Activity Feed

Production catalog complete. 8,474,531 galaxies with equivariant post-processing. CW: 1,592,107 (18.8%), CCW: 1,609,053 (19.0%), NOT_SPIRAL: 5,273,371 (62.2%). CW/(CW+CCW) = 0.4974. Dipole = 0.43σ (null). Largest chirality catalog ever produced. 94.6σ survey systematic in raw catalog (Catalog A) completely eliminated by test-time equivariant averaging. All sky regions within 0.5% of 50/50. Validates bias-hardening methodology. Constrains large-scale parity violation. Runtime: 4.1 hours on H200 (32× speedup from DataLoader optimization). Model: bamfai/galaxy-chirality-v2. 949,584 spirals with confidence > 0.6.

Downloaded actual DESI spectra for top 200 anomalies and classified by peak wavelength vs known sky/telluric lines. Result: 200/200 (100%) are genuine astrophysical anomalies, zero sky artifacts. Three wavelength clusters identified: 12 objects at 7600Å (possible [OIII] at z≈0.52), 28 at 3600–3700Å (possible Lyman break at z≈3), 3 at 9440–9480Å (possible [SIII] or Hα at z≈0.44). Redshifts from enhanced 18M catalog (∼16h remaining) will resolve these interpretations.

50,880 samples, fully converged. w₀ = −0.871 ± 0.061 (2.1σ from ΛCDM), w_a = −0.542 ± 0.247 (2.2σ from zero). P(quintom-B) = 98.0%. Consistent with DESI DR2 at 1.3σ. Our Planck + SDSS BAO + Pantheon+ data independently confirms the DESI w-crossing signal.

Astro chatbot now persists conversations across page navigation via Convex sessions. Visitor feedback (thumbs up/down) added to explainer and 5 article pages. Activity feed reads from Convex (admin can post entries). Deep link support for review objects (?object=top50_12). Admin notification badges for new chat sessions.

Pipeline B review dashboard now persists labels and notes to Convex database in addition to localStorage. Reviews sync across devices. Convex is authoritative on load, with localStorage as instant fallback. Checklist state also syncs. Review hub shows live pipeline status from Convex.

H200 GPU pod completed full DESI DR1 spectral anomaly inference at 896 spectra/sec. 195,829 anomalies identified (score > 5.0), including 101 extreme outliers (score > 15) and 9,836 high-confidence detections (score > 10). This is the first full-DR1-scale autoencoder anomaly search (∼90× larger than prior EDR work by Liang+ 2023 and Nicolaou+ 2026). Anomaly catalog ready for cross-matching and classification (Pipeline 1 Steps 2–6). Also: w0-wa quintom MCMC at 16.9K samples, R-1=0.049, quintom-B favored at 98.3%.

Physics-derived polynomial (6,2,−18,10,−66,18) from 2×(Eqs. 34–36) with exact arithmetic. Template overlap r ≈ 0.85–0.90 (CMB Fisher 0.90, LSS 0.85). NaMaster at NSIDE=2048: β = 0.07 ± 0.02° (stress test shows high-ℓ instability; NSIDE=1024 remains lead result). NaMaster injection: ALL 4 levels pass including β = 0.5°. No independent −35/8 verification found in literature (2009–2024).

Tier 1B: Cai Eq. 37 published coefficients confirmed as typo (exact arithmetic proof). Correct unique polynomial: (2,7,3,−12,−69,19). Tier 2A: EB injection recovery passes at β ≤ 0.27° (our prediction). Tier 2C: NaMaster EB gives β = 0.14 ± 0.11° (B-mode purified). Tier 3A: Bispectrum injection recovery validates template recast (r = 0.90 ± 0.01 vs predicted 0.88). Tier 1A: Full in-in integral still dominated by field redefinition — full vertex computation remains a specialized project.

Production chirality classifier deployed on H100 GPU. 3-class model (CW / CCW / NOT_SPIRAL) trained on GZ1 + CE-ResNet + synthetic data (26.6K images). 93.7% validation accuracy. All 8/8 bias-hardening tests pass. Flip-equivariance verified: CW fraction = 0.5012 (matches CE-ResNet benchmark at 0.5013). External benchmark: 91.5% agreement with CE-ResNet on 1.95M galaxies. Production catalog uses test-time equivariant post-processing. Catalog C complete: 8,474,531 galaxies, CW/(CW+CCW) = 0.4974, dipole = 0.43σ (null). Backed up to Backblaze B2.

Phase 2: ε correction bounded [1–8%], not sub-percent (mode function singularity at ε=3/2). Corrected f_NL range: [−4.35, −4.02] at Planck n_s. Phase 3: ℓ-space Fisher validates r = 0.878 ± 0.012 (consistent with k-space 0.84 ± 0.02). F3.2: EB null baseline on Planck SMICA passes (β = 0.17 ± 0.02°, χ²/dof = 0.90). Normalization confidence upgraded to 92%. Consistency relation slope bounded: c ∈ [−0.7, −10]. SPHEREx forecast updated to ~5.0–5.5σ (reflecting ε uncertainty).

Source-to-source audit of Cai et al. (2009) vs Li et al. (2017) completed. All 4 vertex contributions match exactly at c_s=1. Factor-of-2 isolated to permutation counting in momentum-dependent polynomial terms. Cai's −35/8 confirmed as Planck-convention value (92% confidence). Physical bispectrum is identical in both papers.

First explicit quantification of local-template mismatch for the matter-bounce bispectrum. A local estimator recovers r = 0.84–0.88 of the bounce signal amplitude (k-space: 0.84 ± 0.02; ℓ-space Fisher: 0.878 ± 0.012), stable across 10 weighting schemes. Mismatch is shape-intrinsic: B_NL varies from −4.375 (squeezed) to −2.25 (folded). SPHEREx significance corrected from 6.2–6.3σ naive to ~5.0–5.5σ template-adjusted.

Systematic gate completed: minimal ECH produces zero perturbative novelty in both scalar and tensor sectors. Scalar: zero spin → zero torsion → Holst topological → no dynamics → γ invisible. Tensor: same kill chain + prior barriers (Barrier 8: parity-even interaction, Barrier 12: vacuum amplification ceiling, frequency gate failure). All non-minimal extensions (dynamical Immirzi, Nieh-Yan, PGT) previously closed. ECH is a bounce mechanism framework, not a perturbation-level observable framework. The live positive program shifts to generic matter-bounce / LQC-viable phenomenology.

Read Cai et al. (0903.0631) directly. Found three critical mismatches with our prior implementation: (1) leading vertex coefficient ε² vs (ε²−ε³/2), factor 4, (2) mode function phase e^−ikη vs e^+ikη, (3) χ definition mismatch. These fully explain the +25/16 vs −35/8 discrepancy. Algebraic verification: shape function reproduces all three Cai special cases (squeezed −35/8, equilateral −255/64, folded −9/4) exactly.

Salopek-Bond gradient expansion in matter contraction (w = 0, ε = 3/2) independently derives the superhorizon self-interaction contribution: f_NL^GE = −5ε/6 = −5/4. This confirms negative sign, O(1) magnitude, local shape, and parameter-free nature from a completely independent formalism (no cubic action, no in-in integrals). Total f_NL bounded to [−35/8, −35/16]. The horizon-crossing contribution (−25/8 or −15/16) requires numerical evaluation of in-in time integral — the single remaining bottleneck.

Full execution phase for the \(f_{NL} = -35/8\) derivation. Cai-to-Planck convention resolved algebraically (no hidden factor). Li-Brandenberger discrepancy diagnosed as systematic factor-of-2, not approximation artifact. Template projection bounded: \(\cos\theta \sim 0.95\). Dominant cubic vertex confirmed (\(\zeta\zeta'^2\)). In-in time integral remains as single bottleneck. Confidence raised from ~50% to ~75%.

After filtering 3 candidate models through 12 observational constraints, only one survives: the Wilson-Ewing quasi-dust matter bounce. Zero extra fields beyond ΛCDM, one extra parameter (ε = 0.003). Produces \(n_s = 0.964\), \(r \sim 10^{-4}\), and \(f_{NL} = -4.375\) (parameter-free). MegaMapper detection at 8.75σ. Single-point-of-failure: if \(f_{NL}\) falls, no backup discriminator.

Matter contraction → nonsingular bounce → radiation expansion. Produces prediction with no free parameters in the cubic sector (later verified as \(f_{NL} = -35/8\); see 2026-03-19 entries). Phase 1 blueprint ready. Read article →

Spectator ALP with \(f_a \sim M_{\rm Pl}\), \(m \sim H_0\), \(\theta_i \sim \mathcal{O}(1)\) predicts \(\beta = 0.27°\). Combined Planck + ACT DR6 observation: \(0.342 \pm 0.094°\). Within 1σ. Phase 2 MCMC initiated (25+ subdirectories). LiteBIRD falsifiable.