The Registry

Falsification ledger

Every test we armed against ourselves

Each claim in the registry ships the test that would kill it. 82 are on record. 10 have fired — taking 7 claims to the graveyard and putting 2 in quarantine. Every one of those kills was our own. 27 have never been run, which is the honest weak spot in this table.

10 fired 1 failed, fixed same day 3 closed by enumeration 1 searched, nothing found 40 survived 27 never run

key↑	claim	domain	status▪	re-run↓
USR-2026-0043	A 7.8σ drift we quarantined ourselves Multi-probe bootstrap wₐ departure at 7.8σ. 7.8σ (quarantined ghost)	Dark energy	QUARANTINED	—
USR-2026-0042	A 6.8σ vacuum departure we quarantined ourselves Multi-probe bootstrap w₀ departure from -1 at 6.8σ. 6.8σ (quarantined ghost)	Dark energy	QUARANTINED	—
USR-2026-0016	Both satellites weakly prefer a large-scale cutoff Both Planck and WMAP weakly prefer a low-ℓ power cutoff (Planck Δχ² = 3.72 ≈ 1.4σ at ℓ_c = 2.6; WMAP ≈ 1.2σ at ℓ_c = 2.5) — a curiosity, not a detection. 1.2-1.4σ — curiosity, not detection	CMB	PROVISIONAL	2026-07-31
USR-2026-0012	WMAP's largest angles are missing their power The WMAP large-angle sky is anomalous: quadrupole 243 µK² (p = 0.052), cut-sky S_1/2 anomalous at p = 7.7×10⁻⁴ (3.2σ; 1.1×10⁻³ = 3.1σ under the literal full data path), and a 12.7° quadrupole–octopole alignment (p = 0.025). REVISED 2026-07-31: the earlier 'p < 0.0007' was a 0-of-1500 Monte-Carlo floor; the value above is measured on 10⁵ simulations with a beam-and-pixel-window-corrected null. ~3σ (cut-sky S_1/2) reproduced ≠ discovered	CMB	PROVISIONAL	2026-07-31
USR-2026-0069	We audited ourselves and found one bad error bar Covariance audit of the live registry: of 56 live claims, exactly ONE (USR-2026-0003's local-structure half) uses diagonal errors on a dataset that ships a full covariance — and the registry had already flagged it. Re-running the flagship both ways quantifies the exposure: full covariance gives Δχ² = 6.973 (2.16σ), while dropping only the Pantheon+ off-diagonals inflates it to 17.934 (3.83σ) — a 1.77× significance inflation, the same mechanism that produced this registry's 6.8σ quarantine. 1 of 56 live claims exposed; diagonal errors inflate the flagship by 1.77× in σ reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0068	The dark-energy drift is not the vacuum running The measured dark-energy drift is NOT the vacuum's renormalisation-group running, for three independent reasons. Magnitude: fitting the running-vacuum form to DESI DR2 + Pantheon+ + CMB gives ν = +0.0015 ± 0.0014 (1.07σ), while the QFT running implies ν ≈ 4×10⁵³ — a ratio of 10^56.4. Shape: w_eff(0) = −1 exactly for EVERY ν, an algebraic identity, against a measured w₀ = −0.849. Form: the irreducible m⁴ running renormalises the additive constant, not the νH² term, so it cannot be ν at any magnitude. ν = +0.0015 ± 0.0014 (1.07σ) vs a QFT-implied 4×10⁵³ — a 10^56.4 gap, and the shape is wrong too reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0067	Unimodular gravity is blind to vacuum energy The trace-free (unimodular) form of Einstein's equations is exactly blind to vacuum energy: shifting T_μν → T_μν + ρ_vac·g_μν leaves all 16 components of the trace-free field equation IDENTICALLY zero on a fully general metric — while the standard form picks up a residual of −κρ_vac·g_μν. In FRW the trace-free equation reduces to Ḣ = −4πG(ρ+p), which depends on matter only through (ρ+p), so a vacuum with p = −ρ contributes nothing at all. vacuum-energy shift leaves the trace-free equations exactly zero; Λ becomes an integration constant reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0066	Lorentz invariance splits the vacuum-energy problem in two The vacuum-energy problem splits cleanly along Lorentz invariance. Lorentz-VIOLATING regulators (hard 3-momentum cutoff, spatial lattice) give w = +1/3 AND a quartic divergence; Lorentz-INVARIANT regulators (Pauli–Villars, dimensional regularisation) give w = −1 exactly AND no power divergence at all. The two columns are the same fact: a K⁴ term has no m⁴ to carry it, so it must be built from a regulator scale — and a Lorentz-invariant regulator has no preferred frame to build one from. Lorentz-violating ⟺ w = +1/3 + quartic divergence; Lorentz-invariant ⟺ w = −1, no power divergence reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0064	Why w = −1 is an identity we can test The sharpest empirical handle on the vacuum: w = −1 is an identity for a true vacuum energy (Lorentz invariance forces T_μν = −ρ g_μν), not a fit parameter — so observation can test it. From this registry's corrected CPL fit (w₀ = −0.849, wₐ = −0.548), w crosses −1 at z = 0.380 and is phantom above it, at Δχ² = 6.97 on 2 dof = 2.16σ. That is NOT a detection: 24% of the way to 5σ, which needs σ(w₀) ≤ 0.030. 2.16σ — consistent with a constant vacuum, mildly prefers drift, not a detection reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0063	The anthropic bound explains less than advertised The anthropic bound on the vacuum energy explains far less than advertised: a four-way-validated linear growth ODE gives collapsed fractions 0.561, 0.210, 7×10⁻³, 6×10⁻⁹ as Λ grows ×1, ×10, ×100, ×1000, placing the ceiling at 6.4×, 35.5×, 108× the observed value at 50%, 10%, 1% thresholds. Granting a multiverse ensemble removes ~118.6 of 120.1 orders, leaving ~1.55; without granting the ensemble it explains zero. anthropic ceiling 6–108× observed; ~1.55 of 120 orders left over, and only if a multiverse is granted reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0062	The vacuum catastrophe is a choice of assumption The famous vacuum-energy discrepancy is not a number but a function of an assumption: against the observed 2.24 meV scale it is 10^120.7 at a Planck cutoff, 10^108.4 at GUT, 10^56.4 at 1 TeV, 10^41.6 at QCD, and exactly 1 at the observed scale by construction — four decades of exponent per decade of cutoff, with the loop-factor convention alone shifting every row by 2.2 dex. 10^120.7 at Planck cutoff — but 10^56.4 at a TeV; the exponent is assumption-laden, the problem is not reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0020	The sky is lopsided, but only at large angles The hemispherical power asymmetry peaks at 2.0σ (p = 0.0250) at LMAX = 64 toward (l,b) = (228°, -24°) — 7° from Lepus — and vanishes (0.2σ) by LMAX = 96. 2.0σ at LMAX = 64; 0.2σ by LMAX = 96 reproduced ≠ discovered	CMB	PROVISIONAL	2026-07-26
USR-2026-0003	Every supernova compilation shows the same drift The evolving-DE preference appears in all three SN compilations refit through one pipeline: Pantheon+ Δχ² = 6.96 (2.16σ), DES-SN5YR 8.95 (2.53σ), Union3 11.83 (3.00σ). Separately and more weakly, the local-structure escape hatch falls about 5× short on a diagonal-covariance shell and dipole test. 2.16σ / 2.53σ / 3.00σ across three SN compilations (post-Lyman-α-fix) reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0001	Dark energy is drifting, not constant Dark energy evolves: a CPL (w₀, wₐ) fit to DESI DR2 BAO + CMB + Pantheon+ beats ΛCDM by Δχ² = 6.97 (2.16σ), with w₀ = -0.849 and wₐ = -0.548. CORRECTED 2026-07-26: the earlier Δχ² = 8.2 (2.4σ) used a transposed DESI Lyman-α covariance pairing — see the audit log. Δχ² = 6.97 → 2.16σ (was 2.39σ before the Lyα covariance fix) reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-26
USR-2026-0049	MOND's acceleration scale is mildly disfavoured In the SPARC radial-acceleration relation, the parameter-free MOND acceleration scale a₀ is mildly disfavored. The canonical a₀ sits +2.6σ above the data's own per-galaxy median offset (+0.116 dex); the zero-parameter cosmological guesses cH₀/2π and cH₀/6 land at +1.6σ and +2.0σ; a free-fit a₀ is consistent (+0.3σ). canonical a₀ +2.6σ_median; free-fit +0.3σ_median	Galactic dynamics	PROVISIONAL	2026-07-12
USR-2026-0048	No gap in the GD-1 stellar stream The GD-1 stellar stream shows no statistically significant density gap. The deepest local under-density (z = −5.0 in a single bin at φ₁ = −21°) has a look-elsewhere-corrected global significance of p = 0.587 across 39 bins and 5000 smooth+Poisson simulations — consistent with an unperturbed stream. global p = 0.587 — not significant	Galactic dynamics	PROVISIONAL	2026-07-12
USR-2026-0029	What polarization could add at low multipoles Conditional forecast, ratios only: in this computation σ_i = 2.5 × √(F_i/F_T), so the four quoted sensitivities are one calibration and three mode-count ratios — E-modes are 84% independent of temperature at low ℓ, giving LiteBIRD E ≈ 2.5σ and T+E ≈ 3.4σ relative to a Planck-T value that is an INPUT, not a result. ratios only: σ_i = 2.5 × √(F_i/F_T); the 2.5 is an input	CMB	PROVISIONAL	2026-07-09
USR-2026-0028	Where the asymmetry axis actually points Coordinate recomputation, no significance computed: the pipeline's measured asymmetry axis (221°, −24°) is RA 5.6h, Dec −17°, in Lepus, at ecliptic latitude −41°. Its separations from hardcoded reference directions are 13.9° (published Planck power-asymmetry direction), 23.9° (Dipole Repeller), 34.1° (CMB Cold Spot). coordinate arithmetic only — no significance claimed or computed	CMB	PROVISIONAL	2026-07-09
USR-2026-0027	No excess or deficit on the largest scale of all No detection of either excess or deficit on the largest observable scale: P/P_std = 0.09 ± 1.08, i.e. 0.84σ from standard. The data do not require extra large-scale power and cannot exclude up to about 2.2× standard. Below k = 1.5×10⁻⁴ Mpc⁻¹ the reconstruction is formally unconstrained (errors 262% to 142001%). P/P_std = 0.09 ± 1.08 — 0.84σ from standard, i.e. no detection either way	CMB	PROVISIONAL	2026-07-09
USR-2026-0026	The asymmetry lives only at the largest angles The power asymmetry's excess over ΛCDM is confined to ℓ ≲ 40. The raw amplitude falls from A ≈ 0.107 to ≈ 0.02 across ℓ = 2–220, but the ΛCDM null falls across the same bands and the observed decline has not been shown to exceed it. Under a conservative mask the significance is 0.94σ (p = 0.173). excess confined to ℓ ≲ 40; 0.94σ under a conservative mask	CMB	PROVISIONAL	2026-07-09
USR-2026-0025	The lopsided axis is not our own galaxy The asymmetry axis is not a foreground: raw single-frequency asymmetries (0.212/0.161/0.476) and the 143-217 GHz difference point at the galaxy (within 3°), while the cleaned maps agree with each other — WMAP ILC 0.139 at 25° and Planck SMICA 0.151 at 24° from Lepus. diagnostic — axis is not foregrounds	CMB	PROVISIONAL	2026-07-09
USR-2026-0024	The oldest light varies more on one side The CMB low-ℓ variance dipole is significant in this pipeline: A = 0.102 toward (l,b) = (221°, −24°), p < 0.005 — zero of 200 ΛCDM realizations exceeded the observed amplitude, so 200 sims cap the reportable significance near 2.6σ. Single configuration; no second pipeline, cleaning or mask has confirmed it. A = 0.102 toward (221°, −24°), p < 0.005 (200-sim floor, ~2.6σ ceiling) reproduced ≠ discovered	CMB	PROVISIONAL	2026-07-09
USR-2026-0021	No scar from a collision with another universe No bubble-collision signature in the CMB: global p = 0.903 (0.0σ); the best local candidate is a cold disk of radius 10° at (214°, -27°) with local S/N 2.6, and the Cold Spot itself scores only 1.7. null — p = 0.903 (0.0σ)	CMB	PROVISIONAL	2026-07-09
USR-2026-0019	A mirrored corner reproduces both sky anomalies A corner (reflecting-corner) observer reproduces both halves of the anomaly — ⟨parity⟩ → 0.52 with a low quadrupole, KDE likelihood ratio ≈ 4.8× over ΛCDM — but its predicted anti-mirror correlation (S_min = -0.332 ± 0.102) is not detected: observed S_min = -0.233, the 69th percentile of ΛCDM skies. likelihood ratio ~4.8× — suggestive; predicted signature not detected	Cosmic geometry	PROVISIONAL	2026-07-09
USR-2026-0017	Counting the looks shrinks the large-angle anomaly After look-elsewhere correction the joint large-angle boundary anomaly is 0.8–1.6σ. The 1.6σ is the most favourable end of that range, not a single answer: it assumes 5 effective independent looks (1.24σ at 10, 1.01σ at 15), and excluding the method-fragile alignment term leaves 0.75σ. 0.8–1.6σ post-LEE — a range set by two stated choices, not a single number	CMB	PROVISIONAL	2026-07-09
USR-2026-0015	Polarization does not settle the missing power TE polarization does not decide the missing-power question: S_1/2^TE has p = 0.203, leaning low but not decisive. p = 0.203 — not decisive	CMB	PROVISIONAL	2026-07-09
USR-2026-0014	The missing power survives every map cleaning The large-angle missing power survives all four map cleanings (D₂ = 243/200/184/198 µK²; S_1/2 anomalous at >3σ in every cleaning), while the alignment axis is fragile (3-30° spread). >3σ missing power; axis fragile	CMB	PROVISIONAL	2026-07-09
USR-2026-0013	Planck confirms the missing large-angle power Planck SMICA independently confirms the large-angle missing power seen in WMAP (C₂ = 200 µK², p = 0.030). The quadrupole–octopole alignment does NOT independently confirm: the registered 3.0° came from an unmasked full-sky transform, while this seam's own masked estimator reports 23.8° (p = 0.085). C₂ = 200 µK² (p = 0.030) confirms missing power; alignment is 23.8° masked, not 3.0° reproduced ≠ discovered	CMB	PROVISIONAL	2026-07-09
USR-2026-0011	How far a bigger sky could push a cutoff test Mode-counting scaling argument (not a Fisher forecast — no instrument noise models were used): if the significance of a horizon-scale primordial cutoff scales as the square root of the number of independent modes, then the 1.4σ CMB-temperature baseline tops out near ~2σ once partly-independent E-modes and LSS/ISW are added, because every probe at our epoch samples the same finite set of horizon-scale modes. ~2σ ceiling from a mode-counting argument, not an instrument forecast	CMB	PROVISIONAL	2026-07-09
USR-2026-0010	Quasars and the oldest light correlate, weakly A matched-filter cross-correlation of Planck SMICA with 756k Quaia quasars gives a positive temperature–galaxy cross-power at 2.2σ against a no-ISW null. Only the sign and 'at least ΛCDM level' are claimed: the fitted amplitude A = 5.00 ± 2.28 carries statistical error only. 2.2σ positive ISW cross-power; amplitude normalisation systematic-limited, NOT a tension reproduced ≠ discovered	CMB	PROVISIONAL	2026-07-09
USR-2026-0009	The universe is spatially flat The universe is spatially flat: Ω_k = +0.0025 ± 0.0019 with CMB+BAO+SN (1.3σ from zero); geometry-only data (BAO+SN, no CMB) allow Ω_k = +0.0320 ± 0.0174. null result — flat (Ω_k = +0.0025 ± 0.0019)	Cosmic geometry	PROVISIONAL	2026-07-09
USR-2026-0008	An oscillating dark energy ties the standard fit On this data combination the CPL best fit is not uniquely selected: a five-parameter oscillating form ties it (χ² = 1413.72 versus 1414.11). But at its best fit the amplitude is pinned against the code's own A < 1.5 bound with a 24-e-fold period against 1.25 e-folds of data, so the 'oscillation' is a monotone segment — and CPL still wins on AIC by 1.6. χ² tie (1413.72 vs 1414.11) but the oscillator is pinned at a code bound; CPL wins on AIC by 1.6	Dark energy	PROVISIONAL	2026-07-09
USR-2026-0005	No late-time fix reaches the local Hubble value Of the five late-time dark-energy candidates tested, none reaches H₀ = 73 when SH0ES is included — all cap at 69.25–70.24. Only a free sound horizon does, reaching H₀ = 73.06 at ΔAIC = −13.46 with r_d shrunk 2.96% (~4.4 Mpc). Within CPL the fit crosses w = −1 at z = 0.381. five late-time candidates cap at H₀ = 69.25–70.24; only free r_d reaches 73 reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-09
USR-2026-0004	Dark energy's pressure changes across cosmic time Binned w(z) rises from -0.920 ± 0.040 (z = 0.0-0.4) through -1.078 ± 0.093 (z = 0.4-1.0) to -1.537 ± 0.282 (z = 1.0-2.5); the trend survives dropping all SNe but dies without BAO. BAO-anchored trend — no independent σ quoted reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-09
USR-2026-0002	The Hubble constant disagrees with itself The Hubble tension is real in this pipeline: post-Lyman-α-fix inverse-ladder H₀ = 68.71 ± 0.39 versus SH0ES 73.04 ± 1.04 — a 3.90σ gap. 3.90σ gap (registered as 4.4σ before the Lyman-α fix) reproduced ≠ discovered	Dark energy	PROVISIONAL	2026-07-09
USR-2026-0044	Gravity does not weaken with redshift The growth index γ evolves from 0.41 (z < 0.3) to 0.10 (z > 1) — gravity weakens with redshift. was 4.84σ claimed — 1.03σ on independent reproduction, opposite sign	Structure growth	KILLED	2026-07-26
USR-2026-0007	A sign-switching vacuum is not preferred after all Sign-switching Λ (Λs-CDM) is NOT preferred by this data: ΔAIC = +1.50 versus ΛCDM with SH0ES included, with the vacuum flip at z = 3.11. The registered ΔAIC = −2.5 at z ≈ 2.0 came from the transposed DESI Lyman-α covariance pairing and does not reproduce. ΔAIC = +1.50 — disfavoured (registered as −2.5, preferred)	Dark energy	KILLED	2026-07-26
USR-2026-0047	The wide-binary MOND boost was an estimator artefact A Gaia wide-binary velocity analysis shows an apparent MOND-direction boost, but only under a clip-biased median estimator (residual 1.167). The unbiased mean gives residual 0.948 [0.939–0.958] — consistent with Newton. The boost is an estimator artifact of debiasing in the noise-dominated a₀ regime, where 28.7% of pairs are clipped. residual 0.948 [0.939–0.958] — Newtonian (was 1.167 under clip-biased median)	Galactic dynamics	KILLED	2026-07-12
USR-2026-0023	The quasar dipole excess belonged to the catalogue The Quaia quasar-dipole excess (raw D = 0.0208, 9.9σ over shot noise) is not cosmological: adversarial magnitude and galactic-latitude cuts collapse it to D = 0.0066, inside the kinematic expectation band. was 9.9σ raw — kinematic (0.0066) after cuts	CMB	KILLED	2026-07-09
USR-2026-0022	One giant wave cannot tilt the sky A single super-horizon wave cannot cause the tilt: matching the dipole-free asymmetry requires λ ≥ 308× the observable diameter, while the observed quadrupole caps it at λ ≤ 4× — the window is shut. killed — the vise (308× vs 4×)	CMB	KILLED	2026-07-09
USR-2026-0018	No simple box-shaped universe fits the sky No simple-box cosmic topology works: across torus, Dirichlet, and Neumann boundary conditions the simulated even/odd parity ratio stays at 0.97-1.05 — no box reproduces the sky's odd-parity preference. killed — no box makes odd parity	Cosmic geometry	KILLED	2026-07-09
USR-2026-0006	Holographic dark energy is ruled out Holographic dark energy (Hubble-horizon HDE) is ruled out by information criteria: ΔAIC = +37.5 versus ΛCDM. killed — ΔAIC +37.5	Dark energy	KILLED	2026-07-09
USR-2026-0045	Early results whose scripts no longer exist Early-generation (gen1-~31) Dark Energy pipeline results exist only as ingested summaries — their analysis scripts no longer exist. unverifiable — scripts lost	Dark energy	NOT RERUNNABLE	—
USR-2026-0046	The universe as the inside of a black hole The observable universe is the interior of a black hole. speculation — no measurement	Cosmic geometry	SPECULATIVE	—

Open to the field This registry is not just ours — register any published result you want checked and it enters like everything else: provenance bundle, kill test, status set from a public audit, labeled reproduced ≠ discovered. How submission works →

The graveyard

Claims we killed, and what killed them

USR-2026-0044 2026-07-26

The growth index γ evolves from 0.41 (z < 0.3) to 0.10 (z > 1) — gravity weakens with redshift.

Died by its own kill test, on our own data, at our own hand. The 4.84σ evolution reproduces at 1.03σ with the sign reversed, no jackknife sample recovers it, and the claimed Δχ² is arithmetically unreachable given how well the no-evolution model already fits. The autopsy is the useful part: fσ₈ loses 93% of its sensitivity to γ by z ≈ 2, and γ is 91% correlated with σ₈ — so a high-z γ 'measurement' is mostly the amplitude degeneracy talking. Gravity does not weaken with redshift in this data; it looks like General Relativity (γ = 0.525 ± 0.077 vs 0.55). Post-mortem addendum: fixing a wrong fσ₈ point uncovered during the autopsy makes the evidence against this claim stronger, not weaker — γ_a moves from +0.63 to +1.19 while the claim requires about −0.68.

USR-2026-0007 2026-07-26

Sign-switching Λ (Λs-CDM) is NOT preferred by this data: ΔAIC = +1.50 versus ΛCDM with SH0ES included, with the vacuum flip at z = 3.11. The...

Published as weakly preferred at ΔAIC = −2.5; re-runs at +1.50, which is disfavoured. The whole reversal came from one transposed covariance block — the same Lyman-α pairing error that cost USR-2026-0001 a quarter of a sigma. Killed not because the idea is bad but because our own number was wrong, and the corrected number points the other way. The lesson is propagation: this claim sat unchanged for hours after the fix that invalidated it.

USR-2026-0047 2026-07-12

A Gaia wide-binary velocity analysis shows an apparent MOND-direction boost, but only under a clip-biased median estimator (residual 1.167). The...

Killed by its own estimator audit: the MOND boost lived only in the clip-biased median. Under the unbiased mean the wide binaries are Newtonian (0.948). The 28.7% clip rate in the a₀ regime was doing the work.

USR-2026-0023 2026-07-09

The Quaia quasar-dipole excess (raw D = 0.0208, 9.9σ over shot noise) is not cosmological: adversarial magnitude and galactic-latitude cuts...

The adversarial battery did its job: 9.9σ of raw excess collapsed to a kinematic D = 0.0066 under magnitude and latitude cuts. The dipole belonged to the catalog, not the cosmos.