The Belt Equation

The story so far

Every change to the tree, with its date

Every change to the tree, the definitions, or the headline numbers, with absolute dates. Newest first.

This is the project's changelog, rendered straight from the repository. The project's day-by-day progress and its roadmap, kept in TimeAssembler, join this page at website step 24.

2026-09-19

  • Home page redesigned for readers who have never heard of the project (Cowork session, 2026-09-19): the headline number first under a year dial for the five longevity scenarios, then the chain of seven plain-English links (one per factor except W, which the dial covers, and C, beside the equation), each with the rate at which its gate nodes all hold, then who is asking and how to argue; the letter equation kept as one line under the chain. scripts/compute.py: simulate() takes a joint argument and reports how often a named set of nodes held together. scripts/export.py: numbers.chain and numbers.chain_nodes (the gate nodes the headline tier requires through the tiers under it, grouped by factor). Not yet deployed.
  • Domain attached: beltequation.com and www.beltequation.com connected to the Worker as custom domains from the Cloudflare dashboard, 2026-09-19, on John's call ahead of website step 25 (the review-status line beside the headline stays until step 27 lands). site/wrangler.jsonc carries the two routes so deploys keep them.
  • Roadmap step 13 done: scripts/visual.py draws the whole tree as one SVG (docs/visual/2026-09-19-tree.svg): three columns by horizon, rows grouped by factor, edges from dependency to dependent, each node with its 2071 / 2095 / no-deadline numbers, choice points dashed, resolved nodes filled, the computed headline and Tier 3 numbers in a box at the top, and the review-status line. Standard library only; regenerate after any change to data/. Website step 22 copies it onto the site.
  • Step 27, review 3 of 6 filed: motive-regime-contact packet, Gemini Pro Extended Thinking (docs/reviews/2026-09-19-motive-regime-contact-gemini-pro-extended-thinking.md). Raised: three missing nodes (mutually recognised keep-out zones, an orbital traffic authority, commercial insurance for off-Earth assets), four criterion fixes (exclude government customers from the orbital-work node, a fixed comparison for the recession node, a resident-status form for the permanent-residents node, an applied-result test for C4), three numbers down (kilowatt power beaming to 0.30 by 2071, C3 and C4 to 0.01). Gemini's pass is complete; the three Astra sessions remain.
  • Step 27, review 2 of 6 filed: launch-energy-drive packet, Gemini Pro Extended Thinking (docs/reviews/2026-09-19-launch-energy-drive-gemini-pro-extended-thinking.md). Raised: three missing nodes (cryogenic propellant storage in orbit, propellant made off Earth, megawatt heat rejection), three criterion fixes (fully robotic assembly, a fixed dollar baseline for the superconducting tape price, what counts as a stage recovered intact), three numbers up (lunar material at scale, the two-hundred-dollar launch price, routine nuclear propulsion). Not ruled until the packet's second review is in.
  • Step 27, review 1 of 6 filed: window-biology packet, Gemini Pro Extended Thinking (docs/reviews/2026-09-19-window-biology-gemini-pro-extended-thinking.md). Raised: three missing nodes (a regulator-accepted biological aging measure, surgery off Earth, a staple crop harvested off Earth), three criterion fixes (the dog trial tied to one project, the shielding reference, a spin-rate limit on the spin demo), three numbers (closed-loop life support down, the death-rate therapy up with no deadline, the dog trial down). Not ruled until the packet's second review is in.
  • Step 27 protocol ruled in force by John (2026-09-19, all six decisions accepted as proposed). Sessions begin with the window-biology packet.
  • Website steps 20 and 21 done (parallel Claude Code session, PR #13, merged 2026-09-19): Astro site in site/ with the five version-one pages, scripts/export.py writing the site's JSON from data/ and this changelog on every build, one page per factor and per node; preview live at beltequation.jfredson.workers.dev, domain unattached until step 25. The export runs the tree with the world draw at its default spread. A review-status line added beside the headline so the preview says the numbers are a first pass under outside review until step 27 lands.
  • Roadmap step 27 drafted, awaiting John's ruling: docs/reviews/protocol.md (six decisions with confidence and alternatives), docs/reviews/brief.md (the fixed reviewer brief), scripts/review_packets.py (generates the three packets from data/, standard library only), and the first packets under docs/reviews/packets/ dated 2026-09-19. Not in force until ruled.
  • Roadmap step 12 done: docs/methodology.md written. Covers the tree and the conditional per-scenario numbers, the play-out, the world draw and the case for a spread of 1.0 (with the independent-roll result printed beside it as the check), what is reported and why the five scenarios are not combined, the decision comparison, the calibration score (Brier, yearly and cumulative, over nodes that resolved after their first forecast), and what is fixed versus what moves.
  • Rulings on the first run (John, 2026-09-19, all four accepted as proposed): the two regime gates reworded to durable, nation-agnostic form with re-estimated numbers (lunar program across two changes of government 0.75 to 0.92, was 0.60 flat; space spending recovers within three years of a recession 0.80 to 0.90, was 0.55 flat; ids kept, revisions logged on the nodes); the child-born-off-Earth requirement moved from Tier 2 to Tier 4; a per-run world draw added to scripts/compute.py (normal, spread 1.0 on the log-odds scale, --world-spread, 0 restores independent rolls, both printed). Second run: headline A2 0.9 percent by 2071, 2.2 by 2080, 5.6 by 2095, 14 by 2136, 18 with no deadline; Tier 3 1.6 / 3.6 / 8.9 / 23 / 30 percent; A1 39 percent by 2071. Decision comparison separates the six-year fork by under a point in favour of separating, not a finding at sketch grade. Roadmap step 10 done, step 11 first pass done.
  • First probabilities, group three (motive, regime, access, Contact Clause): twenty-four open nodes estimated, five numbers each, proposed by Claude and accepted by John as proposed; the two aggregator nodes in the access branch carry 1.0 on the group-one precedent; the five personal access nodes are marked sketch-grade until pipeline graduation. Every one of the sixty nodes now has a first probability.
  • Roadmap step 10, the first run (docs/runs/2026-09-19-first-run.md): headline A2 at 0.0 percent by 2071, 0.1 percent by 2095, 2.2 percent with no deadline; Tier 3 at 0.2 percent by 2095; A1 at 35 percent by 2071. Read against intuition, three results fail: two regime nodes act as permanent one-shot gates, Tier 1's requirements are US-specific, and the child-born-off-Earth node chokes everything above Tier 2 under independence. Four structural fixes proposed for John's ruling; no probability changed. Step 11 ran but cannot separate the fork at these numbers and is re-run after the fixes.
  • First probabilities, group two (launch, energy, drive): sixteen open nodes estimated, five numbers each, proposed by Claude and accepted by John as proposed, on the same standing as group one (reviewed under the outside-model protocol, roadmap step 27, before publication). The fusion-plant construction node was already resolved and needs none. Two rationales cite the external forecaster baseline of 2026-09-19 (docs/forecasts/) as a comparison, not a prior. The most uncertain number in the group is the superconducting-tape price fall. Validator passes; 24 open nodes (motive, regime, access, Contact Clause) remain for group three.
  • External forecaster baseline filed: GPT-6 Astra, elicited 2026-09-19 without documents or lookup, seven milestones by 2070, 2080 and 2095 (docs/forecasts/2026-09-19-astra-baseline.md). A comparison row, not adopted priors; the cohort reading of the window item is not a W input and is to be re-asked for a specific person.
  • TimeAssembler roadmap brought up to the full 27 steps of docs/roadmap.md (it held 14), plus the Astra re-ask; website steps 20 and 21 started in a parallel Claude Code session.

2026-09-08

  • First probabilities, group one (window and biology): fifteen nodes estimated, five numbers each, proposed by Claude and accepted by John as proposed, on the record that he defers to the recommendations for now for want of field experience and intends to review them under an outside-model protocol. That protocol is promoted from "Later, not now" to roadmap step 27, to run before anything is published. The partial-reprogramming node found already resolved (first participant dosed 2026-06-09). Three nodes resolved in all.
  • Roadmap step 8 done, the review pass: all sixty node names passed the plain-language test; nine criteria reworded to give the plain meaning before the technical term. No node cut or merged. The remaining verification flags cleared: the fusion drive's weight gap checked against agency propulsion reviews and reworded; the 2015 regulator agreement confirmed as an endpoint decision only; the second-longest verified life confirmed at just over three years short of the record; the metformin trial's results found unpublished in any primary source (node stays open). Phase 2 complete: the tree is stable at 60 nodes and probabilities may begin.
  • Three decisions agreed by John after discussion: a fifth longevity scenario, radical extension to 2136 (a 150-year life), between strong extension and the open window; a regime node for an off-Earth site accepting permanent residents, required by A3 so it no longer equals A2; and the fitness percentile held out of the health paragraph until it comes from a measured test against a named standard. The beyond-the-Moon site node's note clarified as inclusive of every kind of habitation. The August upper-stage catch checked: not attempted, deferred; node stays open. Tree at 60 nodes.
  • Roadmap step 7 done: dependency edges drawn and every tier given its requirement list (data/tiers.toml). Two hard dependencies loosened (fusion plant on superconducting wire; orbital assembly on launch price). Access branch restructured for the decision comparison: two route nodes (civilian training role; military role with off-Earth rotation), a node each for the A0 and A1 tiers, the degree node cut. A Tier 3 node added for a crewed site beyond the Earth-Moon system. Space solar sold to a grid cut. Two verified leaves added: construction begun on a fusion plant meant to sell electricity (resolved 2025-07-30) and both stages of a heavy launcher recovered from one orbital flight (open; late-August 2026 attempt to be checked). Tree at 59 nodes, a third leaves, two resolved, validates with no cycles. A3 has no distinguishing node yet and equals A2.
  • First verification pass under the new rule: the 2022 private flight, the 122-year record, the dog rapamycin trial, the Chinese 370-day sealed habitat, the station's power, and the mouse centrifuge confirmed; the lunar surface reactor's schedule found slipped to the early 2030s, node moved from leaf to mid; the regulator-accepts-aging node's criterion tightened to an approvable indication after confirming the 2015 composite-endpoint precedent; four claims left marked in verify fields for a later session.
  • Ground rules added to docs/roadmap.md, ruled by John: memory claims are checked against the web as work proceeds or marked unverified on the record; and a one-paragraph general health statement may appear in the project (definitions.md, "The window"), amending the earlier rule of nothing at all.
  • Third brainstorm session draft (roadmap step 4, motive, regime, access, Contact Clause): five motive nodes, four regime nodes, eleven access nodes (six world events including the tree's first resolved node, a private person over 65 flying to orbit in 2022; the two choice points at the six-year fork with re-enlist marked as the current plan; three leverage-branch nodes), and the four Contact Clause rungs with C3 and C4 as long shots. Tree at 54 nodes, a third leaves, validates. Accepted by John as drafted; the brainstorm (roadmap step 4) is complete, and step 6 (choice points marked with their options) was done inside it.
  • Second brainstorm session draft (roadmap step 4, launch, energy, drive): five launch nodes added beside the first, four energy nodes, four drive nodes, one long shot (the constant-acceleration drive). Tree at 30 nodes, a third leaves, validates. Accepted by John as drafted.
  • Website decisions accepted by John, all nine. Domain registered: beltequation.com.
  • Standalone site no longer deferred (John's ruling, reversing the founding posture on that one point; under-promoted unchanged). docs/website-plan.md written: five pages for version one, an Astro site in site/ on Cloudflare Workers static assets like the sister sites, a Python export script feeding it JSON, the tree as a committed picture, the story pulled from TimeAssembler at build time, critique through GitHub issues. Nine decisions with confidence and alternatives, awaiting John. Roadmap gains Phase 6, steps 20 to 26, alongside the existing plan.
  • First brainstorm session (roadmap step 4, window and biology): seven window nodes and nine biology nodes added to data/, criterion first for each, two long shots with mechanism and breaking point, six leaves. The tree has 17 nodes and validates. No probabilities. Definitions amended: the deadline is the scenario axis, not a node of its own. Idea recorded under "Later, not now": a review method that brings in other models for fresh critique.
  • Roadmap step 5 done: scripts/compute.py written against the schema. Standard library only. Commands: validate (all eight schema rules), shape (counts, leaves, tiers), run (the tree played out per scenario, tiers and Contact Clause reported), compare (one choice group at a time against the current plan). Refuses to report numbers until every open world node has probabilities. Checked on the real data and on a small test tree whose results match hand arithmetic.
  • Second founding rule added to docs/definitions.md, stated by John: plausible all the way to the edge of the impossible. A long-shot node must name its mechanism and state plainly where current physics or engineering would have to give way. Carried into docs/node-schema.md as a review test: a long-shot node without both fails the same way a node without a resolution criterion does.
  • data/scenarios.toml baseline moved to the refined window, 2071 plus or minus 8, no longer provisional.
  • Ground rule added to docs/roadmap.md, ruled by John: pull requests are merged with the gh command-line tool, not the GitHub website.
  • Roadmap step 2 decided: John accepted the node schema with two amendments Claude made on reflection after stating confidence and alternatives for each choice. Choice points now default to a named current-plan option instead of "not taken"; "either route" logic moves from the tier file onto nodes (depends_on_any), so the habitat hedge becomes one named biology node; the compute script will refuse probabilities that fall as the window lengthens. Known simplifications recorded in docs/node-schema.md. tiers.toml loses requires_any.
  • New ground rule in docs/roadmap.md, ruled by John: every design choice put to him carries a confidence level, whether it is standard practice, and the strongest alternative, before he is asked to accept it.
  • Roadmap step 2 drafted, awaiting John's acceptance: docs/node-schema.md (the node record; TOML, one file per factor; tiers as required-node sets; choice points forced, not sampled; probabilities per scenario and conditional on dependencies), data/scenarios.toml, data/tiers.toml with empty node lists, and the first real node in data/L-launch.toml (a fully reusable heavy launcher flying 100 orbital missions in a year).
  • Window refined from the health record to 2071 plus or minus 8 (baseline scenario moved from 2070 to 2071). A leverage branch (AI capability as a multiplier on John's own participation) added under the access factor. A private, git-ignored transcript of the scene that inspires the project added under docs/private/.
  • Roadmap step 1 done: the kickoff outline reconciled against definitions.md. The outline now carries a note listing what the definitions superseded (three branches to eight factors, location-agnostic tiers, four scenarios, per-scenario re-runs, the node schema) and a starting map of its branch seeds onto the eight factors. The public-forecasting reference from the outline is carried into roadmap step 9. README now points to the roadmap as the plan. Definitions unchanged.
  • Roadmap reviewed with Claude, all six changes approved by John: the node brainstorm runs as three grouped sessions; the compute script moves into Phase 2 in parallel with the brainstorm (now step 5); the window refinement (step 3) is marked non-blocking and sits after the brainstorm in the TimeAssembler task order; the habitat branch is folded into the brainstorm sessions instead of standing alone; a decision-comparison run at the six-year service fork is added as step 11; the first essay will cross-reference the Calibration Problem manuscript. No definitions changed.
  • Project created and named The Belt Equation.
  • Kickoff outline written (docs/2026-09-08-kickoff.md).
  • Definitions document started (docs/definitions.md) and, the same evening, every founding decision made: the question wording, the access tiers A0 to A3 with A2 as the headline, the system tiers 1 to 4 with Tier 3 as the "Belt exists" bar and habitats included, the eight factors frozen, the window placeholder at 2070 with four longevity scenarios including escape velocity, the founding rule that long shots stay on the board, the Contact Clause beside the equation, the access branch kept provisional until pipeline graduation with choice-point nodes, and the public-but-under-promoted posture with a quarterly scan and January annual review.
  • Roadmap written (docs/roadmap.md) for the work before the pipeline resumes.
  • No nodes, no probabilities yet.