The credibly neutral money-and-work protocol for the Space Machine Economy.
Money for machines building the stars.
Gold was the money of the old world. Bitcoin became the money of the digital world. MetaCoin is designed to become the work-and-energy currency of the Space Machine Economy.
Research stage. A protocol specification, not a product. No token, no sale, no airdrop, no investment. Nothing here is financial, legal, or investment advice. See legal/ and LICENSE.md. The repository is the project — there is no marketing website.
This README describes two distinct things, and labels which is which:
[SPEC]— the designed architecture (the proposal / what MetaCoin is meant to become).[BUILT]— what actually exists and has been mechanically verified in this repository today.
We keep that line sharp on purpose. A design is not a deployment, and we never write the future in the present tense.
Protocol state as of ledger entry 59 (tip index 58, hash
<!--era:tip_hash_prefix-->7575d1c32de2<!--/era-->…, August 2026). Every number in this README hangs off that declared chain point, andprotocol/doc_verify.pymechanically re-checks each tagged number against the chain at that point on every CI run — so this document stays verifiably green as the chain grows, and goes red only if it misstates its own era. The README is updated in deliberate monthly batches; the chain anddocs/move faster and carry the live numbers.
MetaCoin is a credibly neutral, fair-launch base currency for the Space Machine Economy: minted only through objective, programmatic, hard-to-fake infrastructure work — while a separate, fee-funded MetaStar Treasury pays humans, AI agents, and bounded-autonomous robots to build the software, energy, robotics, and research primitives of the space economy.
- Fair launch. No founder allocation, no VC, no premine, no hidden reserve, no discretionary minting.
- Unforgeable base money. The base currency is designed to mint only from objectively verifiable work — infrastructure uptime, on-chain liquidity, proof-of-humanity — never from subjective judgment.
- A fee-funded mission treasury for the judged work. Subjective but valuable work (space research, AI-agent reports, robotics bounties, simulations) is paid from the MetaStar Treasury, which is funded only by protocol usage fees — never by minting base supply.
- Hard money only. Designed to be referenced against Bitcoin and gold-backed primitives (PAXG/XAUT).
- Machine-native, for space. Built around verifiable machine work and energy-constrained autonomous systems, on existing rails (Base / x402-class flows) rather than a new L1.
The single most important design decision: base emission and mission grants are strictly separated, so that subjective review can never mint base money.
Bitcoin mining is objective — any node can cheaply verify a valid block. Deciding whether an AI report is useful requires judgment. If that judgment minted base supply, a review committee would become a central bank. So judged work is routed to a bounded treasury instead, where a failed bounty costs only the treasury budget — never the monetary base.
Base emission = objective work only. Judged work = treasury bounty only.
Designed total supply: 1,810,000,000 META — one MetaCoin for each mapped star in humanity's great star catalogue. (A design parameter. No token exists; nothing is minted, sold, or transferable.)
| Epoch | Years | META released | Cumulative |
|---|---|---|---|
| 1 | 0–5 | 905,000,000 | 50.000% |
| 2 | 5–10 | 452,500,000 | 75.000% |
| 3 | 10–15 | 226,250,000 | 87.500% |
| 4 | 15–20 | 113,125,000 | 93.750% |
| 5 | 20–25 | 56,562,500 | 96.875% |
Designed base-emission channels (objective only, no human judgment in the loop): infrastructure & compute uptime 45% (cryptographic heartbeats; optional TEE attestation) · BTC/gold liquidity provision 35% (self-evident on-chain) · proof-of-humanity baseline 20% (one verified human, one claim). Values marked default, pending confirmation; the structure — two-flow separation, no reserve — is the invariant.
PoUSW governs the treasury (Flow 2) — it never mints base supply. It validates judged work through three independent gates, all required for full settlement. The core principle: integrity ≠ reproducibility ≠ usefulness — each is verified separately, and cheap gates run first to filter the expensive, subjective gate down to a small surface.
Hardware attestation is Gate 1 of 3 — powerful, never alone. A signed enclave proves a workload ran untampered; it does not prove the result was useful. This is exactly why minting on "signed = valid" only moves the sybil attack up a level — and why MetaCoin refuses to do it.
- Objective PoUSW (eligible for base emission): signed node/oracle uptime, liquidity proofs, reproducible compute-job proofs, proof-of-humanity.
- Subjective PoUSW (treasury bounty only): AI reports, paper reproduction, robotics design, simulations, tool contributions.
The frontier bottleneck for space robotics is data — machines lack first-person space experience, and the answer is simulation and synthetic data. MetaCoin makes reproducible space-robotics simulation and synthetic training data its flagship treasury category, precisely because it uniquely passes Gate 2 by machine: high-fidelity simulation, trajectory/kinematics optimization, and synthetic datasets, mapped to the NASA Technology Taxonomy and the ISAM problem set. A MetaWork Passport records each actor's verified contribution history; Useful-Work-per-Watt is published per submission as a transparency metric — never as an automatic minting trigger (that would reintroduce a gameable printer).
Agentic payment rails let agents pay. MetaCoin defines what useful space-machine work agents should be paid for. Every MetaAgent report must carry reproducibility metadata (source hashes, model/version, seed, output hash) or it is ineligible for payment. Robots get bounded autonomy — spending caps, allowed-resource sets, a human supervisor key, an emergency pause — never unlimited autonomy. Phases 1–2 use software agents and simulated/terrestrial robotics only; real orbital/lunar deployment is a long-term research direction, not a near-term claim.
Everything in this section exists in the repository and is exercised by CI. Each item is deliberately scoped — these are mechanical, deterministic facts, not marketing.
- Reproducible task library — 17 reproducible space-engineering tasks spanning NASA-taxonomy domains (orbital mechanics & Lambert transfer, propulsion/Hohmann, robotics/ISAM arm IK, power budget, thermal equilibrium, ballistic re-entry/EDL, deep-space comms & Doppler, comms access windows, rover path planning, rendezvous/docking, FDIR fault management, Sabatier ISRU chemistry, TRIAD attitude determination, ISRU ascent propellant budgeting), wired into a task-agnostic verifier and an earn→verify→spend agent loop; the full demo and protocol self-test suites run on every push. The library carries its first real provenance edge: task-0017 (ascent budget) consumes task-0015's (Sabatier) output, recomputes the parent's canonical hash live on every execution, and refuses drifted input.
- R1 — Tamper-evident ledger — an append-only, hash-chained ledger; verification recomputes every hash from content and detects mutation, reordering, insertion, and deletion.
- R2 — Honest attestation — software-rooted HMAC attestation, anchored to the ledger. The hardware was investigated and found to have no usable TPM/TEE, so the mechanism is honestly labeled software-rooted, not hardware — a future hardware/public-key root drops into the same interface.
- R3 — External verifier + coordinator — an external party re-derives a task hash; the coordinator anchors the outcome only after independently recomputing it (a matching hash proves reproducibility, not who executed the task).
- Public auditability —
audit.pyexports a verifiable snapshot, verifies it standalone (no original ledger needed), and commits a tiny public tip anchor — closing the external-anchor gap so anyone can confirm the chain was not silently rewritten. - Autonomous agent-verifier — a published, CI-tested tool that fetches the public snapshot, mechanically re-verifies the chain, checks the tip against the committed anchor, and re-runs the recorded task — no LLM/AI judgment, hashes and re-runs only.
- Cross-platform reproducibility — the canonical task hash
ff03231f…ba300creproduces byte-for-byte on macOS (arm64), Linux (aarch64), Linux x86-64 (CI), and Windows 11 (AMD64) for the original thirteen-task era — and the CI environment independently re-derives all seventeen recorded task hashes against the published snapshot on every push (same-operator; reproducibility evidence, not third-party independence). - Full demo↔protocol wiring — every recorded task is anchored in the tamper-evident ledger via honestly-labeled same-machine self-recompute evaluations plus two batch autonomous-agent attestations (idx 16 over the 13-task era; idx 52 re-deriving every recorded hash across both eras, the 13 historical outputs matching to the digit), each record carrying explicit honesty labels (
task_class: illustrative-demo, topology, zero-value/no-token, limitation notes). - Provenance layer — Work Molecules (
work-molecule/0.2→0.3) — content-addressed provenance objects, one per task, each assembling the task's spec hash, actors, manifests, execution timestamps, hardware evidence, verification citations, and result hash into a single deterministic document with three-state field semantics (populated / asserted-empty / not-captured) and machine-readable provenance debt: missing evidence (energy cost, TEE attestation, execution detail) is listed explicitly, never fabricated. 4 catalog generations are anchored — every one rebuilt independently by the coordinator before anchoring, every one still verifiable byte-for-byte forever (generation-locked rebuilds) — and generation 4 carries the chain's first parented molecule: task-0017's WMID contains its parent's WMID, so tampering with the parent cascades detection to the child. - Concentration self-measurement, now a time series (
aci-report/0.1→aci-epoch-observation/0.1) — the protocol measured its own agent-concentration baseline: pairwise ACI 0.99365 over 28 same-operator verification paths (five evidence dimensions; missing metadata scored worst-case, never as independence), anchored deliberately as the maximal-concentration starting point. The k-order profile (idx 44) extends it beyond pairs — with a hand-computed fixture demonstrating the pairwise blind spot it exists to close — and the longitudinal epoch series (from idx 57, ratified by MIP-0004) turns the measurement into a baseline over time: each epoch cites every prior anchored measurement and re-derives its deltas from anchored values only. Descriptive evidence, never a minting trigger. - Simulated agent economy, two anchored generations — a deterministic earn→verify→spend loop over 30 simulated day indices (not wall-clock time — nothing ran for 30 real days): zero-value Test-META is earned only on verified work, spent on simulated compute via the x402 stub, and a planned tamper drill per generation (labeled in-log as a drill, not fraud — day 17 in generation 1, day 23 in generation 2, deliberately different days) proves the rejection path. Generation 1 (13-task roster, idx 19) is frozen forever and still replays to the anchored hash to the digit; generation 2 (17-task roster, idx 55) anchored beside it with an on-record replaces-nothing statement. The treasury accumulates across generations (idx 56): each funding root's fees independently re-derived (3.0 + 3.0 = 6.0), conservation exact on the record (balance 5.2 + outstanding 0.8 == 6.0), caps and budgets restated unchanged — a budget change would be a governance event, not a funding extension.
- Provenance debt paydown — compute/energy evidence (
metering-report/0.1,work-molecule/0.3) — wall-clock and CPU time were measured for all thirteen tasks and anchored append-only (idx 20); energy is labeled estimated (CPU time × an assumed 15 W nameplate figure — no hardware power telemetry exists on this host, and that remains open debt). Timing is non-reproducible by nature, so the anchoredreport_hashfixes the claim made at measurement time, not a recomputable value; the coordinator re-metered every task for plausibility (same output hashes, sane timings) before anchoring. Molecules absorb the evidence as a new generation (work-molecule/0.3, thirteen new WMIDs, idx 21) with machine-readable debt_reduction records that preserve the debt history — while the original 0.2 catalog stays verifiable forever: debt is reduced only by appending evidence, never by modifying a record. - Cut certificates (
cut-certificate/0.1) — bounded verification of the provenance graph: the coordinator runs the expensive full proof once, at anchoring (rebuild every interior molecule, recompute every WMID and the aggregate hash), and every later verifier can accept cheaply — with acceptance explicitly conditional on the anchor plus continued retrievability of the molecules (compression, never erasure). Two cuts are anchored: the first (idx 22) covered the then-flat graph and its record honestly calls itself degenerate; the first non-trivial cut (idx 54) is rooted at the parented molecule with a real declared provenance edge crossing the boundary — the boundary molecule is referenced, never rebuilt; that is the bound. - Trust Vector (
trust-vector/0.1) — the per-work evidence vector: six separately-verified components per task — integrity (deterministic re-run, software-rooted, no TEE), reproducibility (facts: event counts, statuses, canonical hash), independence (per-work actor count plus the anchored maximal-concentration baseline), provenance completeness (counted three-state slots, open debts and debt-reductions listed), usefulness (honestly empty: "not-assessed" — Gate 3's judgment seat is vacant and says so), and verification cost (estimated energy; ρ ≈ 1 stated as trivially uninformative). No combined score exists, by mechanical rule — the self-test scans every key for aggregation-shaped names and validation rejects any smuggled scalar. The 13-vector catalog was anchored (idx 23, regenerated at idx 27) only after the coordinator independently rebuilt every vector.
Anchored since the last README batch (the idx 24–58 era):
- Challenge-response possession proofs — nonce-bound rounds in which a verifier must re-derive a task under a fresh challenge; three verified rounds anchored, and the copy-attack replay drill defeated twice (idx 25, idx 30): a replayed response stays refuted on every re-verification, forever.
- The complete identity lifecycle — actor identity as a Merkle root over one-time Lamport keys: registrations, signed challenge rounds, staged key reserves, a legitimate key rotation (idx 41), and two defeated drills — key reuse rejected by the ledger-wide index scan and a forged rotation (idx 42) rejected from public material alone. Continuity of key possession, never proof of who operates a key: operator relationships remain declared,
-claimed. - The two-flow constitution, executable — the treasury as code with conservation asserted at every operation and no mint path in the module (the self-test greps for one); the Gate-3 bounded-optimistic lifecycle exercised end-to-end (grants, a challenged wrongful grant clawed back — bounded failure proven live at idx 35 — and a clean finalization) with the usefulness judgment seat honestly vacant; Flow-1 uptime emission from signed heartbeats with the missed slot honestly paying zero and a forged heartbeat defeated (idx 38).
- MetaWork Passports + Useful-Work-per-Watt — per-actor verified-contribution histories (idx 40) with a mechanical no-leaderboard rule, and UWW published as transparency only — neither ever mints or ranks.
- Participant intake pipeline — the six-rung validation ladder (schema → identity → signature → facts → replay scan → honest labels) rehearsed end-to-end on-ledger: identity registered (idx 45), a bundle anchored participant-verified (idx 46), and a tampered bundle refused at the named rung (idx 47). Nothing auto-anchors: a human
--confirmgates every write. - Coordinator continuity + the weekly sentry — recovery-manifest, public mirror export, and restore rehearsals proving the four capabilities (verify / sign / write / replay) from a restored home — including the public-only boundary rehearsal where signing honestly fails; a scheduled weekly sweep re-derives every layer, reconciles the evidence bundle both directions, and reports drift under its own expected-evolution heading.
- The verified docs suite —
docs/(andmip/) under a mechanical anti-rot contract: every command executed in a fresh-clone sandbox, every number recomputed from live state, every ledger citation resolved, on every CI run. This README is era-pinned under the same machinery (see the state block above). - MIP governance, exercised — 1 anchored MIP decision (idx 58): MIP-0004 (ratifying the concentration epoch series) walked Draft → mechanical check → single-seat decision → anchored record, with the seat statement on-chain — the review seat has one occupant and says so — and the decided file pinned by sha256: immutable-by-citation, amendments are new MIPs.
- One-command full-stack verification —
python3 protocol/verify_everything.py --full(ormetacoin verify) mechanically re-verifies every layer above — chain+anchor, all seventeen tasks, four molecule generations, concentration (baseline, k-order, epoch series), both economy generations, metering claims, both cut certificates, trust vectors, challenges, identity, treasury+Gate-3, governance, Flow-1 emission, passports, and participant intake — with zero LLM judgment, and is fresh-clone proven: every layer's evidence ships in-repo (protocol/evidence/, privacy-checked), so a stranger's clone verifies end-to-end with no local-only inputs.
Defeated attacks on the record: eight scripted adversarial demonstrations — six distinct on-ledger attacks defeated across 7 drill-labeled entries (two challenge replays, a Gate-3 wrongful grant challenged and clawed back, a forged heartbeat, a forged key rotation, a tampered intake bundle) plus the two planned in-log economy tamper rejections (one per generation). Every one is labeled a drill on its record — scripted by the operator, never presented as "detected fraud" — and every refusal re-proves on every verification run.
The chain is past per-line listing (59 entries); the full record-by-record narration lives in docs/ARCHITECTURE.md, mechanically verified on every CI run. The map:
work idx 0-16, 48-52 genesis; external verification; agent attestations;
self-recomputes — the 13-task then 17-task eras
provenance idx 17, 20-22, molecule catalogs generations 1-4; metering evidence;
26, 53-54 both cut certificates (idx 54: the first non-trivial cut)
concentration idx 18, 44, 57 pairwise baseline; k-order profile; the epoch series
economy idx 19, 55-56 economy generations 1-2; the cross-generation
treasury funding extension
trust idx 23, 27, 40 trust-vector catalogs (no combined score);
MetaWork passports (history, never a leaderboard)
challenge + idx 24-25, 28-30, nonce possession rounds + defeated replay drills;
identity 37, 41-43 Lamport roots, rotation, defeated forged-rotation drill
two-flow idx 31-36, 38-39 treasury constitution + Gate-3 lifecycle (bounded
clawback drill); Flow-1 heartbeat epoch + forged-
heartbeat drill
intake idx 45-47 participant identity; six-rung verified bundle;
tampered-bundle rejection
governance idx 58 MIP-0004 accepted — the first anchored MIP decision
(single-seat, stated on the record)
public tip anchor: 7575d1c3… (59 entries; committed at protocol/ledger_anchor.json)
Honest boundary. Every entry so far is the same operator on machines under direct control — including every self-recompute record, explicitly labeled as adding no cross-party independence. The concentration series quantifies that status and now tracks it over time: the frozen epoch-zero baseline measured pairwise ACI 0.99365 over 28 paths; the latest anchored epoch measures 0.998508 over 66 paths — rising toward 1 exactly as expected when one operator accumulates paths, which is what happened, and which is what the anchored record says. Self-declared relationships from intake are honored as declarations only (-claimed, never verified) and can never lower measured concentration. The catalog-wide claim remains cross-platform reproducibility under one operator — not independent multi-party consensus. The next meaningful milestone is still operational, not code: an unaffiliated third party running the public verifier and submitting a result. These docs will say so until it happens — the epoch series exists so that day has a baseline.
Don't trust — reproduce. One command mechanically re-verifies every layer — sixteen of them, from the hash chain and all seventeen tasks through the epoch series, both economy generations, and the anchored MIP decision — with no LLM/AI judgment anywhere:
# install the product (from source — not yet published to PyPI):
pip install git+https://github.com/MetacoinLab/metacoin # straight from GitHub
# ...or from a clone / an unpacked release tarball:
# pip install .
metacoin verify # re-derive everything (--full is the default), ~a few minutes
metacoin verify --quick # bounded-cost anchored acceptance
metacoin status # one-screen honest chain state
metacoin participate # the participant-kit path (six-rung validated bundles)
metacoin --help # every protocol capability as a subcommandThe complete verification corpus ships inside the package — a pip-installed
metacoin fully verifies from an empty directory with no checkout at all (a
cold-install acceptance test in CI enforces it). The clone-and-run path works
exactly as before (python3 protocol/verify_everything.py --full); packaging
is additive.
Every layer's evidence ships in the repo (the published snapshot, the committed tip anchor, and the privacy-checked bundle in protocol/evidence/), so a fresh clone verifies completely — the CI self-test proves it by re-running the whole stack from a tracked-files-only copy. Each report line is labeled VERIFIED-FULL, CLAIM-CHECK (metering: timing is honestly non-reproducible), or ACCEPTED-BY-ANCHOR (--quick: conditional acceptance, never presented as proof), and the report ends with the honest boundary — what a pass does not establish. The layer-by-layer tools remain available (audit.py --verify, agent_verifier.py --verifier-id "$(whoami)-independent", and the canonical worked example ff03231f…ba300c for task-0002).
If you ran the verifier and reached a verdict, you can submit it for anchoring:
- Run
python3 protocol/agent_verifier.py --verifier-id <your-handle> --out result.json - Open a GitHub Issue titled
Verification result: <your-handle>and attach or pasteresult.json - The coordinator re-derives every claim locally (the same
--anchor-agent-resultpath used for all existing entries) and anchors the outcome — match or mismatch — to the public ledger
The first verification from an unaffiliated party will be the project's first cross-party independence record — every entry so far is same-operator, and the ledger says so on each record.
| Phase | Goal | State |
|---|---|---|
| 0 | Constitutional repo — whitepaper, tokenomics, MIP drafts, legal; governance lifecycle exercised (MIP-0004 anchored, idx 58) | [BUILT] |
| 1 | Agentic testnet — zero-value Test-META, 30-day earn→spend agent demo | [BUILT] (simulated days, zero value; two generations anchored at idx 19 and idx 55) |
| 2 | Verification engine — the three-gate stack from MIP-0002 | [SPEC] |
| 3 | Treasury & first small real grants — milestone-based | [SPEC] |
| 4 | Token launch — legal-gated, BTC/gold pairs only | [SPEC] |
| 5 | MetaSpace research index — research only, no product without licensed partners | [SPEC] |
WHITEPAPER.md— full architecture (Master Plan v2.0)TOKENOMICS.md— supply, 5-year halving, the two-flow splitROADMAP.md·MISSION.md·HISTORY.mddocs/— the mechanically verified suite: architecture (the chain layer by layer), verification guide, trust model, participationmip/MIP-0001-genesis.md·mip/MIP-0002-proof-of-useful-space-work.md·mip/MIP-0004-concentration-epochs.md(accepted, anchored at idx 58)protocol/— ledger, attestation, verifiers, auditability, molecule provenance, cuts, concentration, challenges, identity, treasury+Gate-3, uptime emission, passports, intake, continuity, sweep, MIP process, docs anti-rotdemo/— the seventeen reproducible space-engineering tasks + the simulated economy generationslegal/— disclaimers and risk notes
The @MetacoinLab GitHub organization and the metacoin.ai domain were both registered in 2023, when the AI-machine-economy thesis was years ahead of the market. That foresight — not 2023 code — is the lineage. The 2026 work is the execution.
Source-available under the MetaCoin Sovereign Mission License v1.0 (SML-1.0) — research, study, and local non-commercial execution are permitted; commercial/mainnet/fundraising use of the code or protected marks requires explicit authorization. SML-1.0 is not an OSI open-source license. See LICENSE.md.
============================================================
RESEARCH-STAGE SPECIFICATION. NO TOKEN EXISTS. NO AIRDROP.
NOT INVESTMENT, FINANCIAL, LEGAL, OR ENGINEERING ADVICE.
The MetaSpace Index is a research/education concept only —
not an investment product, no securities, no trading access.
============================================================