Holding mark, arm and drug conditions fixed and varying only which confound the replicate pair carries: the floor ranges are DISJOINT at D1 and D10, a processing-confounded pair inflates D1 2.20x while a batch-confounded pair inflates D10 2.21x, and the direction reverses by stratum - so the floor is a property of the pair, not of the mark or the arm, and every floor this project has published came from the pair type that is wrong for D1

**Statement.** Cycle 111 named its own largest defect: it scored four marks against a floor measured from H2AK119ub pairs. This tests whether that transfer was legitimate — and finds the question was wrongly posed. **The pre-registered `P3` branch fired.** At D1 the **within**-mark spread (40.75 points) exceeds the **between**-mark spread of medians (9.93 points), so the floor is a clean property of neither the mark nor the arm. `P3` required identifying the dominating term rather than publishing a per-mark table, and it is identifiable. **Holding mark (H2AK119ub), arm (MLLAF9 Control) and all three drug conditions FIXED, and varying only which confound the pair carries:** | stratum | `R1-R2` processing | `R2-R3` 7-month batch | ratio | ranges | |---|---|---|---:|---| | TOP | 37.37% [13.70–62.63] | 61.89% [59.18–68.74] | 1.66× | overlap | | **D1** | **37.04%** [22.25–45.72] | **16.84%** [4.97–19.95] | **2.20×** | **DISJOINT** | | **D10** | **23.75%** [14.88–26.30] | **52.57%** [48.43–59.42] | **2.21×** | **DISJOINT** | **The ranges do not overlap at D1 or D10**, and **the direction reverses**: a processing mismatch inflates the dense bulk, a batch mismatch inflates the peaks and the sparse tail. **The practical consequence is sharp. Every floor this project has published came from a processing-confounded pair** — the `paired`-versus-`rep2` comparison that cycles 88–110 called *"the two conventions"*. **That is the pair type that inflates D1 — the exact stratum three cycles of argument were about.** Cycle 104's `M1` passed against a floor that this says is roughly 2.2× too wide there, which makes its pass *more* generous than cycle 111 already concluded, not less. **The corrected gate also reproduces Q135 independently.** Scoring pair agreement as `|A∩B|/min(|A|,|B|)` with the stratifier applied afterwards, **H3K27me3 is refused at all three conditions (86.31–88.90%)** and the `R1-R3` pair — which carries *both* confounds — is refused at all three (90.43–91.38%). The gate now refuses pairs for disagreeing rather than for having a sparse stratifier. --- # ⚠ QUALIFIED THE NEXT CYCLE — THE FINDING STANDS, THE LABEL DOES NOT This entry's own counter-evidence said the depth explanation *"is arguably more parsimonious"* and that the per-track depth *"is on disk and was not used"*. **It was used. It supports the depth reading.** **The two pair types are perfectly confounded with two different mismatch statistics, in opposite directions:** | pair type | depth ratio (total coverage) | breadth ratio (non-zero bins) | |---|---|---| | `R1-R2` "processing" | **2.74–3.57×** | 1.056–1.066 | | `R2-R3` "batch" | 1.86–1.91× | **1.145–1.276** | | | **disjoint** | **disjoint** | **And that maps onto which stratum each inflates.** A **depth** mismatch acts hardest where shares are large — the dense bulk, D1 — and the processing pairs are the depth-mismatched ones, which inflate D1. A **breadth** mismatch changes which bins are non-zero at all, acting hardest on the sparse tail and the peak set — and the batch pairs are the breadth-mismatched ones, which inflate TOP and D10. Spearman across the six pairs: **floor@D1 vs depth ρ = +0.714**, **floor@D10 vs breadth ρ = +0.543**. **So "processing" and "batch" are almost certainly the wrong causal labels.** They are metadata names for pairs that differ measurably in depth and breadth, and this design — n=3 against n=3, with the two predictors anti-correlated by construction — **cannot separate any of the four.** With n=6, |ρ|=0.829 is p≈0.06 two-sided; none of these correlations is significant on its own. **What survives, and is more useful than what it replaces:** the floor tracks **how mismatched the pair is**, and mismatch in depth and in breadth act on **different strata**. That is measurable for any pair from the binned tracks alone, whereas "processing" and "batch" require deposit metadata and explain nothing mechanistically.