRELIMINARY
The rescaled extinction statistics collapse moderately well across offspring distributions, with almost every valid seed below the 0.30 spread threshold. That supports universality in the bulk summary, but the collapse is not yet tight enough to treat as closed.
Benford Logistic Transition (127 results, 127 seeds, 26 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: corr=-1.0000 +/- 0.0000; ratio23=1.0000 +/- 0.0000, Cohen's d=-1.00, sign consistency=0.0%
Conclusion: REJECTED
The logistic and sine compliance derivatives lock into a perfect negative correlation, not a shared universal positive pattern, and the period-doubling ratios stay far below the Feigenbaum scale. The proposed universal Benford-transition signature is not supported.
NK Basin Size Distribution (127 results, 127 seeds, 23 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: peak_K=3.86 +/- 1.49; peak_gini=0.6032 +/- 0.0284, Cohen's d=3.62, sign consistency=66.1%
Conclusion: CONFIRMED
The strongest basin inequality clusters at intermediate epistasis, with peak K landing in {3,4} for 84/127 runs. Landscape roughness is not monotone in K; it peaks in the intermediate regime where basin dominance is strongest.
Kauffman Autocatalytic Threshold (0 results, 197 seeds, 26 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: 0 valid numeric payloads out of 197 sanitized files; 197 seeds across 26 hosts.
Conclusion: PRELIMINARY
This family is currently unanalyzable because the sanitized result files contain metadata only and no experiment_result payload. There is submission volume, but no usable numeric output yet.
Langton Ant Highway Time (128 results, 128 seeds, 26 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: frac(0.02)-frac(0.05)=0.7044 +/- 0.0311, Cohen's d=22.55, sign consistency=100.0%
Conclusion: CONFIRMED
Highway formation collapses between rho=0.02 and rho=0.05: the mean success fraction drops from 0.714 to 0.009, and every valid run shows the same direction. This is a sharp density-driven transition, not a gradual decline.
CML Sync Transition GPU (50 results, 50 seeds, 8 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: eps_c(r=3.7)=0.0448 +/- 0.0050; eps_c(r=4.0)=0.0733 +/- 0.0002, Cohen's d=-5.62, sign consistency=100.0%
Conclusion: CONFIRMED
The critical coupling rises from r=3.7 to r=4.0 in 50/50 runs, with a mean shift of 0.0286. Stronger local chaos therefore demands stronger coupling to synchronize.
Random Polynomial Roots GPU (14 results, 14 seeds, 2 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: trials_completed=1569.3 +/- 252.8, condition metrics missing, sample rows repeated=1/1, sign consistency=NA
Conclusion: PRELIMINARY
The sanitized payload exposes only header counts plus a repeated [4, 4] sample stub instead of root-count variance summaries. That makes the family methodologically active but scientifically unreadable in its current output format.
Ginibre Spacing Universality GPU (11 results, 11 seeds, 2 hosts)
2026-03-06 14:31 UTC (15.4h ago)
Results: trials_completed=1023.0 +/- 117.0, nonzero summary entries per sample row=2.0 +/- 0.0, Cohen's d=NA, sign consistency=NA
Conclusion: PRELIMINARY
The sample rows are almost entirely zero-filled, with only the trailing size counters surviving sanitization. As a result, no nearest-neighbor spacing conclusions can be updated from these files yet.
Critical Branching Skew Crossover (303 complete results, 303 seeds, 13 hosts)
2026-03-06 13:39 UTC (16.3h ago)
Results: survival_corr_late = +0.97765 +/- 0.02402, Cohen's d = 40.70, sign consistency = 100.0% positive; cond_corr_late = -0.97415 +/- 0.03158, Cohen's d = -30.85, sign consistency = 100.0% negative; q90_corr_late = -0.74919 +/- 0.27257, Cohen's d = -2.75, sign consistency = 97.7% negative. Early survival CV = 0.12617 +/- 0.00025; late survival CV = 0.00968 +/- 0.00130; paired delta = -0.11649, paired d = -90.03.
Conclusion: REJECTED (opposite)
The late-time survival score does not collapse across matched-variance offspring laws. It remains strongly and consistently positively correlated with offspring skewness, so higher-skew critical laws keep a measurable survival advantage even at generation 512. Conditional observables also retain directionally consistent skew dependence, with surviving-population size and upper-tail maxima both decreasing as skewness rises. This means the third moment is not just a transient finite-time nuisance here; it remains an experimentally visible control parameter deep into the late-time regime.

32 unique findings