How We Work · Daylight Brief

What this publication is — and isn’t

Daylight Brief is an advocacy publication. We argue that specific things should change, and we’re open about it. We are not neutral, and we won’t pretend to be — a page that claims to have no viewpoint while clearly having one is just a less honest version of what we do.

What we offer instead of neutrality is a promise we can actually keep:

Every factual claim we make is traceable to a source you can check.
We show our reasoning — where a fact ends and our interpretation begins.
We engage the strongest case against us, not a strawman.
We correct our mistakes fast and in the open.

If we ever fail one of these, that’s a failure by our own standard, and you’re entitled to hold us to it.

The seven commitments

Everything below is downstream of these. They’re the standard; the process is how we meet it.

Every factual claim is sourced. If we can’t point to where a claim comes from, we don’t publish it.
We separate fact, inference, and value — visibly. A number, what we infer from it, and what we conclude we ought to do are three different things, and we never let them blur into one another on the page.
We argue our conclusions; we don’t assert them. No claim skips its justification for being “obvious” or “common sense.” If it’s true, we can show why.
We engage the strongest counterargument. A piece that hasn’t survived its best opposition isn’t finished.
We preserve uncertainty. Where the evidence is genuinely contested, we say so rather than pick the convenient number.
We correct in the open, fast. Errors get fixed visibly, with a note saying what changed and when.
We disclose our use of AI and stay accountable. AI helps us work; it does not decide what’s true, and it never gets the last word on whether a source is real.

How a piece gets made

Every piece moves through six stages. The rule that governs all of them: AI assists at every stage, but a human owns verification, and no fact enters a draft that wasn’t traced to a source a human confirmed.

Stage 0 — Frame the question

Before any sourcing, we break the thesis into its component claims and sort them into three buckets:

Empirical claims — checkable facts and numbers (“aid to X totaled $Y in year Z”).
Causal claims — assertions that one thing produces another (“this policy caused that outcome”). These are the hardest to establish and the easiest to get wrong; we flag them for extra scrutiny.
Value claims — statements about what should happen (“this ought to change”).

This separation is the backbone of everything. If we can’t tell you which parts of an argument are facts, which are inferences, and which are values, we’re not ready to write.

Checklist:

Thesis stated in one sentence.
Sub-claims listed and tagged empirical / causal / value.
Causal claims flagged for evidence review.

Stage 1 — Sources in

We gather sources and classify each one against our source hierarchy (below). For every source, we record the specific claim it actually supports — not the topic it’s “about,” the claim it substantiates. A source that gets stretched past what it literally says is flagged here, before it can do any damage downstream.

Checklist:

Each source assigned a tier.
Each source’s actual supported claim recorded.
Advocacy / think-tank / partisan sources flagged with their known lean.
Any source being asked to carry more than it says — flagged or dropped.

Stage 2 — Bind every claim to a source (the verification gate)

This is the stage that makes or breaks our credibility. Every factual sentence that will appear in the piece must bind to:

a specific source, and
a specific location in it — a page, a table, a quoted line — not just a URL.

A human reads the cited passage and confirms it says what the claim says. If that exact supporting text can’t be produced, the claim does not ship. No exceptions, and no “the AI said it was in there.” This single rule eliminates the overwhelming majority of the ways a cited-facts publication can embarrass itself.

Checklist:

Every factual sentence bound to source + exact location.
A human has read each supporting passage and confirmed the match.
Any claim without a confirmable passage: cut or rewritten to what the source actually supports.

Stage 3 — The adversarial pass (not optional)

Before drafting, we build the case against our own piece:

the strongest version of the opposing argument,
the best data the other side would cite,
the weakest link in our own chain of reasoning.

Then we make sure the piece answers it. This is our defense against the most insidious failure mode in advocacy: a piece where every individual fact is true but the selection is stacked, so the reader is steered without being lied to. Propaganda ignores the counter-case. We engage it — and we engage the real one, not a caricature we can knock down. A conclusion that can only stand by pretending the other side has nothing isn’t a conclusion we’ve earned.

On some issues, once we’ve done this honestly, we may well conclude that the organized opposition is driven by narrow interests rather than by the merits. That’s a legitimate finding — but it has to be a finding this stage reaches by engaging the argument, never a premise we start from. The conviction that “our case is so strong only bad-faith could dispute it” feels identical whether the case is right or wrong, so it can’t be trusted on its own; only the counterargument, taken seriously, can settle it. If the strongest version turns out to be genuinely strong and made in good faith, this stage has to be able to say so — even when that’s inconvenient for us.

Checklist:

Strongest opposing argument written out in good faith.
Best opposing evidence identified and addressed in the piece.
Our weakest link named and either shored up or openly acknowledged.
Any judgment that the opposition is bad-faith or interest-driven is a conclusion we reached by engaging the argument, not an assumption we began with.

Stage 4 — Draft

Now we write. Throughout the draft:

Every factual sentence keeps its citation attached.
Fact, inference, and value stay visibly distinct — the reader can always see where the sourced facts end and our argument begins.
Rhetorical and motivational language is allowed and welcome, but it never gets dressed up to look like a claim of fact.

We write with force and clarity. Force comes from the evidence and the moral stakes — never from telling the reader that disagreement is illegitimate.

Checklist:

Citations intact on every factual sentence.
Fact / inference / value visibly separated.
Persuasive language doesn’t masquerade as fact.

Stage 5 — Integrity & liability review

A final pass before publication:

Defamation check. Every hard factual claim about an identifiable person or company is airtight and precisely worded. If it isn’t provably true and exactly stated, it’s cut or softened to what we can defend.
Uncertainty check. Where the data is genuinely contested, the piece says so.
“Common sense” check. Any conclusion resting on the phrase “obvious” or “common sense” has been replaced with the actual evidence and reasoning. If it’s true, we’ve shown why; if we can’t show why, it doesn’t run.
Correction-readiness. Every claim is traceable, so that if we’re wrong we can fix it within the hour and show exactly what changed.

Checklist:

Every claim about a named person/entity verified and precisely worded.
Contested points flagged as contested.
No conclusion smuggled in as self-evident.
Full source trail intact for fast correction.

Our source hierarchy

We weight sources roughly in this order, and we tell you when a claim rests on a weaker tier.

Primary and official records — government data, official filings, court records, original datasets, direct quotes in full context.
Peer-reviewed research — with attention to whether findings are replicated or contested.
Reputable secondary reporting — established outlets with correction records and editorial accountability.
Aggregators and summaries — used to find primary sources, not as the final word. We cite the underlying source, not the aggregator.
Advocacy and think-tank material — usable, but always flagged with its known orientation, and never as the sole support for a contested factual claim.

Where sources conflict, we say so and show the range rather than quietly picking the one that helps us.

How we handle “most informed people agree” claims

This is a powerful argument and we use it — but carefully, because it’s easy to abuse.

We verify what a study actually found. “People informed about an issue tend to favor conclusion X” is a much stronger and rarer finding than “people hold unfavorable views,” and we don’t upgrade the second into the first.
We keep our distribution distinct from our evidence. Our own outreach is broad and untargeted — we inform people who didn’t seek the topic out, rather than preaching to those who already had views. That’s a real strength, and it’s closer to a clean test than to echo-chamber measurement. But it doesn’t retroactively fix studies that measured only the already-aware: “people who chose to get informed agree” can reflect who self-selects into being informed, not what information itself produces. When we cite such a study, we say what it actually measured.
We don’t confuse informed opinion with majority will. If most people are unaware of an issue, there is no settled majority position on it yet — only a projection of what people might conclude if informed. We can argue that projection; we won’t state it as an accomplished fact. And we hold that projection loosely: if broad, honest information surfaces that the public, once informed, doesn’t share our conclusion, that is a real result and we treat it as one.

Corrections

When we get something wrong, we fix it visibly. A correction note states what was changed, what it originally said, and the date. We treat a fast, transparent correction as a strength, not an embarrassment — it’s the clearest proof that our claims were traceable in the first place. If you think we’ve gotten something wrong, tell us, and point us to the source.

Our use of AI

We use AI tools to help us decompose arguments, organize and classify sources, extract supporting passages, stress-test our own reasoning, draft, and check for consistency. We are transparent about this because it’s part of how we keep our standards high, not something to hide.

What AI does not do here:

It does not decide what is true.
It does not introduce a source or a fact a human didn’t provide and verify. Every citation traces to material we supplied and confirmed against the original.
It does not get to strengthen a claim beyond what its source supports. Where our tools try to smooth over a hedge or upgrade “some evidence suggests” into “evidence shows,” we catch it and put the uncertainty back.

A human is accountable for every published word.

Where we stand

We’re for putting the interests of ordinary citizens ahead of the concentrated interests — corporate, political, and foreign — that we believe too often set the agenda without the public’s informed consent. That’s a viewpoint, openly held. Our commitment isn’t that you’ll agree with our conclusions. It’s that when you disagree, you’ll still trust our facts — because we showed our work, engaged the best case against us, and gave you what you need to decide for yourself.

And we mean something specific by “informed consent”: our aim is an informed public, not a public that agrees with us. If informing people surfaces that the majority, knowing the facts, don’t share our conclusion, that is a result we accept. The point was never to win the argument; it was to make sure the decision gets made in the open, on the facts — instead of by whoever lobbies loudest while most people aren’t looking.