Why we test an explanation we don't believe

Most stories this heavily covered settle quickly on the simplest explanation, and the simplest explanation is usually right. Occam's razor, the rule that the explanation needing the fewest assumptions is probably the best one, exists for good reasons.

This journal breaks it on purpose, in one clearly labelled place. Our main question is about outcomes: what is being built in Canada's resource trade, where it goes, and who pays. Alongside it we run a thought experiment: suppose Washington foresaw, and wanted, the way Canada has responded. What would we expect to see, and do we see it?

The thought experiment runs from a loose version to a strong one. The strong version, a worked-out plan with named projects and a timeline, the author doesn't believe. The looser one, that Washington saw Canada's response coming and welcomed it, is the one the author argues for here, deliberately. Its main rival is plain self-interest, in which the US acted in its own interest without much thought for how Canada would respond.

Arguing the looser version is a steelman, not a conviction. To test an explanation properly, you have to make its best case and hold it as if you believed it, and the author doesn't consider it the most pragmatic reading.

The best case has real arguments behind it. It is hard to picture a room of US trade analysts where nobody raised how Canada would respond, or where more secure Canadian supply would be unwelcome. And we have found no US measure aimed at slowing what Canada is building. Holding that case is also what drives the research: without it, there would be little reason to dig. But the case itself isn't what we write about. The articles are about what the digging turns up, and those findings hold whoever intended what.

Both foresight versions need more assumptions than the plain reading, and Occam's razor would pick the plain one by default. But most of what we grade fits both readings equally, and an event that fits both equally leaves the question where it was: it isn't a point for the simplest story. Only the events that tell the readings apart count, and so far they lean clearly the plain way. So why test the foresight versions at all?

The razor doesn't choose where to look

Applied too early, the razor stops the search before it starts. If the simple story is assumed, nobody checks what a less likely one would predict. An unlikely explanation is useful precisely because it disagrees with the plain one. It forces predictions the plain story doesn't make, and it sends the research into places a heavily covered story rarely gets inspected.

It also widens what we watch. We follow the story as it unfolds, a step behind the headlines. Each explanation we keep in play adds items to the watch list. Is a tariff eased soon after a Canadian law passes? Who pays for the lawsuits against the projects? Which way does the money flow? Watching them together connects developments that usually get reported separately.

Then the evidence decides. Events are graded against each explanation: does this fit one reading better, or both equally? Most of the time the honest answer is "both equally", and we say so. Here is the count so far for the plain self-interest reading against the foresight reading:

Events graded against both readings	Count
Fit both equally	140
Favour plain self-interest	65
Favour foresight	7
Total	212

An event that fits both equally is called non-discriminating: however striking it looks, it can't tell the readings apart. The Method page explains the grading. The grades are made by an AI model working blind: it sees each event, its sources and the readings' definitions, but not earlier grades, our arguments or which reading anyone favours. The author reads them, but they stay the machine's judgment.

The one exception is a public warning by Atlantic Council analysts in January 2025, before any tariff took effect, that energy tariffs could push Canada and other suppliers to diversify away from the US. It shows the response was foreseeable, so "nobody considered it" is a little harder to hold. It doesn't show that Washington wanted it: the analysts called it a loss for the US.

What the searches found

The payoff is in the by-products: findings that stand whatever anyone intended.

Nobody forecast both halves. Testing "was it foreseen?" meant reading what US officials, think tanks and lawmakers said before the tariffs. They mostly said Canada couldn't build: pipelines too hard, a recession coming. We found no one who predicted that Canada would build fast and keep supplying the US.

Pre-registered prediction P14 · scored · registered 2026-09-27"Was it foreseen?" (foresight test 1)

"Was it foreseen?" (foresight test 1)

Window: 2025-06-30. What counts: What counts. Documentary evidence dated before or early 2025 from US officials, advisers, think tanks close to the administration, or trade/Commerce staff, predicting that tariff pressure would push Canada to permit, build and export resources. What each result means: - Found and tied to administration circles → the foresight reading rises (the "foreseen and welcomed" reading may rise with it, since foresight alone …

Result: Nothing found tied to administration circles, in the foresight search and its follow-up. Outside analysts did warn that pressure could push Canada to diversify, but none of them close to the administration, and an insider asked directly (Energy Secretary Chris Wright, February 2025) dismissed the idea. Under this test that is weak evidence for plain self-interest: it counts against the foresight readings, though only a little, because private foresight would not show up in public sources.

This documented search, registered and scored on 27 September, found nothing tied to the administration, and it turned the build-out into a result to explain rather than a truism. See An awkward response.

No hidden money behind the lawsuits. One rival explanation says Washington would rather weaken Canada than see it build, and First Nations litigation is the main legal brake on the build-out. So we looked for money linked to the US government or the US right behind the cases, in British Columbia, Ontario, Alberta, Quebec, Labrador and at the federal level. We found none.¹ The visible money comes from the nations' own funds, Canadian crowdfunds and lawyers, Canadian governments, and US conservation foundations that were giving well before 2025.

The only new US money we saw worked the other way. After signing benefit agreements with Houston-based Western LNG, two First Nations withdrew their court challenges to an export terminal.

Event · 2026-06-09 · action · CA-BCMetlakatla and Lax Kw'alaams withdraw Federal Court challenges to Ksi Lisims LNG after benefit agreements with the project, led by Houston-based Western LNG

Metlakatla and Lax Kw'alaams withdraw Federal Court challenges to Ksi Lisims LNG after benefit agreements with the project, led by Houston-based Western LNG

On 9 June 2026, Metlakatla First Nation, the Lax Kw'alaams Band and Gitxaala Nation signed benefit agreements with Ksi Lisims LNG, whose lead developer is Houston-based Western LNG, along with agreements with the PRGT pipeline. The release announcing the Lax Kw'alaams agreement, distributed by Western LNG, says the band withdrew its Federal Court judicial review (file T-4013-25) as part of signing. Metlakatla's withdrawal of its challenge was confirmed by a Western LNG spokeswoman to The Canadian Press. No amounts were disclosed. The project has preliminary supply deals with German utilities.

Documented by

• Western LNG, “Metlakatla First Nation signs Benefits Agreement with Ksi Lisims LNG and amended Project Agreement with PRGT”, 2026-06-09.
• Western LNG, “Lax Kw'alaams Band signs Benefits Agreement with Ksi Lisims LNG and modernized Project Agreement with PRGT”, 2026-06-09.
• Western LNG, “Gitxaała Nation signs Benefits Agreement with Ksi Lisims LNG and Project Agreement with PRGT”, 2026-06-09.

"Not found" means not observed in public sources: covert money is invisible by design. So the question stays open, as a registered test.

Pre-registered prediction P10 · open · registered 2026-09-27Fragmentation watch

Fragmentation watch

Window: 2026-09-27. What counts: The question: is there confirmed evidence that the US is trying to break Canada apart (the fragmentation reading) rather than wanting it to build (the foresight reading)? What counts as a hit: any confirmed item of these kinds counts for fragmentation over the foresight reading: US tariff offers or rates that differ by province (including who approached Quebec in spring 2026, and on whose mandate); US contacts with …

A strong rally, but no choreography. US insults and annexation jabs clearly rallied Canadians. In a Léger poll at the end of August 2026, 56% said they were very proud to be Canadian, up 11 points from June 2025. Of those feeling prouder, 72% said it was because of things President Trump had said. Only 9% wanted Canada to become a US state.²

But when we dated 53 provocations against Canada's steps to lock in the build-out, the timing didn't fit a script. About 20 came within days of a Canadian act of defiance: counter-tariffs, Ontario's Reagan ad, Prime Minister Carney walking away from the talks. Near Canada's 14 lock-in steps, from British Columbia's fast-track law to the tabling of the federal bill C-39, provocations were rarer than chance, and US tariff actions were at chance level.³⁴

The test has little power, and the steps were picked after the fact, so we read it narrowly: no sign of a scripted campaign. A sharper test, on tariff actions around Canada's new laws, is registered and still open.

Pre-registered prediction P6 · open · registered 2026-09-27Tariff-action test around the new laws

Tariff-action test around the new laws

Window: A = the later of two dates: the Royal Assent of Bill C-39 and the Royal Assent of the bill containing the Productivity Mega Deduction; windows A-120 to A+120 days. What counts: The question: do new US trade restrictions on Canada drop once Canada's two new laws are in place? We compare the 120 days before the date the second of the two laws receives Royal Assent with the 120 days after it. What counts as a hit: strictly fewer new US restrictive actions aimed at Canada in the 120 days after that date than in the 120 days before it (a tie counts as no drop), and at least one easing act in …

The order you heard it in. Testing what was foreseen means dating everything twice: when it was said, and when the public heard it. Applied to the "extreme tail risk" story in September, that showed almost every American dismissal of a military threat to Canada came before Carney spoke, the reverse of how most people received it. That is The order you heard it in.

The rules we hold ourselves to

Outcomes first. The main claims are about outcomes, and they stand whether or not anyone intended them. Every article leads with what the searches found.

Never alone. The thought experiment is always labelled as one and shown beside its rivals: accident, plain self-interest, annexation and fragmentation.

Tests before verdicts. Predictions are registered before the result is known, with what would count against each reading, and scored in public. Misses are expected, especially on the speculative predictions. A miss is information, not a failure, and it stays on the record.

Careful words. "Consistent with", "fits both equally", "not observed in public sources". We never claim to know what anyone intended.

Actions over statements. What costs money or carries legal force counts for far more than words.

The odds, and the markets

The odds on this site are our estimates, not expert forecasts. Where we can, they come from simple models whose rules we write down before running them; the method page says which ones. They are published so you can see where the journal leans, and they move only with a written reason. The readings on why Washington acted are the exception: they carry no odds at all, because no dated event could ever settle them. They decide where to dig; what we publish is what the digging turns up. More in About the odds.

We also watch prediction markets. We don't bet on these markets, and thin markets are weak signals. We use them only to spot gaps: when the crowd and our odds disagree widely, we go and look for what we might be missing. A market price never moves our odds by itself. More in How we use prediction markets.

The method, in short

For anyone who wants to try this on another story: keep the outcome question separate from the unlikely explanation. List at least three rival explanations. Write down, in advance, the predictions on which they disagree. Search where those predictions point, and save the primary documents. Grade every item against every reading, and count the ties. Let the razor decide, using only the items that tell the readings apart. Then publish what the searches found.

That last step is what this site is for. The verdict on the unlikely explanation matters less than what we found on the way to it.