- When every team's diagnosis is locally correct but the company still can't fix the problem, the cause almost always sits between the teams, not inside any one of them.
- Debate doesn't resolve this, because each side isn't just seeing different facts — they're weighing different things as important. You have to change the angle, not win the argument.
- Naming what each person is actually protecting — correctness, their team, the budget, the board — turns an argument about facts into a visible trade, which is the only way to find a fix that satisfies more than one seat at once.
- The fix that looks obvious from one seat usually creates a second problem two steps downstream. Tracing those ripple effects before you commit is what separates a good-sounding fix from a fix that holds.
- Parallax isn't a personality trait or a facilitation trick. It's a discipline — a repeatable way of triangulating position, so the real constraint stops hiding in the gap between everyone's honest account of what they see.
Parallax, in astronomy, is the apparent shift in a star's position when you observe it from two different points instead of one. The star hasn't moved. You have — and that shift is what lets you calculate a distance no single observation could ever reveal.
Most stalled decisions have the same shape. Not "nobody understands what's wrong" — the opposite. Everybody understands what's wrong, from where they're standing. The trouble starts when a company mistakes four accurate local views for one company-wide diagnosis, and starts arguing about whose local view should win instead of asking why they don't match.
Everyone is right about what they see
Picture a company whose AI pilot has stalled. Sit down separately with four people and ask each one what's wrong, and you'll get four confident, well-evidenced, completely different answers.
The engineer will show you the data pipeline: half the fields the model needs are inconsistent, missing, or stored in three different formats depending on which team entered them. From where they sit, this is an infrastructure problem, and it's real — you cannot build reliable output on top of unreliable input.
The frontline manager will show you their team quietly routing around the tool: it makes confident recommendations that are sometimes badly wrong, and once you've been burned twice, you stop trusting it and go back to doing it by hand. From where they sit, this is a trust problem, and it's real — a tool nobody believes is a tool nobody uses, no matter how good the underlying model is.
The finance lead will show you the invoice against the usage report: a meaningful license spend, and a system that three people log into. From where they sit, this is a return-on-investment problem, and it's real — you cannot keep funding a tool nobody uses at the price you're paying for it.
The CEO will show you the board deck: a bet that was supposed to be finished by now, that they still can't point to a number for. From where they sit, this is a credibility problem, and it's real — every quarter this drags on costs them trust they'll need for the next bet.
Four diagnoses. Four different fixes on offer — better data infrastructure, better change management, cancel the license, ship something demoable fast. All four people did their homework. None of them is lying, or careless, or missing the obvious. They're each holding one true observation of a system none of them can see all at once. That's not a failure of intelligence. It's a failure of vantage point — and vantage-point problems don't get solved by whoever argues hardest.
Why the debate never actually ends
The instinctive next move is to get everyone in a room and debate it out. This rarely works, and it's worth understanding why, because the reason is more specific than "people are stubborn."
Each person in that room isn't only working from different facts. They're unconsciously running a different scoring system for what even counts as a good outcome. The engineer is optimizing for correctness — the system should be right, and "right" is close to a moral position for someone who has spent a career being precise for a living. The frontline manager is optimizing for care — protecting their team from a tool that embarrassed them in front of a customer once already. The finance lead is optimizing for fairness in the accounting sense — spend should be proportional to value received, full stop. The CEO is optimizing for the loyalty and standing of the bet they made to the board that funded it.
None of those value systems is wrong. They're just different currencies. Trying to resolve a coordination problem through argument is like four people negotiating a price in four different currencies and refusing to agree on an exchange rate — you can talk for hours and nobody moves, because everyone is right inside their own system and the systems don't automatically translate.
This is the part most facilitation misses. It treats the disagreement as an information gap — as if showing the finance lead the engineer's data pipeline diagram will make them agree on the fix. Sometimes it helps. Usually it doesn't, because the finance lead was never disputing the pipeline. They were weighing a completely different question, using a completely different scale. You don't resolve that by presenting better evidence. You resolve it by explicitly naming which value each seat is protecting, so the room can see that four "wrong" people are actually four "right" people optimizing for four different, legitimate things — and then ask a harder, shared question: given all four are true, what's the one change that moves all four numbers at once?
Naming the currency, then asking one harder question
Shifting vantage point on demand isn't a gift some people have and others don't. It's a sequence you can run in the same meeting that's currently going in circles.
Step one: say the value each person is actually protecting, out loud, before debating solutions. "You're optimizing for the system being correct. You're optimizing for your team not getting burned again. You're optimizing for spend matching usage. You're optimizing for what you told the board." This isn't therapy — it's making the exchange rate visible, so the room stops mistaking a values disagreement for a facts disagreement.
Step two: ask the one question none of the four individual fixes answers — given all four diagnoses are true at once, what's the single change that moves all four numbers, not just the one you personally own? For the stalled AI pilot, that's rarely "more training" or "cancel the tool." It's almost always the data pipeline the engineer flagged first: fix what the model is actually reading, and accuracy improves (helping the engineer), trust rebuilds because the tool stops being visibly wrong (helping the manager), usage climbs enough to justify the spend (helping finance), and there's a real number for the board (helping the CEO). One fix, four value systems satisfied, because it was never four problems — it was one problem observed from four seats.
That's the test for whether you've found the real constraint: it should make more than one person's number move. If your fix only helps the person who proposed it, you haven't triangulated yet — you've just picked a side.
The fix that creates the next problem
There's a second discipline that matters just as much as changing where you stand: following your own fix far enough downstream to see what it breaks.
Take the AI pilot again. The obvious first-order fix for "nobody trusts the tool" is to make it more visible — put it in front of more people, more often, so adoption climbs. That's the fix that looks right from the frontline manager's chair. Follow it one step further, though, and you hit the second-order effect: pushing an unreliable tool in front of more people faster doesn't build trust, it destroys more of it, faster, in public. Follow it a second step and you reach the third-order effect: the team that got burned twice now tells the team next to them not to bother, and the credibility problem that started with one pilot has just spread sideways to two more, before anyone touched the actual data pipeline that caused the unreliability in the first place.
The fix wasn't wrong on its face. It was incomplete, because it solved for the visible symptom one level down and ignored the compounding effect two and three levels down. This is where most well-intentioned interventions quietly fail — not because the first move was bad, but because nobody asked "and then what happens, and then what happens after that" before committing to it.
Tracing those ripples is slower than picking the fix that feels obviously right in the room. It's also the difference between a recommendation that holds up in six months and one that just relocates the same problem to a different team's desk.
Parallax as a discipline, not a personality
Put these together — deliberately occupying more than one vantage point, naming the different value systems in the room instead of arguing past them, and tracing a fix through its second and third downstream effects before committing to it — and you have the actual method behind the name. Not a philosophy. A repeatable sequence you can run on any stalled decision, whether the subject is an AI pilot, a reorg, or a pricing change nobody can agree on.
It's also why we never walk into an engagement with four people's opinions and pick the loudest one, and we never walk in with a fix before we've triangulated where the four accounts actually converge. The convergence point — the place where the engineer's data problem, the manager's trust problem, the finance lead's return problem, and the CEO's credibility problem all trace back to the same root — is very rarely where any single person in the room was pointing. It's usually one step removed from all four. Find that point, fix what's actually there, and every one of those four honest, locally-correct diagnoses gets resolved at the same time, by the same change.
That's the whole principle. The star didn't move. You did. And from two positions instead of one, the distance you couldn't measure before finally has a number on it.
Running this sequence internally still takes someone outside all four seats to hold the room to it. A diagnostic does that job: it triangulates where your four honest, locally-correct accounts actually converge, and hands you the one fix that moves all four numbers — not the loudest opinion in the room.