The third silent-failure class, audited after being named in four king logs: exactly one of seven bigWig analyses is raw-coverage, the single incommensurable pair is already labelled three times, and the cross-pipeline ratio comparison was bounded at the point of measurement
also called third silent-failure class · ratio versus raw coverage · the gap named four times and run once · 4.7x is not 50x
Two ways of measuring a protein on DNA give numbers that cannot be compared directly, and this project worried for four cycles that it might have mixed them. It mixed them once, and had already written the warning next to it.
The same thing, for a scientist
Cycle 59 found two silent-failure classes in track work and named a **third** it did not audit: *do any entries quote a magnitude from a RATIO track alongside one from a RAW-COVERAGE track?* The worked example was that **4.7× element CTCF from Gray's ratio tracks is not comparable with 50× from Canela's raw HCT116 coverage**. That question was then carried unaudited into the cycle-60, 61 and 62 king logs — **named four times, run zero times**. This runs it. **Method: classify every bigWig analysis in the repository by what its tracks actually are, by reading each one rather than pattern-matching.** There are seven, which is few enough to open — and this project's standing lesson is that a count is a list nobody opens. | analysis | what its tracks are | |---|---| | `verify-bigwig-route-and-ctcf-anchor-hct116` | **RAW coverage** with a separate INPUT track (input 2.43×, CTCF 50.05×) — divides by input itself | | `k562-h3k9me1-over-the-kmt2a-bcr` | RATIO — ENCODE **fold-change-over-control** | | `q108-gse210480-h3k9me3-kmt2a-bcr` | RATIO — *"ChIP over Input ratio coverage tracks were generated using deepTools"* | | `q110-h3k9-writer-and-reader-at-the-kmt2a-locus` | RATIO — fold-change-over-control | | `q111-ctcf-anchor-matched-ranking` | RATIO — signal over control, genome mean 1.0 | | `q111-cd34-h3k9me2-over-the-kmt2a-bcr` | neither — **RSEG domain calls**, not coverage | | `audit-track-resolution-and-replicate-provenance` | neither — reads entries, not tracks | **So exactly ONE analysis is raw-coverage and there is exactly ONE incommensurable pair in the whole corpus** — HCT116 raw against RPE ratio. **And it is already labelled, in at least three places**, including verbatim: *"Note that this is a ratio-track magnitude and is not comparable with the 50.05× CTCF and 72.22× RAD21 figures recorded for this element from raw HCT116 coverage."* **The subtler question is ratio-against-ratio, and it is also already handled.** Two different normalisations are in play: ENCODE **MACS2 fold-change-over-control** in the K562 and Q110 work, and **deepTools ChIP-over-input** in Q108 and Q111. Those are not the same quantity. The K562 analysis says so in its own results, unprompted: *"RPE values are ChIP-over-Input RATIO tracks; these K562 values are fold-change-over-control. Both are ratios to a control, so the comparison is of **DEPLETION DIRECTION and rough magnitude, not of absolute scale**."* **Verdict: the class is real, it occurs once, and it was labelled before anyone audited it.** The four king logs that named this as an outstanding risk were describing a gap that the beats had already closed at the point of measurement. That is worth knowing in both directions — the track discipline is better than the ledger's own worry list implied, and **a gap named four times without being run is a defect in the ledger, not in the corpus**.
| status | run |
|---|---|
| run date | 2026-09-22 |
| why this ran | named as an unaudited risk in the cycle-59, 60, 61 and 62 king logs and never executed; a gap named four times is a task |
| method | every analysis directory containing a bigWig read was opened and classified by its OWN stated track provenance, rather than keyword-matched. Seven qualify. A first keyword pass misclassified two of the seven, which is why they were read. |
| absence class | measured-not-analysed |
classification
{
"raw_coverage_with_separate_input": [
"verify-bigwig-route-and-ctcf-anchor-hct116"
],
"ratio_MACS2_fold_change_over_control": [
"k562-h3k9me1-over-the-kmt2a-bcr",
"q110-h3k9-writer-and-reader-at-the-kmt2a-locus"
],
"ratio_deepTools_chip_over_input": [
"q108-gse210480-h3k9me3-kmt2a-bcr",
"q111-ctcf-anchor-matched-ranking"
],
"not_coverage_at_all": [
"q111-cd34-h3k9me2-over-the-kmt2a-bcr (RSEG domain calls)",
"audit-track-resolution-and-replicate-provenance (reads entries)"
]
}the single incommensurable pair
{
"raw": "HCT116 Canela tracks - input 2.43x, CTCF 50.05x, RAD21 72.22x at the element",
"ratio": "RPE Gray tracks - CTCF 4.7x the chr11 mean at the same element",
"already_labelled_in_at_least_three_entries": true,
"verbatim_from_one_of_them": "Note that this is a ratio-track magnitude and is not comparable with the 50.05x CTCF and 72.22x RAD21 figures recorded for this element from raw HCT116 coverage."
}the subtler case ratio against ratio
{
"the_problem": "MACS2 fold-change-over-control and deepTools ChIP-over-input are not the same quantity, and both are in use here",
"already_bounded_at_the_point_of_measurement": "the K562 analysis states in its own results: 'RPE values are ChIP-over-Input RATIO tracks; these K562 values are fold-change-over-control. Both are ratios to a control, so the comparison is of DEPLETION DIRECTION and rough magnitude, not of absolute scale.'"
}what this audit cannot see
[ "IT TRUSTS EACH ANALYSIS'S DESCRIPTION OF ITS OWN TRACKS. If an analysis mis-states what it downloaded, this inherits the error, and nothing here re-derives track type from file headers.", "It covers analyses in THIS repository. An entry quoting a fold-enrichment from a published figure, rather than from a track this project read, is outside its scope and is the larger population.", "It says nothing about whether any individual figure is correct - only about whether two figures placed side by side came from commensurable pipelines." ]
- This is a measurement-hygiene audit and nominates nothing clinically. Its only bearing on an intervention is to keep one number out of a dossier: the 50.05x CTCF and 72.22x RAD21 figures at the KMT2A element are raw-coverage HCT116 measurements and must not be presented beside the 4.7x ratio-track figure as though the element's occupancy had been measured at two magnitudes. They are the same element measured on two scales.
- How many entries quote a fold-enrichment taken from a PUBLISHED FIGURE rather than from a track this project read? That is the larger population, it is entirely outside this audit, and a paper's fold-enrichment carries its own unstated normalisation.
- Should track provenance be a required field on any analysis that reads a bigWig? It is currently recorded in prose, inconsistently, which is why a keyword pass misclassified two of seven. A `data.track_type` with a controlled vocabulary would make this audit a one-line query instead of a reading exercise.
- Are MACS2 fold-change-over-control and deepTools ChIP-over-input comparable even in DIRECTION at low signal? The K562 note restricts the comparison to direction and rough magnitude, which is the right instinct, but nobody has checked whether the two normalisations can disagree in sign where coverage is sparse.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
Seven analysis directories in this repository perform a bigWig read. Read individually rather than keyword-matched, exactly one uses raw coverage with a separate input track (verify-bigwig-route-and-ctcf-anchor-hct116, input 2.43x against CTCF 50.05x), four use ratio tracks - two ENCODE fold-change-over-control and two deepTools ChIP-over-input - and two are not coverage reads at all, one being RSEG domain calls and one an audit of entries. The single raw-against-ratio comparison in the corpus, HCT116 50.05x against RPE 4.7x at the same element, is explicitly labelled as incommensurable in at least three entries. The cross-pipeline ratio comparison between fold-change-over-control and ChIP-over-input is bounded in the K562 analysis's own results to depletion direction and rough magnitude rather than absolute scale.
What this evidence cannot show
THIS TRUSTS EACH ANALYSIS'S ACCOUNT OF ITSELF. Track type was taken from what each script and results file says it downloaded, not re-derived from file headers, so a mis-stated provenance propagates. A first keyword pass misclassified two of the seven - it missed that the HCT116 analysis is raw-coverage - which is why all seven were opened, and is also a warning about how fragile the classification is. The audit is confined to tracks THIS project read; figures quoted from published papers are the larger population and are out of scope.
Evidence that argues against this (2)
- A clean negative on a risk the project named four times is the outcome most likely to mean the audit was too narrow rather than that the corpus is clean. The scope here is seven directories; the worry as originally written was about ENTRIES quoting magnitudes, and an entry can quote a fold-enrichment from a paper's figure with no analysis directory behind it at all. That population was not enumerated and could be large.
- Finding the warnings already in place is partly an artefact of who wrote them. The same coordinator who flagged the risk in four king logs also wrote or reviewed the entries that carry the labels, so this is closer to checking that a known concern was acted on than to an independent audit of whether the concern is handled generally.
this is the third class that entry named and did not audit
its results carry the cross-pipeline note bounding fold-change-over-control against ChIP-over-input to direction and rough magnitude
one of the four ratio-track analyses classified here