The claim that kills you was true when somebody wrote it

I have been doing communications for contested proceedings for four decades. In that time I have watched a lot of campaigns come apart, and I can tell you what it almost never is. It is almost never the strategy. The strategy gets argued over by senior people in a room, for hours, and it usually ends up sound.

What fails is a fact.

Not a lie. Nobody in this chain is dishonest and nobody is bad at their job. What fails is a claim that was true enough when the person who wrote it wrote it, and then kept moving.

Where the work actually gets done

Here is the part that does not appear on any org chart. The research and the first drafts in a live fight get done by the most junior people in the building, on a deadline, under real pressure. That is not a criticism of them. It is the only way the work fits in the time available, and it is how every one of us learned.

The senior people are supposed to catch what comes back. They cannot, and it is not for lack of care. A general counsel in a live proceeding is triaging exposure. The partner is in a hearing. Nobody in that chain has an afternoon to open a fifty-nine page filing and check whether the paragraph a summary is built on actually says what the summary says it says.

So review is where the queue backs up. In the last two years everybody in this business, us included, found ways to produce more material. Almost nobody rebuilt the checking to match, and buying more production is easy in a way that buying more senior attention is not. That gap is the whole risk.

Producing more is a purchase. Checking more is a rebuild. That is why one of them happened.

The three ways a fact dies

They are boring, which is exactly why they get through. I see the same three over and over.

• Paraphrase drift. Each retelling is a fair summary of the last one. Four hops later the sentence asserts something the document never said, and every individual step was defensible.
• A roster read as a position. Somebody has a list of who is involved and treats it as a list of who is on which side. In one closed Virginia case I worked through, that single substitution reverses the read on the largest interest group on the docket: two of the biggest data center operators in the world signed on opposite sides of the same settlement. 1
• A number from a source quoting a source. The figure is real. The primary document says something narrower, or says it about a different year, or attaches a condition that got dropped in transit. This is the one that gets screenshotted back at you.

None of these are caught by proofreading, because none of them are errors on the page. They are errors in the chain behind the page.

The fix is not more review

More review does not scale and everybody has already tried it. What works is separating two things that ordinarily travel together: whether a claim exists, and whether it holds.

So every factual claim in an engagement gets its own record. What it asserts. Which document it came from, down to the page or the paragraph. Whether that document actually supports it, checked against the document and not against a summary. And a verdict, which is allowed to be inconvenient: it survives, it fails, it is contested, or nobody has established it yet.

That last verdict is the one that earns its keep. Most systems have no way to say “we do not know,” so an unchecked claim looks identical to a verified one right up until the moment somebody attacks it.

Then you put a gate in front of the message

A message cannot go out while a claim it depends on is unverified. That rule is easy to write down and genuinely unpleasant to run, because it deletes the line you most want to say.

A real example from the Virginia work. We had a good line: even the industrial users split on this settlement. It is vivid, it is short, and it reframes the fight. It rested on one committee being absent from the settlement document. Our own gate blocked it, because absence from a document is not a position. The committee could have withdrawn earlier, been folded into another party, or declined to take a view. Three different findings, one observation, and that document cannot tell them apart. 2

The line is still blocked. It will stay blocked until somebody reads the procedural orders and resolves it, or it gets rewritten down to what the record actually supports. What it will not get is softened wording so it can slip through, which is the standard way this kind of gate quietly stops working.

You had better have as many agents challenging the work as you have creating it. That is true whether the agents are software or people, and it is the only reason what comes out the other end is grounded in anything.

What you are actually buying

Chief executives do not want a tour of the argument. In my experience they want one thing, and they will say it in the first ten minutes: do we have a team that is going to win this.

This is how you answer that honestly. Not by being cautious, and not by being slower. By making sure that when the other side goes looking for the weakest thing you have published, they find that somebody already went looking, wrote down what they found, and either fixed it or refused to say it.

You do not put brakes on a campaign because you are afraid of speed. You put brakes on it so you can take the corner faster than the other guy.