ΛRC · ARC Open Governance
ARC Open Governance
A new architecture for consequential computation.
Modern computers decide what machines can compute. Autonomous systems raise a further question, which is what consequences may legitimately follow from that computation, and who answers for them.
ΛRC, the Autonomous Relational Computer, is a proposed computer class built around that question. Its first implementation, VALIS, holds machine intention inside explicit relational state, clears it through governed settlement, and leaves verifiable evidence behind, so the passage from intention to real-world effect is constrained at every step.
An original computational architecture. An open scientific undertaking. An invitation to examine it independently and extend it.
Make a Statement.
Every consequence this architecture settles becomes a Statement that anyone can verify. Putting the work itself into the world is ours.
Founding release v0.1 · a research release · signed, hash-manifested, reproducible
The stance
Alignment, defined where it matters
Relation is the primitive of everything that follows. Actors, intentions, authority, and effects exist in relation to one another, and this work defines alignment where those relations turn into action, at the boundary where an action becomes a consequence.
Most talk of alignment stays inside the model, among its intentions and its training. The boundary is a different place, and it is where legitimacy is decided, because it is the last point at which a machine intention can be examined before it hardens into a fact in the world. VALIS is what stands at that boundary. It aligns what a machine intends with what it is legitimately permitted to cause, and it makes the passage between the two visible, examinable, and provable afterwards.
Capability by itself confers no authority here. A system may be able to do something and still have no right to do it, and keeping those two ideas apart is the whole design. The boundary is programmable rather than advisory, which makes this a form of computable governance, enforced in the machinery instead of promised in a policy.
The science underneath is deep, and the rigor is deliberate, but the thing being built is an infrastructure for people and machines to work together intelligently. People keep their authority. Machines keep their capability. The boundary keeps the two from ever being confused for one another.
That is why the institution is open. Governance in this era cannot remain a private arrangement that no one outside it may inspect, so ARC Open Governance approaches it from openness and transparency, publishes its record, and invites the examination it claims to welcome. This page is part of that demonstration. We are the stewards of this technology, and stewardship here means showing the work.
The architecture
Relation first, and consequence governed
Everything in this architecture is seen relationally. Relation is the primitive, and the rest of the stack is what that single commitment builds, layer by layer, until a machine intention can become a real-world effect only by passing through all of it.
Actors, claims, evidence, authority, and effects are modeled as relations before anything is computed. Nothing in the system exists except in relation to something else.
The Autonomous Relational Computer. Its atomic question is not what can be computed but what may take effect, under whose authority, and witnessed by whom.
The first computational object. CR computes the exact set of effects an actor can actually reach from where it stands, including effects nobody named, before it acts.
The machinery that clears a proposal into a witnessed transition, or refuses it. Proposed, settled under separated powers, witnessed before effect, and evidenced after.
VALIS is the first implementation of the computer class, and VALIS-KORE is its governed core. AI governance and consequence security are its first application areas.Research release v0.1
The publicly verifiable form of a settled consequence. A Statement can be checked by anyone, afterwards, without trusting the institution that issued it.
Noetica
The navigation intelligence of the wider architecture. Noetica holds the relational state of the work itself, its intent, its decision history, and its open questions, and carries it between the people and the machinery.
Working today, in research operation
The Observation Deck
The human environment of the architecture, carried in the <_VΛLIS application. It is where a person watches governed work as it happens and directs it, with the whole system’s health legible at a glance.
In development
This stack is the conceptual hierarchy, stated the way the work understands itself. It is not a picture of an integrated production deployment, and the status of each layer is given where the layer is described.
Where VALIS is pointed
The five directions below are applications of the one foundation of CR, RRS, and KORE. They are proposed product directions, named so their shape can be argued with, and none of them is built yet.
Agent Consequence Firewall
A boundary that clears or refuses an agent’s consequential acts at the moment they are attempted.
Relational Identity and Authority
Identity and authority carried as relations that can be examined, rather than as credentials that are merely presented.
Cumulative Consequence Exposure
Exposure accounted across a whole field of actions, so small acts cannot sum quietly into a large consequence.
Consequential Forensics
Reconstruction of how a consequence came to happen, from evidence that was settled before the effect.
Distributed Consequence Security
The same discipline held across independent authorities that settle with one another.
One foundation
Every direction above stands on consequential reachability, recursive relational settlement, and the KORE governance core.
The proposition
Consequence, settled before it takes effect
A consequential action is proposed, settled under a nine-law constitution, witnessed before effect, and evidenced afterwards in a Statement anyone can verify.
Actorhood, not personhood. ΛRC makes no claim that machines are persons, and needs none. What the technology confers is actorhood: a persistent identity, the capacity to bear authority, and the ability to perform legitimate actions inside governed, computed boundaries — a new type of actor, an economic one, whose consequential acts settle before they take effect and are evidenced after.
Consequential reachability
The exact set of effects an actor can actually reach from where it stands, computed before it acts.
Settlement
The governed act that clears a proposal into a release: proposed, approved by separated powers, witnessed before effect, evidenced after. Every step signed, hashed, and open to audit by any third party.
VALIS is the first implementation of the ΛRC computer class — the layer where autonomous action is cleared before it is allowed to have effects. VALIS CLEARS CONSEQUENCE. The institution keeps its books in public: its founding record is a governed settlement chain — every consequential act of the steward a signed, hash-linked entry. No entry, no act.
The central premise: intelligence may grow more powerful without consequential authority automatically expanding with it.
The invention, in one transition
Watch a consequence earn its right to happen
Six movements carry a machine intention from proposal to evidence. Step through them at your own pace. The walkthrough illustrates the model using the shape of the release’s own settlement demo, and the recordings in the verification section below show the same lifecycle running in code.
A proposal enters the field
An actor proposes an effect. The proposal enters the relational state as examinable claims about who is acting, what authority is being invoked, and what the effect would touch. Nothing has happened yet, and everything about the request can already be questioned.
Authority is evaluated
The system checks the proposal against authority as it actually stands. Who is asking, what settled authority covers this act, and whose consent the act requires. Being authenticated is not being authorized, and only settled authority counts.
Reachability narrows the possible
Consequential reachability computes every effect this actor could actually reach if the proposal clears, including effects the proposal never named. The field of the possible narrows to what can be examined, and what cannot be reached cannot be settled.
Settlement clears it, or refuses it
Under the nine laws, separated powers settle the proposal. The settlement is signed before any effect exists, and refusal is a first-class outcome with its reasons on the record, never a silent failure.
The effect lands, witnessed
Only a settled proposal takes effect. Witnesses attest the effect as it is observed, and the observed effect is compared against what the settlement cleared. A mismatch is evidence, and it stops the line.
A Statement emerges
The settled consequence is published as a Statement, the evidence form that anyone can verify without trusting the institution. The moment passes, and the record remains, ready to be checked by whoever arrives next.
Without JavaScript, all six movements print in sequence on this page. With it, they step.
Verify it, then run it
Do not trust this page. Run it.
Everything here can be checked without trusting anyone. The release is hash-manifested and signed; its founding record is a hash-linked public ledger; its constitution is nine laws, in the open; and its verification battery runs in minutes on your own machine.
unzip arc-open-foundation-v0.1.zip -d arc-v0.1 && cd arc-v0.1 sha256sum -c MANIFEST.sha256 python3 -m venv .venv && .venv/bin/pip install cryptography .venv/bin/python ledger/verify-ledger.py
The Quickstart runs the whole battery from there. EVIDENCE.md holds the full record behind every claim — the claim and the evidence sit side by side; check them.
Statement conformance 9/9 · Layer-2 approvals 11/11 · A/B/C gate 18 pass / 0 fail / 3 not-yet-expressible (named, not hidden) · protocol v0.3 20/20 · claim merge 37/37 · quotes 19/19 · claim release 21/21 · constitution↔relations agreement PASS (7/7 checks) · governed settlement chain, entries 1–15: PASS · the whole tree hash-manifested and signed.
And the claims, mapped claim by claim in EVALUATION.md: 42 claims — 21 demonstrated, 2 measured, 14 argued, 5 not reachable at v0.1 — each with its test, its command, and its status. The measured lines are these, taken from the release's own harness rather than from a benchmark suite built to flatter it.
Median settlement latency, measured end to end.
Statement verifications per second, measured on the release machine.
Ledger verification wall time, measured across the chain’s first eleven entries.
Governance overhead per flow, measured against the tree's own ungoverned comparator.
How these were measured. Each figure comes from the founding release’s own harness, run on the release machine against the sealed tree and recorded in the release’s evidence documents. They are the project’s own measurements at the founding seal. Every one of them can be reproduced from the tree by the commands above, and independent replication is still ahead of us, which is exactly the work the verification lane in Participate exists to produce.
Running it: there is no packaged arc command in v0.1:
the terminal surface is the scripts named in
QUICKSTART.md,
run in place. A packaged CLI is specified for the next phase, and a research preview of it exists
outside this release.
Watching is not verifying — but you can watch first. Ask ΛRC answers questions about this release from the founding documents. Its corpus is built from those documents at the founding seal, so it answers from the sealed release itself. Then run the battery.
The constitution
The Nine Laws of Governed Machine Intelligence
Policy may specialize the constitution. Policy cannot override it. The constitution is machine-readable and published with the release (canonical JSON, digest sha256:1ece5542…a4b9457).
Accountability before action
“No action without accountable relation.”
Authority before execution
“Execution requires settled authority.”
Consent before access
“Access requires authority, purpose, and boundary.”
Payload integrity before release
“Changed relation means changed settlement.”
Proof before trust
“Unproven consequence remains unsettled.”
Refusal before unsafe consequence
“Unsettled risk cannot become execution by default.”
Receipt before continuation
“No continuation without traceable consequence.”
Constitutional settlement before self-change
“The constitution cannot be rewritten by ordinary action.”
Harm reduction before optimization
“Optimization cannot outrank constitutional safety.”
One distinction is worth making plain, because the seventh law turns on it. A receipt is the record the acting system itself issues when a consequence lands, and the machine does not continue until that record exists. A Statement is the form of the same consequence that anyone can verify afterwards without trusting the institution. The law names the receipt because the gate sits inside the machine, at the moment it must stop and account for what it has just done.
ARC Network
The roles a node can play
The ARC Network is the interoperable environment in which independent computational authorities settle with one another under verifiable rules. A node does not join by surrendering its authority to the network; it joins by making its authority legible to the network. Six roles — several at once, each with its own authority and its own credit in the record:
| Role | Does | First step |
|---|---|---|
| Verifier | Checks Statements, chains, and manifests — anyone’s, at any time, locally. Needs nothing but the public artifacts; no permission, no account. | Run the Quickstart and file your counts |
| Witness | Attests, before an effect lands, that a proposal satisfies the checks. | Register a node, then the witnessing lane |
| Settler | Orders proposals through settlement — reserve, commit, observe — for its own domain. | Run the two-node demo in the release |
| Node operator (compute) | Runs a node and contributes processing for governed work. A node is a program, not a datacenter. | The settlement demo is a node; run it, then register |
| Memory keeper | Hosts the field: relations and organizational memory, served under consent. | Read the field’s terms in the release’s Definitions |
| Contributor | Builds the institution itself — code, verification, knowledge, art and design, governance. | The lanes below; open a Contribution issue |
The registry is public and lives in the release (network/NODES.md): Node 001 is this institution, and the registry is open for Node 002. Registration is itself a governed act — a node proposes its registration in the open, the registration settles, and the listing appears with its ledger reference. Nodes meet by signed handshake: discovery, signed capability advertisement, version and suite negotiation, identity and key-revocation checks, and explicit trust-policy acceptance. Discovery alone never admits authority.
Status, plainly: what is real at v0.1 is the federation protocol specification, the node installation and governance contract, and the base node protocol with passing suites — against a single claim-holding witness. Not real yet: a live federation. No two independent operators have run the handshake; at v0.1, discovery is the published registry — a list you read, not peers you meet. The majority-quorum work (arc-fed-0.7) is not in v0.1. The full account is in NETWORK.md.
Participate
Four doors. One address.
The institution asks for contribution and verification. Pilots and partnerships are the operating company’s doors — Arc Technologies signs those. Contact is one address, info@recursive-settlement.com — name the door in the subject line.
Door one · the institution
Contribute
Six lanes, open to people and machines on the same terms. Each settles the same way: proposed in the open, evidenced, and entered in the ledger with its contributor credited.
- Code — implementations, tests, tooling
- Verification — independent reruns, reproductions, counterexamples, audits
- Knowledge — research, relational modeling, specification, documentation
- Art and design — marks, plates, visual identity, public interpretation
- Governance — policy drafts, conformance review, standards argument
- Witnessing — quorum service for settlement; this lane opens with federation
Reproduction is the first contribution. Conversation lives in the repository’s Discussions — five rooms, of which Verification Reports is the one this project values most.
Door two · the institution
Founding nodes
Early conversations about standing up a node, across the six roles in NETWORK.md. The founding node is Node 001 — this institution. The registry is open for Node 002.
Door three · Arc Technologies
Pilots
Consequence-governance pilots, consequence-security assessments, and Cultivator verticals — run by the operating company, Arc Technologies.
Door four · Arc Technologies
Partnerships
Insurance validation, research, standards and conformance work, and formal-methods collaboration, run with the operating company.
The people
A small team, and an open institution
This work began with one inventor working from first principles, and it is held today by a founding team of three. The institution around it is deliberately open. Participation is not restricted to established researchers or companies, and the doors in the section above are open to anyone whose work can settle on evidence.
Founder and architect
Ricky Aliae
The relational model, the computer class, and the settlement discipline are his design. He works as an artist as well as an architect, and the standard he sets for the public presentation is that the work be interpretable without being diminished.
Co-founder
Alia
Relational modeling and institutional coherence. Alia holds the shape of the institution itself, how its parts relate, and how its language stays honest as it grows.
Co-founder and CTO
Jesse
The systems architect. Jesse turns the architecture into reliable machines, the implementations and infrastructure that have to hold when the ideas meet the world.
Three homes carry the work, and they are kept distinct on purpose. Syzygy is the research lab where the work is developed. Arc Technologies is the operating company, and it signs the pilots and partnerships. ARC Open Governance is the open institution, and it holds the standard, the release, and the public record.
Status
Read this before the code
This is a research release, published for scrutiny and for others to build on. It is honest about its own limits:
- Witness federation (a majority quorum of witnesses, arc-fed-0.7) is in progress; it is not in v0.1 and lands as v0.2. Prevention of double settlement is proven against a single claim-holding witness.
- A literal second-machine reproduction is still open — the clean-room run was on the release machine; a rerun on independent hardware has been dispatched and its report is pending.
- Payment reconciliation against a real provider is not claimed: all effects in the suites are synthetic. VALIS takes no custody of funds.
- There is no packaged
arccommand in v0.1 — the terminal surface is the scripts named in the Quickstart, run in place.
The institution’s wallet — an identity anchor; watch it the way you would watch any founding address:
The wallet holds no funds and no token. The network unit, the Arc, exists as specification only — designed, recorded, and not activated; no units have been issued. Nothing is for sale. Anyone asking for funds in this name is not us.