bigbounce/spin-torsion cosmology research program
research live

Novelty accounting

contributions

Each contribution is scored on the canonical four-tier novelty scale. Our self-claim ceiling is N3 (first-of-kind demonstration). N4 — paradigm-shifting / Nobel-worthy — is intentionally reserved for outside arbiters and is never self-claimed. Definitions, prior work, and verification links below.

How the six papers fit together

One program, two halves. The theory arm (P1A, P1B, P2) asks where a nonsingular Einstein–Cartan–Holst bounce could leave a falsifiable fingerprint, proves the bounce mechanism itself is invisible to telescopes, closes the enumerated dark-energy routes, and isolates the one surviving handle — a parameter-free matter-bounce non-Gaussianity SPHEREx can test. The data arm (P3, P4, P5) mines 45M+ archival sources for the parity- and anomaly-level signatures any bounce would have to imprint, and reports honest nulls with quantified falsification windows. Negative theory results narrowed the search space; the surveys then went looking exactly where the theory said to look.

P1AECH theory + no-go: perturbation-transparency theorem and a 14-constraint channel-level closure of the four minimal bounce→dark-energy routes.
P1BMCMC + pipeline companion: frozen ΛCDM+ΔN_eff chains (honest null), NaMaster recovery, and an ALP-birefringence consistency check.
P2f_NL = −35/16 forecast: the surviving falsifiable handle — matter-bounce non-Gaussianity, ~300× inflation and opposite sign, testable by SPHEREx at 3–5σ.
P3Multi-survey anomaly catalog: 378,280 unique anomalies across 7 surveys + a NANOGrav free-spectrum slope consistent with matter-bounce γ = 3.0.
P4Galaxy chirality null: 8.47M classified galaxies, a null +0.41σ real-space dipole, refuting the claimed ~3% parity signal at scale.
P5DESI chirality × environment null: spiral handedness is independent of cosmic-web environment, constraining environment-coupled parity models.

Program arc: P1A (ECH theory / no-go) → P1B (MCMC + pipeline companion) → P2 (f_NL forecast) → P3 (anomaly catalog) + P4 (chirality null) + P5 (DESI chirality × environment). See the six papers for full status and PDFs.

Novelty tier scale
N4PARADIGM-SHIFTING (RESERVED)Paradigm-shifting / consensus-breaking / Nobel-worthy. Awarded by the field over time — never self-claimed by the authors.reserved
N3FIRST-OF-KIND DEMONSTRATIONFirst-of-kind demonstration, new constraint, or new direction that meaningfully opens or closes a measurable testbed. Ceiling for our self-claims.7 items
N2NOVEL COMBINATION / EXTENSIONNovel application or combination of existing methods — applying a known technique to a regime it hasn't seen, or combining pipelines to broaden what's measurable.11 items
N1INCREMENTAL REFINEMENT / REPLICATIONIncremental refinement or replication — tightens an existing measurement, fixes a known bias, or reproduces a prior result with a new dataset.0 items
N4

PARADIGM-SHIFTING (RESERVED)

Intentionally empty. N4 is reserved for paradigm-shifting, consensus-breaking, or Nobel-worthy results. That tier is awarded by the field over time — through broad community replication, citation, and consensus — not by the authors. We cap our own claims at N3 (first-of-kind demonstration / new constraint / new direction) and let outside arbiters raise the ceiling if any of this work earns it.

No BigBounce paper self-annotates at N4. Site copy, paper abstracts, and all contribution records are capped at N3 by internal review.

N3

FIRST-OF-KIND DEMONSTRATION

FIRST-OF-KIND DEMONSTRATIONPaper 1A

Perturbation-Transparency Theorem

All-orders proof that the Barbero-Immirzi parameter γ is invisible in every perturbative observable for minimally-coupled scalar matter in ECH.

|ε^{μνρσ} R_{μνρσ}| < 10^{-15} (numerical confirmation of Holst topological identity across 1000 random Riemann tensors with FRW perturbation symmetries)

Why it matters

Tells you what the bounce CAN'T do: the bounce mechanism itself is invisible to telescopes. Every testable prediction must come from the contraction dynamics before the bounce, not from the bounce mechanism. This redirected the entire research program.

What it is

5-step proof chain: zero spin density for scalar matter → zero torsion → Levi-Civita connection at all perturbative orders → Holst term reduces to topological Nieh-Yan invariant → no perturbative dynamics from γ. Extended to scalar, vector, AND tensor perturbations.

What existed before

Hehl et al. (1976); Freidel, Minic & Takeuchi (2005); Calcagni & Mercuri (2009); Mercuri (2009); de Berredo-Peixoto et al. (2012); Långvik et al.

What we did

No prior work combined the known ingredients into an explicit all-orders perturbation theorem for minimally coupled scalar matter in ECH. We formalized it as a 5-step theorem, proved it extends to tensors, verified numerically to machine precision.

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 1A

14-Constraint Channel-Level Closure Map

Catalog of 14 independent structural constraints establishing channel-level closure, under stated assumptions, of the four enumerated minimal-ECH dark-energy routes.

g_eff ~ 10^{-61} · Planck suppression 10^{-122} · graviton-loop fine-tuning 10^{-57} (across the barrier set)

Why it matters

Closes every enumerated minimal route from a nonsingular ECH bounce to late-time dark energy under stated assumptions. Instead of testing one or two mechanisms and hoping, we systematically closed each channel. Tells future researchers exactly where NOT to look.

What it is

7 foundation studies (A-G) + 17 research branches (H-W). Each barrier is named, quantified, and cross-referenced. Mass-coupling lock, Topological-Shift Duality, scalar-tensor universality, Planck suppression, attractor-sensitivity dilemma, parameter immunity, Liouville conservation, and 7 more.

What existed before

Blagojević & Hehl (2013); Weinberg (1989); 't Hooft (1979); Shie, Nester & Yo (2008).

What we did

No prior work tested and closed all standard mechanism classes for connecting a nonsingular bounce to late-time dark energy within a single theoretical framework. Each branch opened only after passing a 4-question filter.

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 2

f_NL = -35/16 Forecast Package

First comprehensive forecast for testing Cai et al.'s matter-bounce non-Gaussianity prediction with upcoming surveys.

f_NL^{local} = -35/16 = -2.1875

Why it matters

Makes the bounce hypothesis testable. Cai et al. derived the matter-bounce f_NL in 2009 (the corrected squeezed value is −35/16 = −2.1875) but nobody built the full machinery to test it. We did: SPHEREx sensitivity, Bayesian model comparison, template mismatch quantification, robustness against systematics. SPHEREx data (~2028) will confirm or kill this at ~5σ.

What it is

Parameter-free local non-Gaussianity from matter-dominated contraction: 300× larger than standard inflation, opposite sign. Forecast: σ(f_NL) = 0.7 (Heinrich+2023 multi-tracer Fisher); 3-5σ after systematic budget; 5.2-5.5σ optimistic pre-GR/b_φ degradation; 4.4σ at MegaMapper even at the worst convention.

What existed before

Cai, Xue, Brandenberger & Zhang (2009); Heinrich, Doré & Krause (2023); Dalal et al. (2008); Li & Brandenberger (2014).

What we did

First combination of (a) SPHEREx + MegaMapper sensitivity specific to bounce; (b) Bayesian model comparison with 600K+ MC realizations (bounce favored at 8-17:1 over tuned multifield); (c) GR-projection robustness; (d) template-mismatch (r ≈ 0.85-0.90); (e) ε-correction bounded [1-8%]; (f) cubic bounce transmission estimate; (g) Li-Brandenberger convention resolution.

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 2

Physics-Derived Full-Commutator Polynomial

Resolves a 15-year factor-of-2 ambiguity between Cai et al. (2009) and Li et al. (2017) by tracing it to the in-in commutator.

Full-commutator polynomial: (6, 2, -18, 10, -66, 18) · template overlap r ≈ 0.85-0.90

Why it matters

Two groups published different answers; nobody knew who was right. We proved both groups are correct at their respective levels. The exact polynomial determines how well SPHEREx can actually detect the signal.

What it is

Using Cai et al.'s own intermediate vertex contributions (Eqs. 34-36), we derive the full-commutator shape polynomial algebraically: (6, 2, -18, 10, -66, 18). Proven by exact rational arithmetic. Published Eq. 37 coefficients (3, 1, -9, 5, -66, 9) are the single-time-ordering values. The factor of 2 is the in-in commutator: i⟨[ζ³, L]⟩ = -2 Im⟨ζ³ L⟩.

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 4

3.2M-Spiral Galaxy Chirality Catalog (8.47M classified)

Largest chirality-labeled galaxy catalog to date: 8.47M DESI Legacy DR8 galaxies classified CW/CCW/NOT_SPIRAL (3,201,160 spirals) by a flip-equivariant ViT pipeline, with a null real-space chirality dipole.

Real-space dipole +0.41σ (rank-p = 0.31) · WLS exclusion of 1.7% dipole at z ≈ −18 · A50 ≈ 0.75%, A95 ∈ (1.0%, 1.5%]

Why it matters

First chirality null at this scale with an honest falsification window. Constrains any cosmological parity-violating mechanism — including the early-universe consequences of ECH parity-odd structure beyond the ALP birefringence channel — and refutes the claimed ~3% parity signal.

What it is

8,474,531 galaxies from DESI Legacy Survey DR8 classified by a flip-equivariant Vision Transformer ensemble with test-time D4 averaging (equivariance suppresses the raw classifier asymmetry 2.98×: +1.576% → −0.529% in A-units). Headline: real-space ℓ=1 dipole at +0.41σ (empirical-rank p=0.31) on the high-confidence sample (N=949,584) plus a block-bootstrap WLS template fit disfavoring a clean 1.7% Shamir-class dipole at z≈−18; injection-recovery brackets A95 in (1.0%, 1.5%].

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 3

Multi-Survey Anomaly Catalog (378,280 anomalies)

First unified anomaly sweep across 7 surveys (37.3M sources): 378,280 unique anomalies (269,317 recommended-tier) after Path-C native retrains and 7-way 5″ positional dedup. The ~141× figure over prior catalogs is a full-instrument-stream vs science-target comparison, not a like-for-like size increase — the like-for-like DESI recount is ≈0.9× the largest single-survey benchmark (2,468 vs 2,685).

Why it matters

Cross-survey continuity is the load-bearing falsification surface for any new physics claim. Population-level rare-class discovery (z>6 QSOs, ultra-rare AE candidates) is the byproduct.

What it is

Path-C native-retrained per-survey tallies: DESI DR1 (195,829), SDSS DR18 (77,905), LAMOST DR10 (113,342, exploratory tier), eROSITA DR1 (298), Planck CMB (200), Gaia DR3 (500), NEOWISE (419, after the |b_ecl|<80° ecliptic-pole mask; 436 pre-mask). These sum to 388,493 raw → 378,280 unique after 7-way friends-of-friends union-find at 5″ (10,213 duplicate detections collapsed). Enrichment statistics (e.g. eROSITA) are reported descriptively, not as inferential significances.

How to verify

FIRST-OF-KIND DEMONSTRATIONPaper 5

DESI Chirality × Environment Null (z-shell corrected)

Galaxy chirality is statistically independent of DESI large-scale-structure environment: 791,635 matched spirals + 56,981 void spirals, with a z-shell selection correction and full covariate robustness.

Omnibus χ² nulls p = 0.31 / 0.99 · covariate-robust Wald p = 0.46 / 0.99 · no class clears Bonferroni significance under any classifier

Why it matters

Cosmic-web environment is the most natural place to look for spin alignment / handedness correlations. The controlled non-detection constrains the whole class of environment-coupled parity models at the ≳25 Mpc/h smoothing scale.

What it is

Matched 2,232,212 deduped DESI DR1 rows against the Paper-4 chirality catalog (duplicate-TARGETID join root-caused and rebuilt; 100% GZ–DESI join). DESIVAST three-algorithm void test, V-Web + T-Web + Tempel + ASTRA cross-checks, 21-shell z-shell selection-corrected rebuild, omnibus χ² environment nulls (p=0.31/0.99), and covariate-robust regression over size/magnitude/morphology/inclination (Wald p=0.46/0.99).

How to verify

N2

NOVEL COMBINATION / EXTENSION

NOVEL COMBINATION / EXTENSIONPaper 1A / Paper 2

ALP Birefringence Consistency

ECH parity structure motivates a Planck-scale ALP. Predicted β = 0.27° matches the 3.6σ Planck+ACT observation at 0.5σ.

β = 0.27° · consistent with Eskilt+ β = 0.342° ± 0.094° (3.6σ published detection)

Why it matters

Our benchmark value sits within 1σ of an actual published observation. LiteBIRD tests at 9σ in early 2030s. Bounce-mechanism independent — falsification target separate from f_NL.

What it is

Numerical ΛCDM ALP field evolution gives Δφ/f_a = 0.65-1.07 across the natural mass range m/H₀ ∈ [1, 3]. Fiducial β = 0.27° (at m ≈ 1.8H₀) is consistent with the published joint WMAP+Planck β = 0.342° ± 0.094° (3.6σ). NaMaster validation (synthetic ΛCDM skies, 500 MC): β = 0.27° recovered as 0.238° — a pipeline-validation figure, not a sky measurement.

How to verify

NOVEL COMBINATION / EXTENSIONPaper 1A

Topological-Shift Duality (Barrier 2)

Mass protection and geometric content are mutually exclusive for pseudoscalar fields coupled to the Nieh-Yan 4-form.

Why it matters

Closes a loophole that keeps coming up — people repeatedly try to build light pseudoscalars for dark energy from Nieh-Yan. Applies beyond our framework: constrains any attempt to use topological gravity terms as sources for light pseudoscalars.

What it is

If Nieh-Yan is topological (standard EC), pseudoscalar mass is shift-symmetry protected but the coupling is a total derivative — no dynamics. If Nieh-Yan is non-topological (metric-affine), dynamics arise but shift symmetry breaks — no mass protection. Cannot have both.

How to verify

NOVEL COMBINATION / EXTENSIONPaper 2

SPHEREx f_NL Fisher Forecast

Multi-tracer Fisher forecast of σ(f_NL) = 0.7 → 4.7-12σ detection of bounce f_NL = -2.1875 by 2027.

σ(f_NL) = 0.36 (Fisher ideal) · σ(f_NL) = 0.93 (Munchmeyer+2019 conservative) · detection 4.7-12σ

Why it matters

Sets a hard deadline on falsification — SPHEREx will report by ~2027-2028.

What it is

Externalized to Heinrich+2023 multi-tracer bispectrum forecast with the bounce template; noise-weighted shape mismatch, ε-correction, b_φ marginalization, GR projection all carried through.

How to verify

NOVEL COMBINATION / EXTENSIONPaper 1B

MCMC Verification Infrastructure (309,189 frozen samples)

309,189 frozen posterior samples across 2 converged dataset combinations (third accumulating). Honest null: ΔN_eff ≈ 0, H₀ = 67.68 (standard ΛCDM); DESI DR2 w0wa chain gives w_pivot = −0.952 ± 0.019 (+2.5σ from −1, twice-verified).

Why it matters

Demonstrates honest negative reporting — we found our own bug and disclosed it.

What it is

Cobaya 3.6.1 + CAMB. Frozen combinations: 176,240 full-tension + 132,949 Planck+BAO+SN, Gelman-Rubin converged. A separate DESI DR2 BAO + Planck NPIPE + DES-Y5 + Pantheon+ w0wa chain reports w_pivot at the decorrelated pivot. Corrected our own earlier artifact (H₀ = 69.2 was a SH0ES prior artifact).

How to verify

NOVEL COMBINATION / EXTENSIONPaper 3, §6

NANOGrav Bounce GW Spectrum

NANOGrav 15-yr real-KDE free-spectrum re-fit: γ = 2.567 ± 0.382. Matter-bounce γ = 3.0 consistent at +1.13σ; SMBHB γ = 4.33 excluded at +4.61σ.

γ_bounce = 3.0 (+1.13σ consistent) · γ_obs = 2.567 ± 0.382 · γ_SMBHB = 4.33 (+4.61σ excluded)

Why it matters

The PTA spectral slope is one of the few currently-measured observables that discriminates bounce-era tensor spectra from astrophysical SMBHB backgrounds.

What it is

Real free-spectrum KDE likelihood (Zenodo chains, emcee re-fit; ESS = 5,507) replacing the earlier synthetic-power-law summary statistic γ = 3.20 ± 0.42. Savage-Dickey Bayes factor decisively favors the bounce slope over the SMBHB slope.

NOVEL COMBINATION / EXTENSIONPaper 1B / Paper 4

NaMaster Pipeline Validation Suite

End-to-end pseudo-C_ℓ validation harness: 500-MC birefringence recovery on synthetic ΛCDM skies (P1B) plus MASTER-deconvolved chirality-field nulls with harmonic completeness anchors (P4).

β bias = −0.032° (500 MC, synthetic skies) · monopole-only null reproduces 99.3% of pre-MASTER ℓ=1 power

Why it matters

Mask-coupled pseudo-C_ℓ pipelines silently manufacture or destroy low-ℓ signals; a committed, rerunnable validation suite is what separates a believable null from an artifact.

What it is

P1B leg: synthetic ΛCDM polarization skies, ACT-like f_sky = 0.32 mask, 10 μK·arcmin noise, 500 MC — injected β = 0.27° recovered as 0.238° (bias −0.032°, sign-symmetric; σ_β(0.32) = 0.046° measured directly). Scope honestly declared: validates the E→B MASTER deconvolution, not the β–α foreground degeneracy. P4 leg: monopole-only generative nulls (99.3% of pre-MASTER ℓ=1 power), label-shuffle backgrounds, and injection completeness — a Shamir-class 1.7% dipole would recover at z ≈ 68–218 vs the observed +7.3.

NOVEL COMBINATION / EXTENSIONPaper 4 / program-wide

Provenance-Audit Methodology (retract-and-rebuild)

A data-provenance audit traced P4's previously-headlined −0.122σ subsample-mask null to a synthetic-footprint catalog; the result was withdrawn, the paper re-anchored on real-space estimators, and the audit trail published.

Why it matters

Most published nulls and detections never face an artifact-level audit. Treating retraction-and-rebuild as a first-class, documented workflow is itself a transparency contribution — reviewers can replay the exact decision chain.

What it is

Artifact-level provenance tracing (file hashes, generation scripts, footprint geometry checks) applied to every load-bearing null. When the audit failed, the claim was withdrawn in-paper (P4 Appendix A), the headline was rebuilt on the real-space +0.41σ dipole + WLS template exclusion, and the corrected equivariance suppression factor (3.86× → 2.98×) was propagated through every surface. P2's irreproducible ~9.9σ SDB joint-Fisher claim was withdrawn the same way and replaced by a committed-code 1.4σ/0.6σ subordinate channel.

NOVEL COMBINATION / EXTENSIONProgram-wide

12-Job Compute Reproducibility Chain

All 12 load-bearing compute closures (fsky sweeps, continuous-prior MCMC, permutation rebuilds, sign-symmetry reruns) executed on a dedicated pod with committed scripts + JSON artifacts, for ~$0.55 total.

Why it matters

A reviewer can rerun any headline number from the committed chain — reproducibility as an artifact, not a promise.

What it is

Each job ships its driver script, inputs, seeds, and output artifact in-repo; results are stamped into the papers via the \artifact{} macro pointing at immutable release tags. The chain covers P1B (fsky sweep, Caγ continuous-prior MCMC, c9f sign-symmetry), P4 (10k-permutation Table III rebuild, harmonic completeness injections), and P5 (closure-recompute scripts 17–19).

How to verify

NOVEL COMBINATION / EXTENSIONPaper 3

BigAE Multi-Survey Autoencoder Anomaly Detector

A deterministic symmetric autoencoder trained per-survey on native data, scoring 269,317 objects across 6 surveys by reconstruction residual — the largest-by-sources autoencoder anomaly search assembled for cosmological-anomaly discovery.

Why it matters

Turns 'anomaly' from a hand-tuned cut into a learned, survey-native reconstruction score, so one pipeline scales across heterogeneous surveys and every headline count is reproducible from committed scripts rather than a promise.

What it is

Per-survey NATIVE retrains (the cross-transfer failure mode — a LAMOST-trained model drifts to an ~98% blue-excess artifact on other surveys — forced a native-retrain protocol); standardized reconstruction-residual score S. Validated not by injection-recovery but by a 5-fold cross-validation robustness gate (mean pairwise Jaccard 0.862) plus an out-of-distribution Jaccard gate — a validation methodology for UNSUPERVISED anomaly catalogs. UMAP of the latent space shows the high-score anomalies concentrate in distinct islands rather than scattering.

What existed before

Autoencoder outlier detection (Baron & Poznanski 2017); single-survey spectral anomaly searches.

What we did

No prior work combined per-survey-native autoencoder retrains into a single 6-survey reconstruction-scored anomaly catalog at this scale, with a cross-validation/OOD validation gate replacing injection-recovery for an unsupervised search. Model + catalog released open-source.

How to verify

NOVEL COMBINATION / EXTENSIONPapers 4 & 5

Z₂-Flip-Equivariant Chirality Classifier (released model)

A flip-equivariant Vision Transformer that classifies galaxy spin handedness with parity symmetry built into the architecture — the correct inductive bias for a parity-violation measurement — with D4 test-time averaging giving 2.98× systematic-dipole suppression. Checkpoint released on HuggingFace.

Why it matters

For a chirality-DIPOLE measurement, an ordinary classifier can bake in an orientation bias that manufactures a signal; making the network exactly equivariant under the parity flip removes that entire class of systematic by construction rather than by post-hoc correction.

What it is

ViT-Small backbone with Z₂-flip equivariance enforced so CW↔CCW predictions transform correctly under image reflection; D4 (dihedral) test-time averaging suppresses residual orientation systematics 2.98×; calibrated on Galaxy Zoo 1 labels. Released as bamfai/galaxy-chirality-v2 with weights.

What existed before

CNN/ViT galaxy-morphology classifiers (Galaxy Zoo DECaLS, Zoobot); standard non-equivariant chirality classifiers (Shamir et al.).

What we did

First use of a built-in parity-equivariant classifier for a cosmological chirality-dipole measurement, with the equivariance itself as the systematic-control mechanism; model + weights released for reuse.

How to verify

NOVEL COMBINATION / EXTENSIONPaper 4

Pseudo-Label-Independence Retrain (GZ1-only null)

A control classifier trained on Galaxy Zoo 1 human labels ONLY (zero CE-ResNet pseudo-labels; val acc 0.978) reproduces the chirality-dipole null at z = −0.04σ — proving the vanishing dipole is not an artifact inherited from the pseudo-labels.

Why it matters

Directly answers the strongest circularity objection to a null from a partly-pseudo-labeled classifier: retrain on fully-independent human supervision and check the physical null survives. It does.

What it is

Retrained the equivariant ViT on GZ1 CW/CCW human labels only (CE-ResNet pseudo-label block gated off), re-inferred the galaxy sample, and ran the IDENTICAL real-space dipole estimator + per-pixel permutation null: z = −0.04σ (rank-p = 0.45), consistent with the canonical +0.41σ null.

What existed before

Self-training / pseudo-label validation typically checks classifier accuracy, not downstream-measurement independence.

What we did

A downstream-MEASUREMENT pseudo-label-independence test: retrain on fully-independent supervision and confirm the physical null (not merely accuracy) survives — a reusable template for validating ML-derived cosmological null results.

How to verify