Read all 76 archived CBX3 and BRCA1 tracks at the KMT2A exon-11 anchor at SIGNAL level against a CTCF-anchor-matched null, and score the anchor's rank among its own class in each - no wet work, no new data, and it decides whether the sub-IDR peak is a foothold or a threshold artefact

**The question this settles.** `ana-q37-the-kmt2a-exon-11-anchor-carries-a-cbx3-and-a-brca1-peak-that-idr-removes` found a called peak on the exon-11 CTCF element in 3 of 180 archived tracks and, in ENCODE's own pre-IDR files, in 2 of 2 replicates of two experiments at global IDR 0.227 and 0.0588. That is a peak-level observation at one threshold. **It does not say whether the element carries a reproducible signal-level enrichment relative to the class it belongs to**, which is the only question a CTCF anchor can be asked honestly (`concept-a-genome-wide-null-flatters-any-interval-that-belongs-to-an-enriched-class`). **Design.** 45 CBX3 and 31 BRCA1 hg38 tracks exist across ChIP-Atlas and ENCODE. For each, read the mean signal over the seven Q35 intervals and over **N = 2,000 CTCF anchors of the same cell type**, drawn from that cell type's own CTCF peak set, matched on peak width and on distance to the nearest TSS. Report the exon-11 anchor's **percentile rank among its own class**, per track, not its fold over a genome-wide tile median - the failure Q35 itself identified, where the *KMT2A* promoter fell from the top 0.07% of 1.51 M tiles to 1.24x against housekeeping controls. **Statistic and pre-declared decision rule.** The anchor **fires** if its class-matched percentile is **above the 75th percentile of same-cell-type CTCF anchors in at least 60% of interpretable tracks of one protein**, and the same statistic computed at the 6 non-anchor Q35 intervals stays below the 50th. It **fails** if the median percentile across interpretable tracks is between the 25th and 75th - i.e. the anchor is an ordinary CTCF anchor for that protein. **What makes this cheap and what makes it fail.** Every input is a public bigWig. There is no antibody, no cell culture and no consent question. The practical constraints are recorded because this project has been burned by all of them: **bin size must be measured per file before any 147 bp number is quoted**; **the build must be read from each FILE's own chrom table across at least five chromosomes**, because padding is a property of files and not of series; **Content-Length and md5 must be compared before two accessions are counted as two replicates**; and **pybigtools 0.3.0 returns NaN for uncovered bases**, so every interval is read twice, once with `fillna` for the honest coverage fraction and once with `nan_to_num` for the statistic. Disk is the real constraint - the CBX3 LP-1 track alone is 467 MB and the DNA2 HCT116 track 585 MB - so the loop must download, analyse and delete one file at a time. **Why it is worth doing even though the likely answer is negative.** Q35's own class-matched read put the element at 0.34-1.25x against 6,000 CTCF anchors, and this beat's peak-level class-matched read found 30.2% of K562 CTCF anchors carry a CBX3 peak. **Both point to 'ordinary anchor'.** The value is that the corpus currently holds two contradictory headline sentences - *no occupancy above background* and *a peak at the element in three tracks* - and a class-matched signal-level number over 76 tracks replaces both with one that can be quoted.