How a piece is built
Long-form research is usually a wall of prose with charts bolted on. Ours is laid out so the reader always knows what is being claimed, how sure we are, and what the next piece of evidence would do to the claim.
The anatomy of a piece
Each part has a fixed place on the page, so a reader always knows where to look.
The claim, stated once
The thesis sits at the top in one paragraph, with its confidence split into mechanism and timing and a dated horizon. The library shows the same line, so a reader can compare theses without opening them.
Exhibits that follow the argument
Charts don’t interrupt the text. In a scrollytelling piece they sit beside it and change as you read; in an editorial piece they break out to full width where the argument needs them. Either way, when a paragraph says the IT bill dominates, the IT bill lights up.
Every model is written out and can be driven by the reader. Change the launch price or the power tariff and the piece’s own figures recompute. The derivations live once, in Mechanisms.
Dated falsifiers
What would change our mind, and by when. A failed test changes the scenario rather than prompting a search for a more flattering comparison.
Metrics to monitor
The readings that would move the odds, with the threshold that favours each side and the first observation required. These feed the watch strip on the front page.
Sources and as-of notes
Primary sources with the date they were checked. Where a figure is a model input rather than a quote, it says so where the number appears.
Four rules, every piece
No trade recommendations
We publish theses, odds and the evidence that would move them. What you do with that is your business.
Timelines are compared, not asserted
Every forward claim is laid over the historical episodes that resemble it, so the reader can see where the analogy holds and where it breaks.
Data is verified and dated
Figures trace to primary sources with the date they were checked. Model inputs are labelled as inputs, never dressed up as quotes.
Formulas and methods are written out
If a number came from a model, the model is on the page. Readers can rerun it, disagree with an input, and see what changes.
Reading the exhibits
Every chart carries one of three classifications and the date its data was checked. The classification tells you how far to trust the line before you trust the argument.
Observed
A published series reproduced as published. Units and frequency are the source’s. Rounded figures are shown rounded and never rescaled to sum.
Constructed
Built from published inputs by arithmetic we show: a ratio, a cumulative sum, a rolling window. The construction is in the caption and the inputs are in the data note.
Proxy
Stands in for the thing we would rather measure. A tracker fund for an index, a size decile for the S&P 500. The caption says what it is not, and nothing is spliced across constructions.
How the odds move
Pieces that carry scenario weights follow one review rule, written here once so every piece doesn’t have to repeat it. Every stated weight and dated test is kept on the record.
1
Review quarterly. Non-event cost and service signals need two consecutive reviews before they move a weight; a final license order or a missed dated gate can count immediately.
2
Require independent confirmation of critical invoices and readings. One source is a lead, not evidence.
3
Use one coherent evidence package per review. Don’t reuse the same evidence twice, and don’t stack the illustrative updates.
4
Transfer weight by judgment, not by formula. The transfers are published as examples; none is automatically today’s revised probability.
5
Mixed signals leave the weights unchanged. Investigate the conflict before moving the odds.
6
When a dated gate fails, the scenario changes. Publish the missing cost or the disappointing result alongside the successful launch.
Example: the orbital compute piece, starting weights
15%
35%
40%
10%
Ground catches up
Rollout stalls
Orbit finds a niche
Broader parity
Each piece that carries weights shows its illustrative single updates in the same bar, so a reader can see what one piece of evidence would do before it arrives.