# Genesis pedigree test: locked exploratory analysis plan

**Date locked:** 2026-08-09 (Australia/Adelaide)  
**Status:** Exploratory, not preregistered. The biblical observations and the earlier Opus 5 calculations were already known. Pilot work before this lock reconstructed the Opus calculation and established that a lifespan shock placed before Nahor is the best one-shock fit. The held-out pedigree test below had not been run at the time this plan was written.

## Question

Conditional on treating the Septuagint-derived figures as individual biological observations, does a genetic-retention model fitted only to the pre-Babel lineage predict the first two post-Babel observations better when it includes a major outcross at Babel?

This is a test of quantitative compatibility and internal cross-prediction. It cannot establish the historicity of Babel, Noah, or the proposed arrival, and it cannot identify actual loci.

## Data

Primary series supplied by Mike Russell:

| Generation | Name | Fathering age | Lifespan |
|---:|---|---:|---:|
| 0 | Shem | 100 | 600 |
| 1 | Arphaxad | 135 | 565 |
| 2 | Cainan | 130 | 460 |
| 3 | Shelah | 130 | 533 |
| 4 | Eber | 134 | 504 |
| 5 | Peleg | 130 | 339 |
| 6 | Reu | 132 | 339 |
| 7 | Serug | 130 | 330 |
| 8 | Nahor | 79 | 208 |
| 9 | Terah | 130 | 205 |

The vault's earlier lifespan table gives Shelah as 460 rather than 533. Both versions will be run; neither will be silently selected.

## Babel boundary

The primary boundary is immediately before Nahor (transition 7 to 8). This follows the chronology rather than the lifespan fit: Babel is placed near Peleg's death; Peleg dies 339 years after his birth, which is 77 years after Serug's birth and 53 years before Nahor's birth under the supplied fathering ages. Nahor is consequently the first named member conceived after that date.

Sensitivity boundaries are immediately before Peleg and immediately before Reu. A scan of all possible shock boundaries will be reported as exploratory and corrected for the scan.

## Longevity model

For an assumed ambient lifespan `A`, Shem's excess lifespan is `D = 600 - A`. Longevity is represented by `N` equal-effect effective factors. `K_g` is the number retained in the focal individual at generation `g`:

1. `K_0 = N`.
2. Before Babel, `K_(g+1) ~ Binomial(K_g, r_pre)`.
3. Observed lifespan is `A + D K_g/N + Normal(0, sigma_env)`.

This is a pedigree-correlated version of the Opus independent-binomial approximation. Its `N` is an effective number of equal-sized segregating factors, not a literal locus count.

The model is fitted only to generations 0 through 7. Generations 8 and 9 are held out.

### Held-out models

- **Continuation:** transitions 7 to 8 and 8 to 9 both use fitted `r_pre`.
- **Babel pulse:** transition 7 to 8 has retention `h = 0.5`; transition 8 to 9 returns to fitted `r_pre`.
- **Persistent Babel regime:** both held-out transitions have `h = 0.5`.

`h = 0.5` represents the additive expectation for one parent carrying the focal genetic component and one ambient parent carrying none. Sensitivity values `h = 0.35` to `0.75` will be scored without selecting the best value as though it had been fixed in advance.

Primary ambient values are 40, 60, 80 and 100. Effective-factor values are 2, 3, 5, 8, 10, 15, 20, 30, 50, 75, 100, 200 and 500. `r_pre` and non-genetic scatter are estimated from the training observations. Profile likelihoods, not a single best grid cell, will be reported.

### Primary comparison

The primary statistic is the held-out joint log predictive density of Nahor and Terah under each model. The one-step Nahor prediction is also reported because Terah introduces an additional unobserved spouse.

Simulation-based parameter recovery is required. Synthetic eight-generation training series will be generated at known `N`, `r_pre`, and `sigma_env`; the analysis will be asked to recover broad categories (few: <=5; tens: 8-50; many: >=75). Failure means the biblical series cannot identify locus-count categories under this model.

## Babel half-excess check

For an additive full outcross, the first-generation prediction is

`L_Nahor = A + (L_Serug - A)/2`.

The ambient lifespan implied without using a chosen value is therefore

`A_implied = 2 L_Nahor - L_Serug`.

This will be calculated at every parent-child boundary. The Babel alignment is assessed against the range 40 to 100, fixed before the calculation. Because the exact numerical match was noticed during pilot reasoning, it is descriptive rather than confirmatory.

## Maturation model

The exact ages are first reduced to the classification fixed by the hypothesis: delayed (`>=120`) or early (`<120`) for Arphaxad through Terah. This yields seven delayed observations, Nahor early, and Terah delayed.

Under a full Babel outcross, a homozygous recessive founder genotype becomes heterozygous in Nahor, predicting loss of delayed maturation. If `M` independent recessive factors are all required, and Terah's mother transmits the founder allele at each with probability `q`, the probability of an immediate Terah restoration is

`P(rebound | M,q) = (q/2)^M`.

Values `M=1..6` and `q=0.1..1.0` will be tabulated. This is conditional, since the mother's genotype is unknown.

The location-only probability that the sole early value among nine falls at the independently specified Babel boundary is 1/9 under exchangeability. This is not treated as a complete null model for the extraordinary pre-Babel clustering.

## Combined evidential standard

The analysis may conclude:

1. **Contradicted:** Babel models predict the held-out data worse, or the maturation reversal is effectively impossible throughout the few-locus parameter range.
2. **Compatible but non-identifying:** Babel models predict better, but factor count/endogamy cannot be recovered or the result depends strongly on unmeasured environmental variance.
3. **Conditionally discriminating:** the held-out prediction favours a Babel outcross, broad architecture categories are recoverable, and the same outcross gives non-negligible maturation loss-and-rebound probability only in a restricted few-locus range.

No result will be described as genetic evidence for the historical event. The strongest permitted wording is evidence that the numerical trajectories are quantitatively consistent with—and comparatively better predicted by—the specified inheritance model.