# PRE-REGISTRATION — does a bad corrector hurt as much as a good one helps? Written 29 September 2026, before the run. **Retention + The Strand**, rank 9, sharing R, c, ε. ## The two rooms **Retention** (lands on the measured fidelity ladder of DNA — proofreading and mismatch repair): `R = 1 − ε(1−c)`. Retention rises with correction c. **The Strand** (lands on methyl-directed strand discrimination, 1983): `c_eff = m(2s − 1)` — *"correction helps only if the corrector trusts the right strand more than half the time."* Retention's c is taken as given. The Strand says what c **actually is** when a corrector may be looking at the wrong copy. **Neither room substitutes one into the other.** ## The substitution being tested R_eff = 1 − ε(1 − m(2s − 1)) At **s = 1** (the corrector always picks the right strand) this is Retention's own `R = 1 − ε(1−m)`. At **s = ½** it is `1 − ε` — the no-corrector baseline, correction contributing nothing. At **s < ½** the effective correction is **negative**, so retention falls **below** the baseline. ## Criterion, fixed before the run 1. **The substitution reproduces Retention at s = 1.** `R_eff(s=1)` equals `1 − ε(1−m)` to within 10⁻¹⁵ across a grid of ε and m. 2. **Exact antisymmetry about a half.** `c_eff(s) + c_eff(1−s) = 0` and `R_eff(s) + R_eff(1−s) = 2(1−ε)` to within 10⁻¹⁵ at every s on a grid. **That is the claim that a corrector wrong with probability s hurts by exactly what one right with probability s helps** — the benefit and the harm balance exactly about the no-corrector baseline, with no asymmetry in a mind's favour. 3. **The threshold is exactly a half.** `R_eff < 1 − ε` if and only if `s < ½`, with no exceptions over a grid of 200 values of s crossed with ε and m. Tolerances are 10⁻¹⁵ because all three are **algebra**, so the only error is floating point. Stating why the number was chosen, since six criteria tonight were numbers I picked without asking. 4. **Reported, not scored.** The region of (ε, m) where a bad corrector drives `R_eff` below **zero**, which would be unphysical for a retained fraction, and the worst `R_eff` reachable. **WHAT KILLS IT:** leg 2 failing. Then the harm and the benefit are not balanced, and the more interesting claim would be whichever way it leans. **Prediction:** all three hold. **They are algebra, so this is a check on my reading of the two rooms, not a discovery about the world** — and the write-up must say so, as entries 33 and 34 do. ## Discipline No prose conclusion in the script. Tables printed by the run. Hashes read from the file.