The Rule
Gregor Mendel counted peas until the math was undeniable: traits come in pairs, one from each parent, and they sort predictably. Set the parents below and watch the offshoot fill in. This is the pattern you'll need — because the whole lab is about the place it breaks.
NULL does not speak. But NULL lines up two rows of pebbles, one row per parent, and sorts the offspring into four little piles — the same four piles, every time. NULL taps the predictable pile. Then NULL looks at you. Remember this feeling of "of course." You're going to lose it in two tabs.
Two letters, A and a. Each parent hands down one of its two copies, picked at random. Cross two Aa parents and you get the famous 1:2:1 — one AA, two Aa, one aa — every time, in the long run. The trait shows up exactly where the letters say it should. Nothing appears that wasn't already in a parent. Nothing one allele does changes its partner.
The Mental Model
Think of the DNA letters as source code. Mendel described how that code gets copied and handed to the next machine: clean file transfer, run it, get the expected output. For 150 years that was the entire story of inheritance, and it earned its place — it is right, as far as it goes. The next tab is about the part it doesn't reach.
The Layer
There's a second layer sitting on top of the code. It doesn't change a single letter. It decides which letters get read out loud and which get whispered to silence. Add methyl marks to the promoter and watch the gene go quiet — without one base ever changing.
NULL does not speak. But NULL holds up a book, then a highlighter, then a roll of tape. NULL covers a paragraph with tape. The words are still there. You just can't read them anymore. NULL points: the letters didn't change. The reading did.
Methylation is a tiny cluster of carbon and hydrogen that clips onto the gene's on/off switch — the promoter. Pile enough of them on and the cell stops reading the gene. The sequence underneath is untouched; you could spell it out letter for letter and find nothing wrong. What changed is whether anyone is listening. This is the layer that runs on top of the code, at what a programmer would call runtime.
Why It's Not in Mendel's Ledger
Mendel's rules govern the letters. Methylation isn't a letter — it's a note in the margin. The old assumption was that even if these marks got passed down, they'd still sort and dominate and recede just like genes do. That assumption is exactly what the next tab puts to the test. And it's where the field gets genuinely contested: in mammals, most of these marks are supposed to be wiped clean twice on the way to the next generation. So how does anything survive? Hold that question.
The Break
Here are two parents. Both clean — no methylation on this gene, either copy. By Mendel, the offspring inherits what the parents had: nothing. Before you run the cross, commit to a prediction. The pile you tap is the whole point of this lab.
NULL does not speak. But NULL sets down two blank tiles — one per parent, both unmarked. NULL slides them together. Then NULL stares at the offspring tile for a long moment, and very slowly turns it over to show a mark on both faces that came from neither parent. NULL does not explain. NULL never explains the good ones.
What Just Happened — "Emergent" Methylation
The offspring is silenced on a gene neither parent carried silenced. It didn't come from mom. It didn't come from dad. It wasn't in the code. The act of combining two genomes generated a state that existed in neither input. The researchers' own word for it was that the methylation seemed to show up out of nowhere. In their three-generation mouse data, this happened 54 times.
And the Stranger One — Paramutation
Hit "Show paramutation" and watch one allele's mark spread to its clean partner sitting right next to it. No DNA change. One copy rewrote the behavior of the other. This had been seen in plants and flies for decades but never confirmed as a natural event in a mammal — until it turned up at a gene called Capn11, which helps build sperm, in a stretch of repetitive DNA known to react to the environment. In the whole dataset, paramutation showed up in just 3 cases. Small number. Big deal.
The Tally
Across three generations of mice, about 522 inheritance events — roughly 7% — broke Mendel's rules: new imprinting, the 54 emergent marks, and the 3 paramutations. 93% behaved. The lab is named for the rest.
The Commit
You saw a mark appear from nowhere and an allele rewrite its neighbor. The seductive story writes itself: the body remembers, and it teaches its children. Before you sign that story, a skeptic is going to hand you a knife. This is the second commit — and the honest one.
NULL does not speak. But NULL holds up the offspring tile from the last tab, then holds up a second tile printed with the parents' raw DNA letters — and the two strains' letters don't match. NULL sets them side by side and tilts its head. NULL is asking the only question that matters here: did the mark appear from nowhere, or from the letters you weren't looking at?
The Skeptic's Three Knives
1 · The confound is baked in. To read each allele separately, the study had to cross two genetically different mouse strains — you need the letter-differences to tell mom's copy from dad's. But letter-differences can direct methylation. So a "non-Mendelian epigenetic" pattern could be ordinary genetics acting through sequence, wearing an epigenetic costume. The method's strength is its deepest confound.
2 · Is it even transgenerational? Mammals erase methylation twice between generations. The canonical objection (Heard & Martienssen) is that marks shouldn't survive that at all. And a real "memory" has to show up in generations that were never exposed — not just in a single cross.
3 · Small n, and maybe directionless. The headline rests on 3 paramutations and 54 emergent marks. And even where epigenetic inheritance is undisputed (plants), the marks are mostly random, not adaptive. The mechanism can be completely real and still be noise, not wisdom.
The same offspring tile, two readings
This is Knife 1 made tangible. Here is the identical result from Tab III. Read it as an epigeneticist, then read it as a skeptic. The data doesn't change. The story does.
Real / Mine — the honesty split
Real: the 2026 mouse study (Davidovich, Feinberg, Threadgill, Hansen et al., Nature Genetics), the ~7% / 522 figure, the 54 emergent marks, the 3 paramutations, Capn11, long-read sequencing, and the skeptic literature it has to answer to. Mine: the OPA framing, the two-stage commit, the malware-and-emergence stylization, and the characters — Patel, Martinez, Volkov — who carry the argument. The line between them is in the modal at the bottom-left, and it does not blur. The science is real. The university is made up. Both are labeled.