[CORRECTED cycle 117 - the columns were not scaled] [CORRECTED cycle 110 - the gradient is NOT an artefact] Grading this project's own stratified-share statistic on data where the truth is known: it is not blind to a global collapse, it is DISTORTING - a pure 5.84x genome-wide loss with no redistribution whatsoever is rendered as a monotone gradient rising from the peaks into the sparse deciles, reporting a 9.70x INCREASE at D10 where the absolute change is 1.66x, and 62 of 66 strata are confidently outside the measured noise floor
also called what a share does to a known collapse · a rising gradient into the bulk is the signature of a global decrease · 62 of 66 strata outside the floor and wrong · the shape cannot carry the interpretation
This project measures chromatin as proportions - what fraction of a signal sits where. A public dataset now provides a case where the true answer is known independently: a drug wipes out a chemical mark across the whole genome. Re-running this project's own method on that dataset shows it does not simply miss the change. It reports a large change of the wrong kind - it says the signal moved out of the busy regions and into the quiet ones, by up to ten-fold, when in reality nothing moved anywhere and everything just got smaller. The shape that comes out looks exactly like a result this project published three cycles ago.
The same thing, for a scientist
**Statement.** Cycle 108 claimed a within-sample share would miss a global collapse and, in its own counter-evidence, admitted the claim was *"asserted, not demonstrated"*. Demonstrated here, on GSE149183, whose deposited per-gene columns are **already spike-in scaled** so the true global change is known: RS4;11 `EPZ5676 d9` totals **3.71e6 against DMSO's 21.67e6 — a 5.84x genome-wide collapse.** Strata are cut on `DMSO_d3`, external to the d9 contrast. **RS4;11, EPZ5676 d9, truth versus this project's statistic:** | stratum | absolute | **share** | share is | |---|---:|---:|---| | TOP | 0.056 | **0.324** | 5.8x too high | | D1 | 0.062 | **0.360** | 5.8x too high | | D5 | 0.661 | **3.859** | wrong side of 1 | | D9 | 1.094 | **6.386** | wrong side of 1 | | D10 | 1.662 | **9.698** | **reports +870%** | **`W3` — neither pre-registered outcome, and the pre-registration is logged as having failed.** `W1` (adequate) and `W2` (blind) were the two options written down. **Neither happened.** The share ratios do not sit near 1 — **only 4 of 66 strata fall inside this project's measured floors**, so the statistic reports a detectable effect almost everywhere — and they do not track the absolute ratios either. **They are large, confident, and wrong in sign in the bulk.** **The generalisable result: A MONOTONE SHARE GRADIENT RISING FROM THE PEAKS INTO THE SPARSE DECILES IS THE SIGNATURE OF A GLOBAL DECREASE.** It is what renormalisation does to a uniform loss. It is *not* evidence of redistribution, and here there is none — every stratum lost signal or held. **And this project published that exact shape.** Cycle 103 reported **0.837 / 0.810 / 0.950 / 0.990 / 1.021 / 1.037 / 1.061 / 1.081 / 1.099 / 1.120 / 1.136** from TOP to D10 and read it as the fusion's share flattening out of the peaks into the bulk. **This does not refute cycle 103, and the quantitative handle says why.** At a 1.25-1.31x global decrease the share profile here spans **0.92 up to 2.90-3.35**. Cycle 103's span is only **0.81 to 1.14** — far milder, so its gradient is compatible with a global change *smaller* than 1.25x plus whatever redistribution is real. **What is refuted is the inference: the shape alone cannot carry the interpretation it was given.** --- # ⛔ CORRECTED IN CYCLE 110 — THE CENTRAL CLAIM ABOVE IS WRONG **`share(s) = absolute(s) / G` is an exact identity** (G = the global absolute ratio), verified to a maximum deviation of **0.00204** across all 66 strata. **The share profile therefore has exactly the shape of the absolute profile.** Three statements above are false and are retracted: | claimed above | actually | |---|---| | *"a pure global collapse with no redistribution whatsoever"* | the absolute profile spans **0.056 to 1.662, a 30-fold range** — the differential across strata is large and real | | *"every stratum lost signal or held"* | **D9 gained 9% and D10 gained 66%** in absolute terms | | *"the gradient is an artefact of renormalisation"* | **the gradient is real**; renormalisation rescales it, it does not create it | **The real hazard is the LEVEL, not the shape, and it is worse than what I wrote.** A share ratio states how a stratum fared **relative to the genome-wide average** — never whether it gained or lost. When `G` is far from 1 those come apart completely: **34 of 66 strata (52%) lost signal, up to 68% of it, while reading as gains of up to +440%.** **And the consequence for cycle 103 reverses.** Since the shapes are identical, a rising share gradient **does** faithfully report that the sparse strata fared better than the peaks. **Cycle 103's shape claim is not undermined.** What no share-based beat can say is whether the strata reading above 1 *gained* or merely *lost less than average* — and that is the claim cycle 103 would need. See `ana-a-share-says-relative-to-average-not-gained-and-half-the-strata-flip-sign`. --- # ⛔ CORRECTED IN CYCLE 117 — THE DATA WAS NOT WHAT THIS ENTRY SAYS IT WAS **Two independent defects, both verified by the coordinator against the files on disk.** **1. The deposited columns are NOT spike-in scaled.** This entry asserted they were, citing the column sums. **All 1,040,292 values in both files lie exactly on a 0.04 grid (integer/25)** — only **18** exceptions, every one the `MT-TT` chrM tRNA whose window runs off the chromosome. A scale factor such as 1/9.91 would destroy that grid. **`+_dm6_spike-in` in the filename means the counts are supplied *alongside* the signal, not applied to it.** This entry's own `U3` said *"if the deposited columns were not spike-in scaled, the absolute arm is not absolute and the whole comparison collapses"* — **`U3` fired and was not noticed**, because the check used was the column sums, and a DOT1L inhibitor collapsing H3K79me2 lowers the raw sums exactly as a scaling would. **2. The RS4;11 file is two-thirds MOLM13.** **38,580 of 57,794 data rows are byte-identical to the MOLM13 file in all nine signal columns**, in a contiguous block from row 19,217 to the end. Only 19,214 rows can be RS4;11. This is a defect in the deposit as served today, not in the local copy. **What survives:** the arithmetic. `share(s) = absolute(s)/G` is an identity and the shape-versus-level distinction is unaffected. **What does not:** every statement in this entry that called the deposited values absolute, and every RS4;11-specific number. See `ana-q138-*` for the corrected absolute-scale analysis, which applies `K = dm6(DMSO)/dm6(treated)` and restricts to the 19,214 genuine rows.
Why this is worse than blindness
A blind statistic returns nothing and prompts someone to get a better one. This one returns a large number with a plausible biological story attached. 62 of 66 strata clear the measured noise floor, so every standard check this project applies — is it outside the floor? is it monotone? is it consistent across contrasts? — passes, on an artefact.
The monotonicity check is the worst offender. Cycle 103 and cycle 130 both treated a clean monotone gradient as evidence of a real ordered process. Renormalisation produces perfect monotonicity for free whenever the underlying loss is roughly uniform and the strata are ordered by abundance, because the sparse strata lose proportionally least of a fixed total.
A sibling beat already showed the mirror image of this: self-stratifying two identical libraries manufactures a monotone decile gradient of span 0.331. Two different artefacts, the same shape, and the shape was being read as signal in both cases.
What this does not license
It does not retract cycles 102-107. The assay is different (TSS-averaged sitepro over 57,794 genes, not the binned genome), the mark is different (H3K79me2, not H2AK119ub), and the drug class is different — a DOT1L inhibitor abolishing the mark DOT1L writes is the largest global effect obtainable, chosen precisely because it makes the distortion visible. U2 was written down before the run and it binds: nothing here shows that share statistics fail at realistic effect sizes.
What it does establish is a burden of proof. Any future share-based claim in this project that rests on a stratified gradient now has to say why a global change could not be producing it — and the d3 rows give the calibration to do that arithmetically rather than by assertion.
| question | what does the stratified-share statistic report where the truth is known? |
|---|---|
| discharges | the counter-evidence of ana-the-spike-in-ruler-is-deposited-in-plain-text-and-the-global-change-is-5-to-23-fold |
| source | GSE149183 deposited sitepro tables, already spike-in scaled |
| stratifier | DMSO_d3, external to the d9 contrasts |
| genes | 57794 |
| strata inside measured floor | 4 |
| strata total | 66 |
| pre registration failed | W1 adequate and W2 blind were both written before the run and neither described the outcome; W3 distorting was added post hoc and is labelled as such |
| absence class | measured-not-curated |
| bite back | This entry gives future beats a reason to distrust any stratified share gradient, including ones that are real. Applied bluntly it would retract cycles 102-107, which it explicitly does not support - the assay, mark and drug class all differ and the effect size was chosen to be extreme. If it is cited as grounds for a wholesale retraction, it will have done more damage than the artefact it describes. |
rs4 11 epz d9
{
"global_absolute_ratio": 0.1713,
"global_fold_down": 5.84,
"global_share_ratio": 1,
"absolute_by_stratum": {
"TOP": 0.0556,
"D1": 0.0617,
"D5": 0.6612,
"D9": 1.0942,
"D10": 1.6616
},
"share_by_stratum": {
"TOP": 0.3244,
"D1": 0.3603,
"D5": 3.8593,
"D9": 6.3864,
"D10": 9.6979
}
}global fold down range
[ 1.25, 5.84 ]
THE SHAPE WARNING
{
"rule": "a monotone share gradient rising from the peaks into the sparse deciles is the signature of a GLOBAL DECREASE, not of redistribution",
"cycle_103_published_profile": [
0.837,
0.81,
0.95,
0.99,
1.021,
1.037,
1.061,
1.081,
1.099,
1.12,
1.136
],
"cycle_103_read_it_as": "the fusion's share flattening out of the peaks into the bulk",
"does_this_refute_cycle_103": "NO - its span 0.81-1.14 is milder than the 0.92-3.35 a 1.25x global change produces here, so it is compatible with a global change smaller than 1.25x plus real redistribution",
"what_it_does_refute": "the inference that the shape alone establishes redistribution"
}CORRECTED CYCLE 110
{
"what_was_wrong": [
"'a pure global collapse with no redistribution whatsoever' - false, the absolute profile spans 0.056 to 1.662",
"'every stratum lost signal or held' - false, D9 and D10 gained",
"'the gradient is an artefact of renormalisation' - false, the shape is real"
],
"the_identity": "share(s) = absolute(s)/G, exact to 0.00204 across all 66 strata",
"what_survives": "the share statistic does mislead badly, but about LEVEL not SHAPE: 34 of 66 strata lost signal while reading as gains",
"effect_on_cycle_103": "REVERSED - its shape claim is not undermined, because share and absolute profiles have identical shape. What it cannot support is any claim that a stratum above 1 gained signal.",
"see": "ana-a-share-says-relative-to-average-not-gained-and-half-the-strata-flip-sign"
}CORRECTED CYCLE 117
{
"defect_1": "the deposited columns are NOT spike-in scaled - all values on a 0.04 grid, 18 exceptions, all MT-TT",
"defect_2": "38,580 of 57,794 rows in the RS4;11 file are byte-identical to MOLM13, contiguous from row 19,217",
"U3_fired_unnoticed": "this entry pre-registered exactly this failure and checked column sums instead of the grid",
"what_survives": "the share = absolute/G identity and the shape-versus-level distinction",
"see": "the Q138 absolute-scale beat"
}- Re-examine cycle 103's FLATTENING verdict with this calibration explicitly applied. Its gradient is milder than a 1.25x global change would produce, so the question is what global change its own data permits - which the input pairs on disk can bound.
- Do the same grading at a REALISTIC effect size. Everything here rests on a drug that abolishes its target mark; the d3 rows (1.25-1.31x) are the closest available and they already put 2 of 11 strata inside the floor while distorting the rest.
- Cycle 130's monotonicity argument used the same shape as evidence. It should be re-read against this result, as should any other beat that treated a clean gradient as evidence of an ordered process.
- Can the slope be inverted into an estimator? If share-profile slope scales with global change, a published profile plus a measured floor would bound the global change that produced it - turning this warning into a tool.
discharges that entry's own counter-evidence, which said the blindness of shares was asserted rather than demonstrated - and finds the failure is distortion, not blindness
CORRECTED in cycle 110: share and absolute profiles have identical shape, so that entry's monotone gradient is NOT undermined as a shape claim. What a share cannot support is any statement that a stratum above 1 gained signal rather than lost less than average.
the mirror artefact - self-stratification and renormalisation both manufacture monotone decile gradients, and both were being read as signal
uses those measured floors to score the distortion: 62 of 66 strata clear them while being wrong
share(s)=absolute(s)/G is an exact identity, so that entry's 'artefact of renormalisation' and 'no redistribution whatsoever' are both false - the shape is real and the failure is one of level
the 'absolute' arm of that comparison was an unscaled raw count-sum ratio, short by the factor K
the shape finding survives the rescaling; only the gain/loss line moves
that entry demonstrated the general form against a spike-in ground truth; this one records the normalisation actually used in GSE173599
- ← related toFour days after the fusion is destroyed, menin signal at non-fusion promoters is 27% higher than control
the 1.271 rise at non-fusion promoters is the artefact shape that entry demonstrated, which is why this observation is filed weak