Record
Every probability we have stated and every dated test we have set, with the date it was made and what happened. Nothing here is deleted. When a weight moves, the old weight stays on the page as a revision.
Calibration
Each stated probability becomes a point once it resolves: its probability on the horizontal axis, and the share of claims at that probability that came true on the vertical. A well-calibrated record sits on the diagonal. The Brier score summarizes the same thing as one number; lower is better, and 0.25 is what coin-flipping earns.
Brier score over N resolved claims, p the stated probability, o the outcome (1 or 0)
BS = (1/N) Σ (pᵢ − oᵢ)²
Rules of the ledger
1
A probability enters the ledger the day it is published, with the piece and the resolution rule beside it. It can be revised at a quarterly review, and every revision stays visible.
2
A dated test enters with its due date and what each outcome would favour. It resolves as passed, failed or not observable; a test that could not be observed is recorded as such, not quietly dropped.
3
Resolution uses the sources named in the piece. Where a definition is still open, the test shows as definitions pending and does not count toward the score until it is fixed.
4
Non-event cost and service signals need two consecutive reviews to move a weight; a license order or a missed dated gate can count at once. Independent confirmation is required for load-bearing invoices.
5
Scores are computed from resolved claims only. We do not grade our own confidence words; we record them so readers can.
6
When a thesis fails, the autopsy is published alongside the record, not instead of it.
Stated probabilities
Scenario weights with their revision history. Each column is a quarterly review; filled columns are published weights, empty ones are reviews still to come.
Technology
Resolves when a terminal state is confirmed at a quarterly review, or at the end of the 2030–2040 horizon. The gates below are the evidence that moves these weights.
Published
Growth or Grind
Economics
No weights stated. The piece proposes an end-2030 classification into four terminal branches without weights; weights enter the ledger only if a future review assigns them.
Terminal branches, in order of precedence: an enacted, attributable market-pressure fiscal turn; otherwise documented financing accommodation; otherwise growth with retained fiscal capture; otherwise residual grind. Definitions and the attribution protocol are still to be fixed, so this is recorded as logic, not as a scored claim.
Dated tests
The falsifiers each piece set for itself, with the date by which they must be observed and which side a result would favour.
| Due | Piece | Test | A pass would favour | Status |
|---|---|---|---|---|
| Due by end of 2027 | ||||
| Dec 31, 2027 | What a free rocket wouldn't fix | Actual upper-stage reflight. Catching a vehicle alone does not establish reuse. | Orbit | Open |
| Dec 31, 2027 | What a free rocket wouldn't fix | Louisiana construction, financing and project-specific review record.No Louisiana application reported as of Oct 28, 2026, single source. | Orbit | Open |
| Dec 31, 2027 | What a free rocket wouldn't fix | A delivered launch offer with payload, orbit and recovery terms.Posted $3,364/kg reference is not an offer for a compute mission. | Orbit | Open |
| Dec 31, 2027 | What a free rocket wouldn't fix | A reproducible satellite bill of materials with fully annualized IT replacement costs. | Orbit | Open |
| Dec 31, 2027 | What a free rocket wouldn't fix | Ground projects that actually deliver power, with complete costs and usable-service dates. | Ground | Open |
| Dec 31, 2027 | Growth or Grind | The end-2027 core-CPI test the timing claim is held to.Threshold as stated in the piece; recorded here by reference. | Timing claim | Open |
| Due by end of 2029 | ||||
| Dec 31, 2029 | What a free rocket wouldn't fix | A full year of thermal telemetry with failures reported separately; at least 100 GPU-years and one flight-year.Silent corruption and permanent loss reported as separate lines. | Orbit | Open |
| Due by end of 2030 | ||||
| Dec 31, 2030 | What a free rocket wouldn't fix | Paid launch invoices, compute-only manufacturing mass and observed hardware costs; workloads assessed on useful output and availability. | Orbit | Open |
| Dec 31, 2030 | What a free rocket wouldn't fix | Launch cadence checkpoint: 1,000 or more completed flights a year favours orbit; fewer than 300 weakens the large-ramp branch.Current reading: 25 authorized per year, Texas. Completions not yet observable. | Orbit at ≥1,000; Ground below 300 | Open |
| Dec 31, 2030 | Growth or Grind | Branch classification into one of four terminal states, by precedence.Definitions and attribution protocol not yet fixed. | Unweighted | Definitions pending |
Stated confidence
Recorded as written. These are not scored; they are here so a reader can compare what we said with what happened.
Resolved
Nothing has resolved yet.
The first dated tests are due December 31, 2027. This section will show each one with the evidence used to resolve it.