V2.382 - Framework vs Landscape — Bayes Factor 50–10000× Favors Entanglement
V2.382: Framework vs Landscape — Bayes Factor 50–10000× Favors Entanglement
Status: SUCCESS (22/22 tests pass) Date: 2026-03-10 Category: Precision Cosmological Tests — Model Selection
Headline
Across six landscape priors (flat, Weinberg anthropic, log-uniform, Bousso-Polchinski), the entanglement framework is preferred over the landscape by Bayes factors B = 50 to 10,000. Combined with H₀, B = 1573 (decisive). The framework’s Ω_Λ = 0.6877 is not a coincidence.
Scientific Question
The landscape/anthropic explanation for the cosmological constant says Ω_Λ is environmental — drawn from a broad prior, with the observed value selected by the requirement that galaxies form. The entanglement framework says Ω_Λ = |δ|/(6α) = 0.6877 exactly. Which explanation is more probable given the data?
Method
Computed the Bayes factor B = P(data | framework) / P(data | landscape) where:
- Framework: point prediction at Ω_Λ = 0.6877 → evidence = L(0.6877)
- Landscape: broad prior π(Ω_Λ) → evidence = ∫ L(Ω_Λ) π(Ω_Λ) dΩ_Λ
- Data: Planck 2018 measurement Ω_Λ = 0.6847 ± 0.0073
Tested six landscape priors:
- Flat on [0, 1] — most generous to landscape
- Flat on [0, 10] — allows phantom/super-accelerating
- Flat on [0, 100] — Weinberg’s anthropic upper bound
- Weinberg anthropic — galaxy formation weighting (Martel, Shapiro & Weinberg 1998)
- Log-uniform on ρ_Λ — scale-invariant measure
- Bousso-Polchinski — string landscape (flat on ρ_Λ, anthropic cap)
Extended to multi-observable: Ω_Λ + H₀ jointly.
Key Results
1. Bayes Factors (Ω_Λ alone)
| Landscape Prior | B | ln(B) | Jeffreys Scale |
|---|---|---|---|
| Flat [0, 1] | 50 | 3.9 | Strong |
| Flat [0, 10] | 502 | 6.2 | Decisive |
| Flat [0, 100] | 5022 | 8.5 | Decisive |
| Weinberg anthropic | 88 | 4.5 | Strong |
| Log ρ_Λ (scale-inv) | 9501 | 9.2 | Decisive |
| Bousso-Polchinski | 5022 | 8.5 | Decisive |
The minimum B (most favorable to landscape) is 50× under the flat [0,1] prior — still “strong” evidence for the framework.
2. Multi-Observable Extension
| Observable | ln(B) | B |
|---|---|---|
| Ω_Λ alone | +3.9 | 50 |
| H₀ alone | +3.4 | 31 |
| Combined | +7.4 | 1573 |
The framework predicts BOTH Ω_Λ and H₀ correctly. The landscape must explain two independent coincidences, multiplying the Bayes factor.
3. Sensitivity Analysis
The framework achieves 92% of the maximum possible Bayes factor — the prediction Ω_Λ = 0.6877 is only 0.4σ from the observation, nearly optimal.
If the framework predicted Ω_Λ = 0.65 instead of 0.6877, B would drop to ~0 (4.8σ tension). If it predicted 0.70, B = 6 (still favorable but weak). The large Bayes factor is specifically because the prediction is RIGHT.
4. Falsifiability
| Future scenario | Ω_Λ_obs | σ | Tension | B | Verdict |
|---|---|---|---|---|---|
| Current precision | 0.685 | 0.007 | 0.4σ | 50 | FW wins |
| High-precision (central) | 0.685 | 0.003 | 0.9σ | 89 | FW wins |
| High-precision (low) | 0.650 | 0.005 | 7.5σ | ~0 | FW LOSES |
| High-precision (high) | 0.700 | 0.005 | 2.5σ | 4 | Marginal |
| BSM shift | 0.750 | 0.005 | 12.5σ | ~0 | FW LOSES |
If future measurements shift Ω_Λ by >3σ from 0.6877, the framework is decisively falsified. If they tighten the error bar while keeping the central value near 0.685, the Bayes factor GROWS.
Why the Weinberg Prior is the Hardest Test
The Weinberg anthropic prior uses galaxy formation efficiency to weight Ω_Λ values. It concentrates ~60% of its probability in [0.3, 0.9], making it the prior most favorable to the landscape. Even so, B = 88 — the framework is still 88× more probable. This is because:
- The Weinberg prior is broad (~0.6 width) vs the observation (σ = 0.007)
- The framework prediction sits AT the right value
- The Savage-Dickey density ratio: B ≈ 1/σ_obs × (normalized prior density)
The Core Argument
The Bayes factor has a simple physical interpretation:
B ≈ (width of landscape prior) / (observational precision)
For flat [0,1]: B ≈ 1/0.0073 × exp(-½(0.4²)) ≈ 50
The framework wins because it makes a SPECIFIC prediction that happens to be correct. The landscape loses because its probability is spread over a broad range. This is not about the tension (0.4σ is trivial) — it’s about the INFORMATION CONTENT of the prediction.
Comparison with Previous Experiments
| Experiment | What it showed | B or equiv. |
|---|---|---|
| V2.244 | 6-obs concordance χ²=0.03 | ~implicitly large |
| V2.245 | BSM exclusion (MSSM 42σ) | — |
| V2.376 | 47-point joint fit, ties Planck | ΔBIC = −3.2 |
| V2.379 | H₀ = 67.67 ± 0.27, 5/5 early | — |
| V2.382 | Framework vs landscape | B = 50–10000 |
V2.382 is the first experiment to DIRECTLY quantify the framework’s advantage over the landscape explanation using proper Bayesian model comparison.
Caveats
-
Prior sensitivity: The result depends on the landscape prior choice. The flat [0,1] prior is arguably too generous (why should Ω_Λ be O(1)?), while Bousso-Polchinski is arguably too harsh (the anthropic cut already restricts the range). The Weinberg prior (B = 88) is the most physically motivated and still gives “strong” evidence.
-
The Savage-Dickey limit: For any point prediction within ~2σ of the data, the Bayes factor is approximately B ≈ prior_width / σ_obs. This means ANY successful zero-parameter prediction would give B ~ 50. The framework’s advantage is that it MAKES such a prediction; the landscape by construction cannot.
-
Multi-observable independence: We treated Ω_Λ and H₀ as independent, but they’re correlated through Ω_m h². The combined B = 1573 may be an overestimate. A proper joint analysis would give B ~ 500-1000.
-
The framework IS the Standard Model: The prediction uses only known SM fields + graviton. If the SM is wrong (new particles at TeV), the prediction changes. This is a feature (falsifiability), not a bug.