E5: The Precision Covenant

The post you were about to write

Imagine you have just read a news story about a new study. It found that people who do something you already disapprove of are somewhat more likely to report a bad outcome. The story says the effect is modest, that the study shows an association rather than a cause, and that the researchers want larger trials. You open a reply box, or a group chat, and type: "Science confirms what we all knew."

None of that is a lie, exactly. But the study confirmed nothing, and what "we knew" was a bigger claim than the researchers made. You have passed a careful finding on at a higher voltage than it left the lab. If someone on the other side did that with a study you disliked, you would notice at once. When you do it, it feels like summarizing.

The move

The Precision Covenant is a tool, which on this site means a mental operation you can run in real time. Most of the tools here are aimed mainly at what you read. This one is aimed at what you write: the post, the reply, the argument at dinner. It is a promise to whoever reads you: you will say the strongest version of what you actually mean, without hiding in vagueness, and nothing stronger.

It has two halves. The first is quantifying where you can: "most," "some" and "a few" are different claims, and a number beats all of them when you have one. The second is marking your confidence honestly, with "I think" or "the evidence suggests" when that is the truth, instead of borrowing a certainty you do not have.

Running it takes four steps.

1. Before you send or publish, find the sentence that carries your main claim.
2. Ask what you actually have behind it: a number, a study, a story you heard, a feeling. Name it to yourself.
3. Rewrite the claim so that its size and its confidence match what you have. Replace "always" with how often, "proves" with what the evidence shows, "everyone" with who.
4. Check that you have not overcorrected into mush. If you believe something firmly and have good reason to, say it plainly. A claim hedged until it cannot be wrong also cannot be checked.

Why it works

The tool belongs to the family the site calls Language Precision: language as power, handled carefully. The site counts it among the five tools that carry the most weight across every subject, because it is the one most squarely aimed at what a writer puts out rather than what a reader takes in.

In March 1951 a US National Intelligence Estimate, a formal forecast agreed by the American intelligence agencies, judged that an attack on Yugoslavia that year "should be considered a serious possibility." In his 1964 essay "Words of Estimative Probability", Sherman Kent, of the CIA board that prepared the estimate, recalled being asked by a State Department planning chief what odds the phrase meant. Kent's answer was "around 65 to 35 in favor of an attack"; the planners "had read 'serious possibility' to mean odds very considerably lower." Among Kent's own colleagues who had agreed the wording, "the low man was thinking of about 20 to 80, the high of 80 to 20." One phrase had carried every view from unlikely to likely. Readers of public reports show a related problem, though not the same one: rather than scattering, their readings drift toward the middle. A 2014 study in 24 countries reports in its abstract (the full paper was not read for this page) that readers took the UN climate panel's probability words "as conveying probabilities closer to 50% than intended."

Does honesty about uncertainty cost a speaker trust? Less than many fear, but not nothing. In five experiments with 5,780 people, four of them online studies of people in the UK and one with readers of the BBC News website, Anne Marthe van der Bles and colleagues found in 2020 that giving a number as a range "does not substantially decrease trust in either the numbers or the source," while a verbal hedge ("it could be somewhat higher or lower") appeared to lower trust in both. Stating uncertainty "never significantly increased perceived trust." A US study of COVID-19 forecasts by Sarah Kreps and Douglas Kriner, run in May and June 2020, found small drops in support for science both when a death forecast was shown as a range and when people were told it had doubled since April. Giving the range, they write, "may be more intellectually honest, but it nonetheless comes at a cost." In a separate experiment in the same paper, a dire warning that left out the uncertainty raised support for letting the models guide policy, mostly among people who knew least about science. Putting their findings together, the authors suggest that "the more exaggerated narrative may produce short-term gains in trust," but "may come at a longer-term cost when those prognostications are falsified." They measured trust only once, so that longer-term cost is their expectation, not something they observed.

So the covenant is not a persuasion trick. Sometimes the right-sized claim convinces fewer people, at least at first. What it buys is a record that holds up, and readers who can decide for themselves how much weight to give you: what the site elsewhere calls "precision as an act of respect for the reader." And if you say "always," one counterexample defeats you, taking down the smaller claim you could have defended.

Using it well

The legitimate use is quantifying claims where possible and marking your own confidence honestly instead of borrowing certainty you don't have.

A real case shows the covenant kept, after two statements that broke it in opposite directions. Each is judged here against what was known when it was made, not against what came later.

Early in the COVID-19 pandemic, a statement in The Lancet published on 19 February 2020, chiefly expressing solidarity with Chinese scientists, said its signatories "strongly condemn conspiracy theories suggesting that COVID-19 does not have a natural origin," and that scientists who had studied the virus's genome "overwhelmingly conclude that this coronavirus originated in wildlife." Reporting what most scientists then concluded was fair. In this page's reading, the overstatement was in the first sentence's reach: it filed every origin other than a natural one, including an accident at a laboratory, under conspiracy theories, which treated an open question as closed. Ten weeks later the US intelligence agencies were still listing a laboratory accident as one of two possibilities to examine.

On 30 April 2020, CNBC reported, President Donald Trump was asked whether he had seen anything giving him "a high degree of confidence" that a Wuhan laboratory was the source. "Yes, I have," he said; asked what, "I can't tell you that. I'm not allowed to tell you that." His overstatement ran the other way: he said he held high confidence and offered nothing a listener could check. CNBC also reports that, asked if he was suggesting the virus was not naturally occurring, he said "No," and that he spoke of "a lot of theories." The same day, the Office of the Director of National Intelligence, which speaks for the US intelligence agencies, said that they would keep examining "whether the outbreak began through contact with infected animals or if it was the result of an accident at a laboratory in Wuhan."

The office's unclassified summary of 27 August 2021 is a model of the covenant. "All agencies assess that two hypotheses are plausible," it says. Four agencies and the National Intelligence Council favored natural exposure "with low confidence," one favored a laboratory incident "with moderate confidence," and three could not choose. Each side of the split carries a label saying how sure it is, and the summary names the evidence that would change its view.

Those views have moved since. In January 2025, the Associated Press reported, the CIA released an assessment that a laboratory origin was more likely, while saying it had "low confidence" in that conclusion and that both origins "remain plausible." The AP added that an Energy Department report had favored a laboratory leak, also with low confidence, and that the FBI's director had said in 2023 that his agency believed the virus "most likely" escaped from a laboratory. This page takes no view on where the virus came from. The 2021 summary is here as an example of how to word a judgment.

This site tries to keep the covenant too. The G12: Fixed Pie or Growing Pie page calls a popular evolutionary explanation "a plausible story, but so far a story." Whether the site keeps it consistently is fair for readers to check.

When it becomes a weapon

The weaponized form is demanding forensic precision from an opponent while granting yourself loose language. Its signature phrase is "well, technically," used to derail an argument about substance into an argument about wording.

It works because it looks like the covenant: imprecision is a real fault. The giveaway is the asymmetry. Two cases from 2021, one from each side of American politics, show it.

On 15 February 2021, Senator Ron Johnson, a Wisconsin Republican, said on Milwaukee radio that the January 6 attack on the Capitol "didn't seem like an armed insurrection to me." As the Associated Press reported, he asked: "When you hear the word 'armed,' don't you think of firearms?" The AP noted that court filings described guns, bombs and other weapons found on people who stormed the Capitol. Johnson condemned the violence in the same interview. But later in it, accusing Democrats of hypocrisy over the protests of summer 2020, the man who had held his opponents to a strict reading of one word described them loosely: "their side is the one that's been saying defund the police."

On 6 July 2021, Randi Weingarten, president of the American Federation of Teachers, a teachers' union and so an advocacy source, told its conference in a speech the union published: "Let's be clear: critical race theory is not taught in elementary schools or high schools. It's a method of examination taught in law school and college that helps analyze whether systemic racism exists—and, in particular, whether it has an effect on law and public policy." As a definition of an academic theory, that is defensible, and her complaint that the label was being stretched may have had something to it. But it does not settle what schools were teaching about race, which was the question in dispute. And in the next sentences the precision ran out: "culture warriors are labeling any discussion of race, racism or discrimination as CRT to try to make it toxic. They are bullying teachers and trying to stop us from teaching students accurate history."

Neither speaker is shown here to be wrong about the larger question. The pattern is the double standard: one word held to a strict reading, the other side described in sweeping terms.

Some signs help tell the legitimate form from the weaponized one:

• The legitimate form is applied first to your own claims. The weaponized form is applied only to an opponent's.
• The legitimate form asks for the precision the question needs. The weaponized form asks for more precision than any claim could supply, and moves the bar when it is met.
• The legitimate form, having corrected the wording, goes on to engage the strongest version of what the person meant. The weaponized form treats the correction as the end of the argument.
• The person using the legitimate form would accept the same correction of their own loose phrasing without complaint.

Take the weaponized form as seriously as the legitimate one: it is the version that feels best to use. Catching an overstatement gives a sharp pleasure that answering the real point seldom does.

Run it on your own side first

This tool is already pointed at you. The test is whether you keep it when it costs you something. Try these:

• Find the last thing you posted or said about a subject you care about. Is there a word in it, such as "always," "everyone," "proves" or "destroyed," that you could not defend if someone asked you to?
• Think of a claim from your own side that you have repeated often. Do you know the number behind it, or only the adjective?
• Of the senator and the union president above, which one did you find yourself defending? Would you accept the same correction of your own side's slogans?

If the honest answer to the second question is "only the adjective," you are in large company.

Grammars it works against

A grammar, on this site, is a recurring pattern of argument that turns up across many different subjects. The Precision Covenant is also one of four general tools that the site treats as working against every grammar (a different list from the five that carry the most weight), because nearly every grammar involves claims made bigger or surer than the evidence allows. It matters most in three places.

• G11: Contested Epistemic Authority. This grammar asks who gets believed, and on what basis. The site's entry on it warns of "expert overclaiming that spends down public trust" and contrasts the confident amateur with "the boring specialist with error bars." The covenant asks the same of both.
• G9: Language as Territory. Fights over which word applies, such as "armed" or "critical race theory," are often fought with words chosen to carry more than they can bear. Asking each side to say what it means, at the size it means it, often shows whether a real disagreement remains.
• G6: The Hidden Hand. Explanations built on secret coordination borrow credibility from the confirmed cases, and the people who debunk them sometimes overreach by dismissing every claim as conspiracy thinking. The covenant asks both to say exactly what the evidence shows about this case, and nothing more.

Related tools

• E2: The Operationalization Demand asks what, specifically, would count, and measured how. It is the covenant's reader-side twin, with a matching weaponized form.
• B1: The Effect-Size Reflex asks how big an effect is, for whom, compared to what. It supplies the numbers that let you keep the covenant.
• F4: The Scout Report asks you to declare whether you are advocating or inquiring. The covenant asks you to declare how sure you are. Together they tell your reader how to read you.
• A4: The Mirror Check asks whether you scrutinized a claim or just liked its conclusion. It is a strong guard against the covenant's weaponized form, which is a double standard about wording.