4D DNA Blueprint #3 — How to read a genome without fooling yourself

[ Where we are ]

We have the idea (position is written, quantity is run) and the vocabulary (cis/trans, gamma/phi, element vs boundary). Before we read a real sequence, we install the rules that keep the reading honest. This is not throat-clearing. The single most common way to be wrong about a genome is to mistake a pattern for a cause, and one rule here exists only to stop that. A reading method is only as trustworthy as the mistakes it refuses to make.

3.1 The constitution, in three articles

Everything in this series is bound by three commitments.

I. Evidence, not authority. A claim earns standing only from the structure of the data and the direction of cause — not from who said it, not from how elegant it sounds.

II. The burden. Every claim must be defined without circularity, must survive the removal of confounders, and must be split explicitly into what is fact and what is hypothesis. If removing a confounder removes the result, it was the confounder’s result, not yours.

III. Honesty toward the result. We follow the data even when it refutes our own prior claims. A predicted negative is a result. A correction is kept visible, not quietly erased. (You will see real corrections in this series, labeled as such — that is a feature, not an accident.)

3.2 Article IV — the rule that throws out most “discoveries”

Now the hard one, and the one that makes this method different.

Article IV. Counting similarities is not admissible as proof of an interpretation.

Here is the trap it guards against. It is wonderfully easy to line up two sequences, count how many short words (k-mers) they share, compute a conservation score, and announce that the shared bits “must be functional.” This feels like evidence. It is not. Similarity counting tells you that two things resemble each other; it does not tell you what either one does, or that your interpretation of it is right. Resemblance is not function, and a high score is not a mechanism.

So in this series, shared k-mers, conservation percentages, motif-match counts, and the like are ruled out as the basis of any claim about what a region is or does. What is admissible? Two things: (1) the interpretation pipeline itself — feeding a sequence through the defined reading and seeing what structure it returns — and (2) hard coordinate, length, and geometry facts (where a gene body starts and ends, how long a region is, the shapes the pipeline computes). Those are not pattern-matching; they are measurements. Claims stand on those, or they do not stand.

This rule is expensive. It throws out many tempting results. That is exactly why it is worth having: a method that lets you keep every flattering pattern will eventually let you keep a wrong one.

3.3 Lock, derive, gate

The day-to-day discipline has three moves, and they run in order.

• Lock the data. Before analysis, every input is pinned: a named source, a version, and a checksum so the exact bytes can never silently change underneath a result. A number you cannot trace to a locked input is not a number you can defend.

• Derive the result. Run the defined pipeline. No hand-tuning a knob until the answer looks nice; the parameters are fixed in advance and, where possible, derived from physics rather than chosen (we will watch one such knob get derived in #5).

• Gate the output. The code checks its own result against pre-set conditions and refuses to report a result that fails its gates. A pipeline that always returns something encouraging is not measuring anything.

3.4 Reproduce, or it didn’t happen

The final rule is the bluntest. Every number in the whitepaper traces to a specific script and a specific, locked input, and re-running that script reproduces that number — bit for bit, from a fixed random seed. If a result cannot be reproduced from the bundle, it does not count, however appealing it is. This is why the paper ships with a reproducibility bundle and a map from each claim to the code that makes it: not as decoration, but as the thing that lets you — or a skeptic — check.

3.5 Try it yourself

Take any genomics claim you have read lately — in a paper, an article, a thread — and run it through one question: what is this resting on? If the answer is “these sequences are similar” or “this region is conserved,” Article IV says you do not yet have a claim about function; you have an observation of resemblance. Then ask the locked-data question: could I reproduce the number, from a named input, myself? You will be surprised how often a confident claim evaporates under those two questions. Doing that to your own ideas, before anyone else does, is the whole game.

— WHERE THIS STOPS — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — A word, a method, a rule — each is a tool, and the honest move is to state exactly what it can and cannot do. This constitution does not make the theory true; it makes the theory testable, and it makes a wrong turn visible when it happens. It deliberately rules out arguments (similarity-counting) that much of the field treats as evidence, which means some claims you have seen elsewhere will not be supportable here. That narrowing is the point. A reading tool that never fails is a horoscope, not a tool. — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

— CHECK IT IN THE PAPER — — — — — — — — — — — — — — — — — — — — — — — — — — — — — - This episode is the teaching version of the whitepaper’s Constitution, including Article IV and the line it draws between pattern-counting (inadmissible) and pipeline plus coordinate/length/geometry facts (admissible). The reproducibility bundle and the claim-to-code map are the constitution in practice. Full paper, proofs & reproducibility bundle (always-latest, now v8): https://doi.org/10.5281/zenodo.20471407 — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

Next up — #4: How a sequence becomes a shape. With the rules in place, we start the mechanism. Letters have no shape — so we follow how composition turns into a physical structure inside a cell, and meet the two knobs, gamma and phi, in action.

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — — Part of the 4D DNA Blueprint series. Full paper, proofs & reproducibility bundle (always-latest): https://doi.org/10.5281/zenodo.20471407 Project: https://jamming-physics.org/ 4D DNA Blueprint — what the sequence fixes, and what it does not. © 2026 Young Jae Lee — CC BY 4.0