# PRE-REGISTRATION — the pressure leg, attempt 2: the window derived from the time constant Written 29 September 2026, after attempt 1's third leg failed, before this run. ## Why there is an attempt 2 `PREREG_nothing_arriving_is_lost_2026-09-29.md` legs 1 and 2 **HOLD exactly** (0.0×10⁰) and are not re-run. Leg 3 **FAILED** and that verdict stands. Both of its failures were mine: 1. I integrated `dΠ/dt = P − rΠ` to **t = 200** at every r. At r = 0.05 the time constant is 1/r = 20, so t = 200 is only **10 time constants** and the remaining gap is `P/r · e^(−10) = 8.2×10⁻⁴` — which is precisely the "error" I measured. **The integration had not converged; nothing was wrong with the law.** 2. I required held pressure to exceed **10⁴** by t = 10⁴ with the drain closed. With inflow `max(0, −a·L̂) = 0.9`, linear growth gives `0.9 × 10⁴ = 9,000`. **My arithmetic, not the model's.** **That is the fifth threshold I have set tonight without deriving it from the quantity being measured.** The correction here is to derive both, not to tune them. ## Corrected design, derived **Convergence window.** For `dΠ/dt = P − rΠ`, the gap to steady state decays as `e^(−rt)`. To leave a relative residual below 10⁻⁸, integrate to `t = 20/r` (since e^(−20) = 2.1×10⁻⁹). **The window is set by the time constant of each run, not by one fixed number.** **Unboundedness.** With the drain closed, Π ≈ P·t, so the right test is **linearity, not magnitude**: the ratio Π(10t)/Π(t) must approach 10. A magnitude threshold depends on P and tests nothing. ## Criterion, fixed before the run 1. **Steady state.** Integrating to t = 20/r, the result matches `Π* = max(0, −a·L̂)/r` to a **relative** error below **10⁻⁶** at all 20 combinations of a·L̂ and r. Relative, because Π* spans 0.1 to 18 and an absolute bound would be a different test at each r. 2. **Unbounded with the drain closed.** At r = 10⁻⁶, the ratios Π(100)/Π(10), Π(1000)/Π(100) and Π(10000)/Π(1000) each lie in **[9.9, 10.1]** — growth stays linear, so it does not saturate. 3. **Zero in, zero held.** For every a·L̂ > 0 the held pressure is exactly 0. **WHAT KILLS IT:** leg 1 failing at a *relative* tolerance would mean the steady state is not P/r and the load does not go into held pressure. Leg 2 failing would mean it saturates and the debt statement is wrong. **Prediction:** all three hold. This is a test of an ODE against its own closed form, so a failure would mean my integrator is wrong. ## Discipline Attempt 1's FAILS on leg 3 is published beside this. Legs 1 and 2 of attempt 1 stand on their own and are not re-run or re-reported.