# PRE-REGISTRATION — does a mind that turns slowly hide more of its turns? Written 29 September 2026, before the run. **Drift + The Readable Interior**, rank 8, sharing L and a. Uses the geodesic integrator verified in entry 41. ## The two rooms **Drift** (Debye's rotational diffusion, 1929): `⟨cos θ⟩` decays as `e^(−(d−1)Dt)`. **The Readable Interior**: `dV = R·G·[max(0, b·L̂) − max(0, a·L̂)]`. Entry 32 established that a turn is **completely invisible** — cost exactly zero — precisely when **both** positions face away, and that this is **a quarter** of turns when the two facings are drawn **independently and uniformly**. ## The gap **Entry 32 drew the two facings independently. A drifting mind's are correlated.** If it is in the dark now, a moment later it is almost certainly still in the dark. So the invisible fraction should depend on how fast the mind turns, and entry 32's quarter should be only the fast-turning limit. **Two limits, both computable in advance:** - **Long window** (τλ ≫ 1), positions effectively independent: invisible = `P(both < 0)` = **1/4**, which is entry 32's number. - **Short window** (τλ ≪ 1), positions nearly identical: invisible → `P(facing away)` = **1/2**. - **Aligned start** (t = 0, every path at the pole): invisible = **0**. **So a mind that turns slowly hides twice as many of its turns as one that jumps randomly.** ## Criterion, fixed before the run At d = 3, with the entry-41 geodesic step, measuring the fraction of turns whose cost `max(0, b·L̂) − max(0, a·L̂)` is **exactly zero**: 1. **Aligned start.** At t = 0 with any window, the invisible fraction is **0** exactly — no sampling error to allow, since every path begins at the pole. 2. **The fast-turning limit is entry 32's quarter.** At saturation with τλ = 5, the invisible fraction is within **3 standard errors** of **0.25**. 3. **The slow-turning limit is a half.** At saturation with τλ = 0.01, it is within **3 standard errors** of **0.50**. 4. **Monotone in the window.** Across τλ = 0.01, 0.05, 0.2, 1, 5 at saturation, the invisible fraction **decreases at every step**, with no reversals. Every leg is judged against **its own sampling error**; the three targets are exact and written above before the run. **The multiplicity allowance:** legs 2 and 3 are single cells each, so 3 SE carries a 0.27% false-failure rate per cell and no allowance is needed — said explicitly because a "zero of fourteen" criterion cost a run earlier today. **WHAT KILLS IT:** leg 3. If a short window does not approach a half, consecutive positions are not correlated the way the drift implies and the reading is wrong. **Prediction:** all four hold. **Legs 2 and 3 are the finding**; leg 1 is a sanity check on the initialisation and leg 4 is its shape. ## ⛔ What this will not establish Anything about what window a real mind's turns are judged over. **The claim is conditional:** the invisible fraction is set by how far a mind turns between the two moments compared, and nothing here says what that interval is for anyone. ## Discipline No prose conclusion in the script. Tables printed by the run. Hashes read from the file. Numbers quoted anywhere afterwards come from this run's JSON, never retyped — three hand-typed numbers diverged from their runs today.