# PRE-REGISTRATION — runaway, attempt 2: against the room's actual system Written 29 September 2026, after attempt 1's first leg failed, before this run. ## Why there is an attempt 2, and it is a different error from the others `PREREG_runaway_kills_readability_2026-09-29.md` leg 1 required `V = 2/(kL(t_c − t))` to satisfy **`V′ = kLV²`** to a relative residual below 10⁻⁸. Measured residual: **exactly 0.5. FAILS, and the verdict stands.** **The room is right and my pre-registration was wrong.** The Singularity's system is V′ = G , G′ = kL·V·G so `V″ = kL·V·V′`, which integrates to **`V′ = (kL/2)V²`**, whose solution is exactly the published `V = 2/(kL(t_c − t))`. Testing it against `V′ = kLV²` is testing a different equation, and the factor of two I measured **is** that difference. **The five earlier failures tonight were thresholds set without derivation. This one is worse in kind: I wrote the wrong differential equation into a criterion and would have reported a published room as defective had I not checked why the number was exactly one half.** The lesson is the same either way — derive the criterion from the object, not from a recollection of it. **Legs 2 and 3 of attempt 1 HELD and are not re-run:** the turn cost is independent of V (spread exactly 0.0 across ten orders of magnitude), and the decay exponents are −0.9985 against n under the cap and +1.000000 against (t_c − t) under the feedback. ## Criterion, fixed before this run 1. **The coupled system produces the published solution.** Integrating `V′ = G, G′ = kL·V·G` from V(0) = V₀, G(0) = G₀ with a step small enough to be checked (see leg 2), the trajectory matches `V(t) = 2/(kL(t_c − t))` with `t_c` determined by the initial conditions, to a **relative** error below **10⁻⁶** at every recorded time up to 0.99·t_c. 2. **The integrator is trustworthy at that tolerance.** Halving the step changes the reported V at 0.99·t_c by less than **10⁻⁶** relative. Without this, leg 1 measures my step size. 3. **The first integral holds.** `V′ = (kL/2)V²` to a relative residual below **10⁻⁶** along the trajectory, confirming the derivation rather than assuming it. **WHAT KILLS IT:** leg 1 failing at a tolerance leg 2 has shown the integrator can meet. That would mean the published solution does not solve the published system. **Prediction:** all three hold, and the room is vindicated. ## Discipline Attempt 1's FAILS is published beside this, with the reason named as mine.