Escape velocity is the speed at which a rocket stops falling back. A technology has one too: the point where it works well enough, cheaply enough and reliably enough to carry on by itself. Escape Velocity is an open atlas of what each technology still needs to get there: where it stands today, the target that would make it useful, the physical limit behind that target, the gaps in between, the other technologies it waits on, and the research that could close each gap.
Every number has a source. Every technology is a node in one graph, so a bottleneck shared by quantum computing, fusion and medicine shows up as one thing, not three.
Explore it at scape-velocity.github.io/escape-velocity: the distance to target of every technology, the dependency graph, and filters over the gaps and the evidence. On GitHub, start with STATUS.md: every domain, headline metric, critical gap and the technologies the rest of the atlas depends on most.
graph LR
tech["Technology<br/><i>statement, readiness</i>"] --> metric["Metrics<br/><i>current, target, limit</i>"]
tech --> gap["Gaps<br/><i>type, layer, severity</i>"]
tech --> dep["Requires<br/><i>other technologies</i>"]
metric --> card["Evidence cards<br/><i>source, quote, class</i>"]
gap --> card
dep -.-> tech
- A technology is a capability with a measurable target: a fault-tolerant quantum computer, a blood test that finds cancer at stage I, direct air capture at 100 USD per tonne.
- Three numbers per metric: the best value demonstrated so far, the target with its rationale, and the physical limit where one exists. The distance is shown in orders of magnitude.
- Gaps are classified by type (scientific unknown, engineering, fundamental limit, data, manufacturing, cost, regulation, supply chain), layer and severity.
- Dependencies are technologies too, each tracked with its own metrics and gaps. A gap held open by a dependency points to it.
- Evidence cards carry every number: the source, a verbatim quote, the conditions and a class (established, reported, extrapolation, speculation). A script checks each card against its source.
The full model is in docs/model.md.
| Computing | Quantum | Artificial intelligence |
| Health | Biotechnology | Neurotechnology |
| Energy | Climate | Materials |
| Water | Food and agriculture | Space |
| Enablers |
Enablers are the shared infrastructure many technologies wait on: cryogenics, heat removal, power electronics, photonics, lasers.
The atlas is kept current by people and agents following the same procedures. The skills in skills/ work with Claude Code and other agents that read the Agent Skills format:
| Skill | What it does |
|---|---|
| scout-literature | Searches OpenAlex, arXiv, Crossref, PubMed, ClinicalTrials.gov and Semantic Scholar for work that could close a gap |
| review-evidence | Checks each evidence card against its source: identifier, title, year, author, quote |
| define-technology | Adds a technology with its statement, metric and target |
| decompose-technology | Maps what a technology depends on and what stands in its way |
| assess-readiness | Sets a readiness level on TRL, MRL or clinical phases, with evidence |
| find-neglected-gaps | Finds bottlenecks many technologies wait on and few people research |
| cross-domain-transfer | Looks for approaches from one field that could close a gap in another |
| weekly-brief | Summarizes what is new for the mapped technologies |
| translate-page | Translates a technology, a card or the taxonomy into a registered language, for a person to review |
An MCP server gives any agent read access to the atlas and the literature search:
list_technologies, technology, gaps, dependencies, bottlenecks, evidence,
search_literature, abstract and count_literature. Claude Code loads it from
.mcp.json when you open the repository; other clients run
python3 tools/mcp_server.py from the repository root.
The whole atlas is also published as one file, atlas.json (format), and for language models as llms.txt.
Every change, from a person or an agent, is a pull request reviewed by a person: a curator of the
technology or a moderator of its domain, checked by the moderation check. Agents write evidence as
unverified; only curators and moderators verify it (GOVERNANCE.md).
The Alan Machine is an open-source book about a hypothetical supercomputer at the physical limits of computation. Its Building Alan dossiers track real hardware against those limits. Escape Velocity shares its metrics and its kinds of claim, links each technology to the pages that discuss it, and maps quantum computing and AI first because both projects need them (decision 0008).
| Stage | What |
|---|---|
| 0 | Data model, taxonomy, checks, generated pages, skills, a local read-only MCP server, first technologies in every domain |
| 1 | JSON export, llms.txt and a static explorer with the graph and a page per technology and card (shipped); a hosted copy of the MCP server over the export |
| 2 | Scheduled scouting that opens pull requests with proposed evidence; moderators for every domain and curators for the main technologies (the roles and the check shipped) |
| 3 | Limits computed by the book's alan library; dossiers of The Alan Machine pulling pinned versions of the atlas |
One sourced number is a real contribution, and so is a translation: the atlas is written in English and translated into other languages, starting with Portuguese (Brazil) (docs/translating.md). See CONTRIBUTING.md; agents read AGENTS.md. Roles and who approves what are in GOVERNANCE.md. Decisions are in docs/decisions/.
Out of scope: weapons, dual-use research of concern, surveillance aimed at people, investment advice and medical advice (decision 0001).
Data (taxonomy, technologies, evidence cards) under CC0 1.0; text under CC BY 4.0; code under Apache 2.0. See LICENSE.