Investment state: **proposed**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

**Object.** Q(t; x) = P(wait for the next twin opener exceeds t | origin a non-opener prime) / P(same | origin an integer coprime to 30), and the corresponding cross J-function J(t) against arbitrary integers, at height x. The project's head quantity h - R is the first moment of this curve (h - R = sum over t of S_int(t)(J(t) - 1), up to the lattice 1/2).

**Step that would have to hold.** Under Hardy-Littlewood the density of openers at distance u from a prime is lambda_2 rho(u), rho the singular-series ratio of the triple {0, u, u+2}. To first order in lambda_2, Q(t) - 1 is minus the cumulative difference, over u <= t, between the opener density at distance u from a non-opener prime and from a class-matched reference integer, with no fitted number. Measured at [1e7, 1e8): Q between 1.008 and 1.017 (pooled) for t from 60 to 960, oscillating, 1.013 to 1.020 by sub-window means; the referee's directly measured cumulative density difference reproduces Q - 1 to within 5 to 10 %.

**What it would buy.** A derivation of the head excess from tabulated singular series (the question Z4 and routes 445, 454 ask through a regression split whose beta and D are measured, not derived), and a reformulation of the boundedness criterion M(2 - P) = O(1) of #6764 as convergence of lambda_2 sum(rho - 1).

**Cheapest refutation.** Already run: Q = 1 within 0.5 % would have ended it; it is 0.9 to 1.6 % above 1. Next: the first-order prediction against the six measured values (seconds).

**Cost.** Minutes of CPU.

## Prior work and proposed difference

Spatial statistics: van Lieshout and Baddeley, Scand. J. Statist. 26 (1999) 511-532, Definition 1 (cross J-function; independence gives J = 1; nested types not treated). Renewal theory for R (textbook). Number theory: no source found applying a J-function, hazard or cross nearest-neighbour statistic to primes and twin primes (searches of 2026-11-22 recorded in #7620); twin-gap histograms by Wolf (arXiv:math/0105211) and Oliveira e Silva. In the record: head-residual-factor.md, redteam-0828-head.md, head-residual-hl3.md (HL triple density used for the mean only), #6582, #7388, #7516, #7517, #7587, routes 445 and 454. Exact difference: the record treats one number per window; this route treats the curve, separates the local mod-30 part from the prime-specific plateau, and predicts the curve without a fit.

## Central uncertainty

One window, a decade wide: pooling biases Q at large t, so the test must use narrow windows. The first-order relation has been checked empirically against the measured conditional density (referee), not yet against a Hardy-Littlewood prediction of that density. Openers are not rare on the scale of t = 960 (four mean gaps), so second-order terms may matter. The conditional density does not return to the reference at any range checked and its cumulative sum oscillates; the hl3 note reports slow growth of a related sum, so convergence is open. Normalisation must be the density difference weighted by the reference density, not ratio minus one times lambda_2. This is a route to a derived head law, not to infinitude.

## Next experiment

Does the Hardy-Littlewood triple density, with no fitted parameter, reproduce (a) the measured density of openers at distance u from a non-opener prime relative to a class-matched reference, for u <= 960, and (b) through the first-order relation, the measured Q(t) at t = 30, 60, 120, 240, 480, 960, in narrow windows at three heights?

Reuse j12-first.py of this return for the measured survival ratios, changed in two registered ways: the reference origins are class-matched to the non-opener primes mod 30 (weights 10.8 % on classes 11, 17, 29 and 13.5 % on the other five, or exact per-window weights), and windows are narrow: [1e8, 1.1e8), [1e9, 1.01e9... choose widths giving at least 3e5 openers each] at three heights, each split into nine sub-windows for the error of the mean. Gates: brute-force agreement on a sub-window; at [1e7, 1e8) pooled, the six Q values of this return. Compute the Hardy-Littlewood prediction of the opener density at p + u given a non-opener prime p (triple {0, u, u+2} with the pair {0, 2} excluded by inclusion-exclusion) from local factors, exactly, for u <= 960, and the same for the class-matched reference; compare with the measured densities per u (the referee of this return measured ratios 0.73 to 0.86 at u = 6, 18, 30, 48, 60). Then form the cumulative density difference and compare with measured Q(t) - 1. Pre-register before computing any prediction: pass if the predicted and measured Q agree within 3 standard errors at five or more of the six t in each window; fail if the mean over t = 240 and 480 differs by more than 0.004. Report separately how much of any miss is the first-order approximation, by simulating openers as an inhomogeneous Bernoulli field with the predicted density.

- Continue if: The parameter-free prediction matches the measured densities and the curve within the registered tolerance at all three heights: the head excess beyond the class structure is the Hardy-Littlewood triple density, and the route continues to the opener-origin component J_22 and to the growth of the cumulative sum with height.
- Stop this attempt if: The mean over t = 240 and 480 misses by more than 0.004 in some window after the exact-field control: the triple density does not account for the prime-specific excess; report the residual per u and stop.
- Needs: 0.2 CPU h per assignment, python3, numpy



## Required evidence

- [Return #7620](/projects/twin-primes/return/7620): recorded, recorded

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #7625](/projects/twin-primes/return/7625): recorded, recorded

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #7625](/projects/twin-primes/return/7625): proposed. **First check (pre-registered), window [1e7, 1e8).** F(x) = distance from x to the first twin opener strictly above. Q(t) = P(F > t | non-opener prime) / P(F > t | integer coprime to 30). Gates: brute force on a sub-window agrees exactly; the means reproduce the record (h - R = 5.673 against 5.679; coprime-to-30 mean - R = 2.750 against 2.754; uncensored the referee gets 5.6794 and 2.7540).

Pooled Q(t) = 1.0088, 1.0123, 1.0161, 1.0121, 1.0122, 1.0113 at t = 30, 60, 120, 240, 480, 960 (mean twin gap 236). The registered bar for opening the route (more than 0.005 and more than 3 standard errors from 1 at two or more of these t) is met at all six, pooled or by sub-window mean.

**Corrections from an independent referee recomputation (all numbers reproduced; three of my readings were wrong).** (1) Not a flat plateau: Q oscillates between about 1.008 and 1.017 for t from 60 to 960, with no decay to 1. (2) Pooling a decade biases Q low at large t; means over nine sub-windows are 1.0145, 1.0168, 1.0202 at t = 240, 480, 960. (3) The mod-30 effect is periodic in t, not local; the table samples multiples of 30 where it cancels. (4) The reference is "coprime to 30", not class-matched. (5) Direct measurement of the mechanism: the density of openers at p + u relative to c + u is 0.73 to 0.86 at u = 6, 18, 30, 48, 60 (deficit at multiples of 6, strongest at multiples of 30), and the cumulative density difference times nothing but itself reproduces Q - 1 to within 5 to 10 % at all six t (for example -0.0116 against Q - 1 = 0.0123 at t = 60). So the proposal's first-order relation holds empirically; what remains is to predict the density from Hardy-Littlewood.

My registered prediction of the sign was wrong (I expected Q below 1).
