# PRE-REGISTRATION — does a turn stay readable when value runs away? Written 29 September 2026, before the run. **The Readable Interior + The Singularity**, rank 11, sharing G, L, V. ## The two rooms **The Readable Interior**: `dV = R·G·[max(0, b·L̂) − max(0, a·L̂)]` — the cost of a turn, paid in a currency the system does not set. **The Singularity**: *"if V′ = G and G′ = kLVG (grace growing with value, **a feedback the house does not adopt**), then V = 2/(kL(t_c − t)) — exponents −1, −2"*. Value reaches infinity in a finite time. ## The question neither room asks Grace is a bit: **G ≡ 1**. So the cost of a turn does **not** depend on how much value a mind holds — it is the same for a pauper and for a mind whose ledger is exploding. But what a reader can *detect* is the cost **relative** to the value it sits in. **So: under runaway, does the cost of a turn stay readable, or does it drown?** ## What is tested **Under the house's own model** (per-cycle cap, G ≡ 1), V grows at most linearly in n, so the relative cost `turn cost / V` should fall as **n⁻¹**. **Under the feedback the house does not adopt**, `V = 2/(kL(t_c − t))`, so the relative cost should fall as **(t_c − t)¹** — reaching **exactly zero at the finite time t_c**. That is the difference that matters: under the cap the relative cost is always **positive**; under the feedback it is **extinguished at a finite time**. ## Criterion, fixed before the run 1. **The Singularity's form is right.** `V = 2/(kL(t_c − t))` satisfies `V′ = kLV²` to a relative residual below **10⁻⁸** at twenty times before t_c, and the printed exponents (−1 in (t_c − t) for V, −2 for V′) are recovered by a log-log fit to within **0.01**. 2. **The turn cost does not depend on V.** With G ≡ 1, the cost of a fixed turn is identical (to 10⁻¹⁵) at V spanning ten orders of magnitude. 3. **The two decays.** Log-log fits give the relative-cost exponent **−1.00 ± 0.01** against n under the cap, and **+1.00 ± 0.01** against (t_c − t) under the feedback. **Why these tolerances are tight, stated in advance:** all three are exact algebra, not sampling. Linear growth gives cost/V ∝ 1/n identically, and V ∝ (t_c−t)⁻¹ gives cost/V ∝ (t_c−t) identically. The only error is floating point and the fit itself, so 0.01 is loose rather than strict. **This is the check I failed to make five times tonight: ask what precision the quantity has before fixing the number.** 4. **Reported, not scored.** The relative cost at t = t_c under the feedback, and at n = 10⁶ under the cap. **WHAT KILLS IT:** leg 3 failing at either end. If the cap's exponent is not −1, V is not growing linearly and my reading of the Ledger is wrong. If the feedback's is not +1, the Singularity's published form does not give the decay I claim. **Prediction:** all three hold. ## ⛔ What must not be overclaimed If this holds it says **runaway value extinguishes readability in finite time**, and that the per-cycle cap preserves it. That composes with entry 15 (the cap keeps value finite) and entry 20 (the ledger cannot see a counterfeit) — **a composition, not new evidence**, and any write-up says so. ## Discipline Verdict from measured quantities. Tables printed by the run. Hashes read from the file, never typed.