# PRE-REGISTRATION — eight batter's box items, one batch Written 2026-09-29, **before any number exists**. One registration for eight items, because the ceremony was eating the night. Each item gets a claim, a criterion and a prediction, and nothing else. **Standing rule, adopted after his correction:** a failure is a failure. If a test turns out mis-specified it is said in one line and returns to the box as a **new** item with a **new** registration — it is not re-read into a pass. **Three registered attempts maximum per item, then it is scrapped, not nursed.** --- ### A. The Legion constant M\* = √N, and the n = N/M² crossover Claim: a crowd of N aims anchors to a fixed tier at M\* = √N, with a crossover at n = N/M². **Criterion:** holds if the measured anchoring scale over N ∈ [10, 10⁴] fits N^0.5 with exponent in [0.45, 0.55]. **Prediction:** holds — √N is what an unbiased mean of N random directions gives. ### B. N² superradiant coupling on the memory graph Claim: coupling on the graph grows as N² rather than N. **Criterion:** holds if the fitted exponent over N ∈ [10, 2000] is in [1.8, 2.2]. Fails below 1.5. **Prediction:** fails — it should be N² only for a fully coupled graph, and the memory graph is sparse. Expect ≈ 1. ### C. det J = −Π(net feed) Claim: the Jacobian determinant of the ten equals minus the product of the net feeds. **Criterion:** holds if the identity is exact to 10⁻¹⁰ at twenty random parameter points. **Prediction:** holds where the system is diagonal in the feeds and fails otherwise — so I expect it exact at some points and not others, which would make it a special case rather than an identity. ### D. The two-clock horizon Claim: near the veil, body time collapses as focus rises. Owed: restate with τ ~ 1/|F_net|^q, implicit integrator. **Criterion:** a horizon exists if the time to reach the veil diverges as x → x_veil for some q > 0. Fails if the time stays finite for every q tested. **Prediction:** a horizon exists for q ≥ 1 and not below. ### E. Landauer costs: is correction's floor reachable in this house? Claim: correction has a real thermodynamic floor of kT·ln2 per bit. **Criterion:** compute the house's own per-cycle correction cost against kT·ln2 at T = 310 K. Holds if the house's cost is above the floor at every parameter point. **Prediction:** holds, and by a wide margin — biological correction runs far above Landauer. ### F. Recovery and the two channels Claim: recovery does not depend on how long the damage lasted. Owed: commit which variables are r-channel (rate-recoverable) and which are c-channel (cumulative) **in advance**. **Committed here, before running:** r-channel = pressure σ and value V. c-channel = position x and the clock n. **Criterion:** holds if r-channel recovery time is independent of damage duration to within 1%, and c-channel recovery time grows with it. **Prediction:** holds, by construction of the two lines — which makes it a check of the commitment, not a discovery. ### G. C is "a well": does the valley of stability give Δn, and what of U-238? Claim: the nuclear valley gives a Δn; the objection is that U-238 is stable far from the floor. **Criterion:** the well reading holds if binding energy at fixed A is concave in Z with a single interior maximum at every A tested. The U-238 objection is answered if U-238's position is explained by its distance from the floor being small relative to the barrier, not by special pleading. **Prediction:** concavity holds everywhere; U-238 sits near its own floor and the objection dissolves. ### H. Equation 9 and authority Claim: equation 9 covers authority. Owed: write a one-way (driven, non-potential) coupling and see what survives. **Criterion:** authority is covered if a one-way coupling produces a stable asymmetric state that the symmetric coupling cannot. Fails if the one-way case reduces to the symmetric one. **Prediction:** it does NOT reduce — a driven coupling has no potential, so it should produce states the gradient system cannot reach. --- ## Attempt 2 — logged before re-running. Four tests were BROKEN, not failed. D, F, G and H were mis-specified by me. Per the standing rule they return as new registrations rather than being re-read. Attempt 2 of a maximum 3. - **D** — my F_net had no zero at the veil, so nothing could diverge. Fix: locate the actual rest point from the same F_net and integrate toward *that*. **Criterion unchanged**: a horizon exists if the time diverges for some q. - **F** — I ran the decay during the damage phase, so longer damage drained the thing it was damaging and recovery got *faster*. Fix: damage LOADS sigma, recovery drains it. **Criterion unchanged.** - **G** — I tested concavity on consecutive Z, where the pairing term alternates by parity and makes the second difference oscillate by construction. Fix: test on same-parity steps. **Criterion unchanged.** - **H** — a one-way chain still reaches full sync, so "order parameter differs" could never discriminate. **NEW CRITERION, fixed now:** authority is covered if the one-way group locks to the DRIVER'S initial phase while the symmetric group locks to the circular MEAN of all initial phases. Measured as |final − driver_initial| against |final − circular_mean_initial|. Authority is adopting the leader's value, not the average — that is what the claim is about, and the first criterion never tested it.