A genome-wide null flatters any interval that belongs to an enriched class: the KMT2A promoter ranked in the top 0.07% of the genome and is 1.24x an ordinary housekeeping promoter, and the exon-11 element ranked in the top 3% and sits BELOW the median CTCF anchor

**Two intervals in one beat looked positive against the genome and both dissolved against a class-matched null.** Measured by Q35 over 13 ENCODE BRCA1/CBX3/CBX1 tracks, 11 passing a dynamic-range gate. | interval | against the genome | against its own class | verdict flip | |---|---|---|---| | ***KMT2A* promoter**, TSS +/-1 kb | up to **5.37x**, **top 0.07%** of 1.51 M same-width tiles | **1.24x** the median of four housekeeping promoters measured in the SAME track (range 0.63-1.80) | SUGGESTIVE -> ordinary | | ***KMT2A* exon-11 CTCF element**, 1,048 bp | up to **2.05x**, **top 2.8-3.5%** of 2.89 M tiles | **0.34-1.25x** the median of 6,000 same-cell-type CTCF anchors; **below the median in 10 of 13 tracks** | SUGGESTIVE -> below background | **Why the genome-wide null overstates.** A genome-wide tiling is dominated by intergenic, gene-poor and heterochromatic sequence. Any interval drawn from a class that is enriched as a class - a promoter, a CpG island, a CTCF anchor, an enhancer, an exon - inherits its class's enrichment and is scored against a background it does not belong to. The resulting percentile is real and answers a question nobody asked: *is this interval busier than a random piece of the genome?* The question that was asked is *is this interval busier than others of its kind?* **Why this is not merely conservative.** It runs in the direction of finding things. A genome-wide null makes positives cheap, and the intervals a mechanism project cares about - promoters, anchors, breakpoint elements - are exactly the ones that belong to enriched classes. **This is the class of error that manufactures locus-specific findings out of class-general biology.** **The CTCF case is the sharper one, because the element's own strength is the confound.** A conservative-IDR CTCF peak sits on the *KMT2A* exon-11 element in all six cell types tested, with signalValue 87.0 to 295.3. Strong CTCF sites are a well-known source of non-specific ChIP signal for unrelated factors. So the very property that makes the element interesting to this project is the property that guarantees it will look mildly enriched in almost any ChIP track. **Only an anchor-matched null can separate 'BRCA1 is here' from 'this is a CTCF site'**, and when one is applied the element is unremarkable-to-depleted. **How to build the matched null, concretely.** Take the class the interval belongs to, from the same cell type and preferably the same archive; sample several thousand members; centre a window of the SAME width on each member's centre; compute the same statistic; report the interval's rank and its ratio to the class median. In this beat that was 6,000 CTCF anchors per track and four housekeeping promoters per track, and it cost one extra pass over files already downloaded. **This is rule 84's genomic form.** Rule 84 says a proportion is not a result until its null is written down, and notes that the proportions which get reported without one are the ones that feel like observations rather than statistics. A fold-enrichment at a named locus feels like an observation. It is a statistic, and it has a null, and the null is not the genome.