Four Gates — Draft, Self-Review, Source-Check, Balance-Check

Whether you can let someone release materials alone is not decided by the final polish. It is decided by whether they pass four gates — draft, self-review, source-check, balance-check — in order, skipping none. The way they pass is the evidence.

Gates are inspection stations in a line

Picture airport security. One station to put your bag down, one for the X-ray, one for the metal detector, one where a person looks with their eyes. The stations are separate and the order is fixed. You cannot skip one to reach the next. The work of making materials (the explanatory documents handed to doctors and patients) has the same kind of ordered gates. This section splits it into four: G1 drafting, G2 self-review, G3 source-matching, G4 balance-check.

What matters is not only whether all four were passed, but where a person tends to get stuck. The place they stop tells you their weak point. Someone you can trust to work alone is someone who sets up the four stations, runs through them, and notices on their own when they stop. Conversely, someone who always walks straight past one station is not yet ready to work alone. Walking past is not always laziness; often the person believes "I covered that." So the gates must leave an outside-visible trace of passage, separate from the maker's good intentions.

G1 Draft — the station that gives it shape

In cooking terms, this is cutting the ingredients before they go in the pot. The question is simple: have you laid out once, properly, what to say, to whom, and in what order? The common stumble at the draft station is fixing the conclusion first. When "this drug is superior" is already in your head as the answer you want to sell, you start picking material toward it. This is one psychological driver, motivated reasoning (the answer you want to sell comes first, and how you read the data is pulled toward it). The person does not feel they are lying. That is what makes it dangerous.

What G1 stops is the conclusion-first design itself. In a reported case, a maker prepared no material for the primary endpoint (the point the trial most wanted to confirm) and explained only a secondary item that showed a significant difference (a gap too large to be explained by chance). When you lay the draft's blueprint on one sheet and the most important item is missing, you can catch it at G1. With the habit of asking at the draft stage, "Am I starting from the primary endpoint?", you avoid from the start the shape that pushes only the convenient result to the front. Think of the draft not as a station that races for speed, but as one that keeps the order from going wrong.

G2 Self-review — the station where you doubt your own draft

This is close to the writer re-reading their own proof print (a test print made to catch errors before printing) with a red pen. But the hard part of self-review is that your own writing looks correct to you. The "my material is fine" conviction seen in the seventh piece shows up most at this station. Because you wrote it, you know its content, and that comfort stops the checking hand.

In a reported case, a maker wrote in a document that "the risk of death does not increase" with no evidence for it. When checked, they answered, "It is not clear at present, so I wrote it this way." They had not proven the absence of risk, yet they stated it flatly. In the same set of cases, an important potential risk was stressed only as "a low risk is expected." Both are signs of local rationalization (justifying a deviation in one part by telling yourself "just here is fine"). At G2 you ask, sentence by sentence, "Where in the source is this written?" If a flat assertion has no source, that sentence stops at G2. This is where you catch words like expected, probably, or should slipping in without source backing.

G3 Source-matching — the station of returning to the original paper

This is like a health check where you do not just stare at the numbers but match them against the table of reference values. G3 is the station where you take each claim in the draft back to the original material (the paper, the package insert, the review report) and check whether it truly says that. The "can return to the source" ability this section calls the floor (the minimum condition that must always be met) is tested here.

Among reported cases, one set a graph's vertical axis to start at 0 instead of its proper 0.8, making two drugs look no different. Another argued an effect from a graph of just 9 cases (4 versus 5) with no statistical analysis. Both stop if you return to the source at G3. When you go back to the original numbers, the original axis, the original case count, any gap from how the draft showed it is the evidence. Think of source-matching as the station that peels off the clever presentation once and returns to the bare fact. A person who passes here can, when asked, point with a finger: "this sentence is this row of this table in this document." A claim you cannot point to is still below the floor.

G4 Balance-check — the station that weighs benefit against caution

Picture a balance scale (a tool with a pan on each side to compare weight). Put the benefit on one side, the side effects and cautions on the other, and see whether the tilt matches the facts. G4 is the last station, where you confirm that even if each sentence matches its source, the whole has not lost its balance. Each sentence can be correct, yet if every caution is made small, the impression that reaches the reader drifts from the facts.

In a reported case, although pre-administration screening (a test done beforehand to check who may use the drug) was mandatory, the product information summary wrote "screening or testing is not needed" as if it were a product feature. In another, a side effect that should have been a warning (a certain component becoming excessive) was explained as a strength, "you can replenish that component." The first is the sin of omission (defending yourself by not saying it, not saying it until asked); the second is rewording a side effect. Read as one sentence each looks unbroken, but on the scale the caution pan is too light. G4 corrects this tilt.

The four gates are built so that skill at appeal (ease of understanding) cannot fill, after the fact, a hole left at an earlier station. A claim that cannot return to the source at G3 remains below the floor no matter how well G4 tidies the balance. The non-compensatory gate (a design where the whole fails if even one part is missing) refers to this order and this floor. So when you judge a person, look at which of the four stations they set up with their own hands, and where they can stop themselves.

The four gates are tools for judging a person by order and floor, not by an average of polish. The draft stops conclusion-first, self-review doubts flat assertions, source-matching returns presentation to bare fact, and the balance-check restores the weight of the cautions. No station can be skipped. Someone you can trust to work alone is someone who sets up these four stations and notices on their own where they stop.

Deviation is not the act of bad people but a circuit ordinary makers fall into under pressure. So use the four gates not to punish, but as a mechanism to watch that circuit in yourself.