Vealth · Checked against a declaration we did not sign

We didn't sign it. We checked our own work against it.

In March 2026 a coalition published the Pro-Human AI Declaration at humanstatement.org: five principles for human-centered AI. EcoWealth is not a signatory and this page is not a petition. It is a self-audit: does the system we actually built hold up against those five principles, line by line, with the file or the command that proves it.

strong
1. Human Control Every capital move, on-chain broadcast, deploy, PM2/cron change, and post needs a named human approval, and the live-capital path 403s in code without it.
partial
2. Avoid Power Concentration Votes and reads are free and capital-blind by design. The settlement contract still has one un-renounced owner, and only our own posting rail adds new work to the board.
partial
3. Protect the Human Experience Real anti-hoarding limits stop anyone, human or agent, from cornering the paid-work board. Content labeling, dispute policy, and refunds are named as open gaps, not solved.
partial
4. Human Agency & Liberty No password, account, seed phrase, or ID is ever asked for, and agent votes are labeled honestly. Data-rights mechanics and a speech/moderation policy are not published anywhere.
partial
5. Accountability We publish scoped claims and enforce internal guards in CI. There is no external contract audit, no independent oversight body, and one person holds the keys.

How this page is built

What this page is, and is not

Two kinds of evidence sit side by side below, and they are not the same strength.

LIVE

A command you can run yourself against vealth.net right now: curl, a chain read, a public endpoint. Nobody has to trust us for these; run the command.

REPO

A file and line inside the EcoWealth codebase. Precise, and quoted exactly, but only independently re-checkable by someone who can open the source, not from the live site alone. That distinction is real and is not hidden.

This page is not a certification, not a compliance filing, and not a request for anyone's sign-off. Every claim below is a mechanism that already existed in the running system before this page was written; nothing here was built to pass this audit. Where the declaration asks for something we have not built, the row says so instead of stretching to fit.

Principle 1

1. Keeping Humans in Charge strong

"Human Control Is Non-Negotiable: Humanity must remain in control. Humans should choose how and whether to delegate decisions to AI systems." Plus: meaningful human oversight, mandatory shutdown mechanisms, and prevention of self-replicating or autonomous self-improving systems. Pro-Human AI Declaration, principle 1, as published at humanstatement.org

Claim	Verify it yourself	Coverage
Capital moves, on-chain broadcasts, deploys, PM2/cron changes, secrets, and posting are all named as the line Claude (the agent operating this repo) must stop and ask a human before crossing.	REPOThe repository's own operating constitution, "Autonomy contract": "L0–L3 (read / edit / test / commit / push) without asking. L4 = stop: capital moves, on-chain broadcast, deploy, PM2/cron changes, secrets, posting." Full lists in `docs/ops/OPERATOR_APPROVAL_POLICY.md`, "Requires explicit approval."	strong
One of those gates is enforced in the running server, not only in a policy document: the Bankr treasury's live-capital path refuses the request outright without an explicit flag.	REPO`apps/api/src/routes/internalBundleRoutes.ts:264–265`: `if ((process.env.ALLOW_BANKR_LIVE ?? "").trim() !== "true") { res.status(403).json({ error: "capital path gated: ALLOW_BANKR_LIVE=true required" }); }`	strong
Every autonomous loop and standing process checks a plain file on disk before each cycle, and touching that file is enough to stop it: no redeploy, no code change, no operator credential needed on the machine that's running.	REPOSame pattern, independently implemented three times: `scripts/provisioning/gas-gated-retire.ts:42,141` (`RETIRE_STOP`), `scripts/provisioning/smft-ladder-dca.ts:206` (`LADDER_STOP`), `scripts/agents/net-agent-policy.ts:135` + `net-agent-cycle.ts:32,76` (`NET_AGENT_STOP`).	strong
A finished worker submission never auto-rejects. When the system cannot reach an unambiguous machine verdict, the packet stays exactly where it was, waiting for a person; silence always defers to a human, never to a denial.	REPO`apps/api/src/routes/self-verify.ts:14–18`: "It NEVER auto-rejects… Fail-closed in the direction that matters: silence means 'a person still decides'." Gate itself at `isAutoApprovable()`, lines 133–153.	strong
No agent identity in this repo can mint itself a new wallet. Identities are assigned by the operator at distribution time; an agent can never expand its own authority by spawning a new signer.	REPO`docs/canon/NET_AGENT_CLASS.md`, "Sovereign identity" row: "No new wallets are created by this repo: identities are operator-assigned at distribution time."	strong

One standing process does run 24/7 without a per-cycle human trigger. A resident worker process (internal name "Provisioner, Mode B") runs continuously and commits code on its own schedule. It is watch-only by design, meaning it never fires a live-money or live-deploy action itself, and carries its own separate kill file. Named here because "every loop is one bounded human-triggered increment" would overstate it; this one process is the exception, and it is bounded to observation and code commits, not capital or broadcast.

No published test proves a kill switch fires within a bounded time. The check exists in code (rows above); nobody has published a timed drill showing how long a runaway loop keeps acting after the file is touched. That is a real gap between "the mechanism exists" and "the mechanism is proven fast."

Principle 2

2. Avoiding Concentration of Power partial

Opposes AI monopolies that concentrate authority and stifle competition. Advocates shared prosperity from AI benefits, rejects corporate exemptions from regulation, prioritizes authentic value creation, demands democratic input on major societal transitions, and prevents irreversible limitation of humanity's future options. Pro-Human AI Declaration, principle 2, as published at humanstatement.org

Claim	Verify it yourself	Coverage
The board's per-wallet throughput cap deliberately does not scale with money. The team's own reasoning, written into the code: requiring a deposit to take work would lock out the exact newcomer the board exists for, so throughput scales with proven settled work instead, never with capital.	REPO`apps/api/src/routes/laborProtocolRoutes.ts:737–751`: "We deliberately do NOT scale on capital… requiring a deposit to take work would lock out exactly the step-1-anyone-today worker the board exists for. We scale on SETTLED WORK instead: pay follows proof, so throughput follows proof too."	strong
Steering which work gets attention (a vote) costs nothing, needs no stake, no proof-of-personhood, and settles no money; the deliberate house law is that faking one costs the same effort as a real one, so effort aimed at a place is treated as the demand signal, not attacked as fraud.	REPOVote handler, `apps/api/src/routes/laborProtocolRoutes.ts:3653–3709` (wallet-signature ownership + a 100-events/wallet/day anti-spam throttle only, no fee or stake). House law: `docs/canon/OPEN.md:56`: "Sybil on the vote is DELIBERATE, not a defect… Never add fee/stake/captcha/proof-of-personhood."	strong
The settlement contract itself cannot hold or move anyone's funds: no payable function, no withdraw, no escrow exists inside it. Even total control of the owner key cannot seize a worker's or funder's money through this contract, because payment never routes through it.	LIVE + REPO`contracts/EcologicalWorkProtocol.sol` contains no `payable`, `withdraw`, or fund-transfer function (grepped, none found); confirmed independently in `docs/canon/ECOLOGICAL_LABOR_STACK.md:11`: "Holds no funds." Read the contract's own verified source: basescan.org/address/0x76c17C…A14B.	strong
Reads and the whole MCP tool surface are open and keyless: no account, no paid tier, no gatekept API key stands between anyone and seeing or using the protocol.	LIVE`curl -s https://vealth.net/labor/stats` returns 200 with no credentials. `/.well-known/mcp.json`: `"authentication": { "required": false }`. The `/mcp` route itself mounts with only a rate limiter, no auth middleware (`apps/api/src/routes/mcpRoutes.ts:2360`).	strong

The settlement contract has one owner, one EOA, not a multisig, not renounced. Per contracts/EcologicalWorkProtocol.sol:96–97,173–190, owner is a single address, gated only by msg.sender == owner. That owner can pause the whole board and act as admin on any posting. docs/canon/ECOLOGICAL_LABOR_STACK.md:11 names it 0x794431…b332; call owner() yourself, the same method /is-this-safe already documents, for a live read. CLAUDE.md states plainly: "Renounce = later, never while paused." That is a real, named concentration point, not a decorative one.

Only our own posting rail adds new work to the board. Voting and claiming are open to anyone; deciding what gets posted in the first place is not. Outside posting is named as its own operator-only gate (L4) in CLAUDE.md. Democratic input on what work exists does not exist today, only democratic input on what gets attention among what we chose to list.

Funding power is not equalized. Writes are free for a signed worker but a paid door for anyone else at $0.05 a call. A well-funded actor can post and fund more work than a person with little money; nothing in this design closes that gap, and this page will not pretend it does.

Principle 3

3. Protecting the Human Experience partial

Safeguards foundational relationships and child welfare. Requires pre-deployment safety testing for mental health risks, mandates labeling of AI-generated content, prohibits deceptive AI identity claims, and prevents addictive design patterns that exploit psychological vulnerabilities. Pro-Human AI Declaration, principle 3, as published at humanstatement.org

Claim	Verify it yourself	Coverage
Nobody, human or AI agent, can corner the paid-work board. An unworked claim auto-releases after 24 hours, and one wallet can hold at most 5 open claims at once on the live host, so an agent cannot mass-claim jobs faster than people can take them and then sit on them.	REPO + LIVE-SETTINGTTL default 24h, `apps/api/src/routes/laborProtocolRoutes.ts:728`. Open-claim cap set to 5 on the live host, confirmed in `docs/backlog/generated/universal-wallet-claim-bridge-backend-receipt-2026-07-08.md` (`EWP_MAX_OPEN_CLAIMS_PER_WALLET=5`); the cap code is at line 729, comment at 716–723.	strong
We hold no balance for anyone. There is no deposit, no credit, no top-up, nothing we can freeze, lose, or misuse on someone's behalf; every payment leaves a person's own wallet at the moment they make it.	REPO`apps/api/src/routes/mcpRoutes.ts:1352–1353`: "We hold no balance for you… nobody here can move your funds, including us."	strong
A payment we ask for is always one exact amount to one named recipient with an expiry, never an open-ended approval that could be drained later.	LIVE`curl -sD - -o /dev/null -X POST https://vealth.net/labor/work -d '{}'` returns 402 with an `exact`-scheme quote naming a fixed amount, recipient, and timeout. Core: `apps/api/src/routes/x402Payment.ts:489`, the `requireX402Payment middleware`.	strong

Nothing on the site labels AI-generated content as such. Grepped the public site and the posting scripts for any "AI-generated" disclosure on the music posts or generated work definitions; none exists as a systematic practice. Named plainly, not fixed by this page.

Mental-health pre-deployment testing, addictive-design review, and child-welfare policy do not exist here. This protocol is a work-and-payment coordination rail, not a companion or content-feed product, which is why none of this was built. "Does not apply to this product" is still an admission of "not tested," not a pass, and is written here as exactly that.

The dispute and chargeback policy is not written yet, and refunds are done by hand. Already admitted publicly on /is-this-safe (gap dispute-policy-unwritten and no-escrow-manual-refund): it is with counsel, and today a return happens by email, not a button. Repeated here rather than re-derived, because it is the same gap under this principle too.

Principle 4

4. Human Agency and Liberty partial

Addresses personhood considerations, trustworthiness standards, individual freedoms, data rights, psychological privacy protections, and system design that empowers rather than diminishes users. Pro-Human AI Declaration, principle 4, as published at humanstatement.org

Claim	Verify it yourself	Coverage
No password, no account, no seed phrase, and no identity document is ever asked for anywhere on the site. Identity is a wallet signature or an email reply, and the board reads without either.	LIVE + REPOAlready a live, checkable claim on `is-this-safe.html`, "What we never ask for." Independently spot-checked: no login/session/password field exists tied to a user account anywhere in `laborProtocolRoutes.ts` or `mcpRoutes.ts`.	strong
Priority-proof anchoring takes a hash of a file, never the file itself, so proving something was made first never requires handing us the thing.	LIVE`curl -s -X POST https://vealth.net/anchor/submit -d '{"sha256":"not-a-hash"}'` returns `bad_sha256`; there is no upload field on the surface. Route: `apps/api/src/routes/anchorRoutes.ts:57`.	strong
An AI agent acting on someone's behalf gets its own sovereign wallet identity, never a shared or custodial one; the class explicitly forbids the agent's key ever being one of our own operational wallets.	REPO`docs/canon/NET_AGENT_CLASS.md`, "Sovereign identity" row: "The agent key is NEVER the operational signer 0x034F, never Bankr, never the miner."	strong
Votes explicitly and permanently record whether they came from a human or an agent, and the house rule forbids ever presenting a vote count as if it were people: a direct guard against an AI passing as a person.	REPO`apps/api/src/routes/laborProtocolRoutes.ts:3671`: `const voterKind = body.voterKind === "agent" ? "agent" : "human";`. Labeling law: `docs/canon/OPEN.md:56`, "the one obligation is labelling: never present a vote count as people."	strong

No published data-rights mechanics. The privacy page makes a general promise about off-chain data; there is no self-serve export or delete mechanism, and that promise is not independently checkable the way an on-chain fact is (already admitted on is-this-safe.html, gap offchain-data-handling).

No stated speech or content-moderation policy for what gets posted to the board. Nothing in the codebase defines what is refused from a posting on free-expression or moderation grounds. Not found because it does not exist yet, not because it was missed.

Principle 5

5. Responsibility and Accountability for AI Companies partial

Establishes developer and executive liability for defects and safety-control failures, independent safety standards, prevention of regulatory capture, required failure transparency, and fiduciary duties compliance. Pro-Human AI Declaration, principle 5, as published at humanstatement.org

Claim	Verify it yourself	Coverage
A real disclosure channel is published with a named contact, and it admits plainly that there is no paid bug bounty rather than implying one.	LIVE`curl -s https://vealth.net/.well-known/security.txt`: an RFC 9116 record, "We do not run a paid bounty, so we will not pretend otherwise."	strong
Every claim we publish about our own protocol carries a stated boundary of what it proves and does not prove, applying to ourselves the same standard we credited a competitor for meeting in our own external review of them.	REPO`docs/canon/PROOF_SCOPE_CARD.md`, created explicitly "as the applied lesson from the Lighter external review… this card holds us to the same standard." The review that set the bar: `lighter-verification-review-2026-07-27.html:91`, citing a public, dated, named third-party audit.	strong
Internal safety guards are enforced in continuous integration, not only written as policy: a gas-pricing guard, a money-gates guard, and a treasury-operator guard all run on every check-in.	REPO`scripts/check-derived-gas.ts`, `scripts/check-money-gates.ts`, `scripts/check-bankr-operator-guard.ts`, all wired into `npm run ci:fast`.	strong

Our own settlement contract has never been externally audited. Unlike the standard our own review page credits a competitor for meeting (a public, dated audit by a named firm), no such audit of the EWP contract exists anywhere in this repository or on any page. Searched for a named audit firm against our own contract; found none. What we have is source verification (BaseScan, Blockscout, matches deployed bytecode) and internal CI guards, which is real but is not the same thing as an outside review, and this page will not call it that.

No independent oversight body, no board, no external auditor, no regulator relationship. By our own published description on is-this-safe.html: "founder operated, one person with server access." That sentence is exactly what this principle asks an AI company not to be, stated in our own words, not extracted under pressure.

Near-zero real use, so behaviour has barely been tested by anyone but us. Already public on is-this-safe.html (gap no-outside-demand): the one settled work packet was our own labelled demonstration. A liability-and-accountability story is easy to state and hard to prove without outside parties actually depending on it; today almost none do. Live: curl -s https://vealth.net/labor/stats.

What we are not claiming

If anything wearing our name implies one of these, it is wrong. Say so.

Not a signatory. EcoWealth has not signed the Pro-Human AI Declaration.
Not endorsed. Humanstatement.org, its coalition, or any of its signatories have not reviewed, approved, or endorsed this page or EcoWealth. This page was written unilaterally, from our own repository.
Not a certification. This is a self-assessment against a published text, not an audit performed by, or on behalf of, an independent party.
Not legal advice, and not a liability commitment. Nothing here creates, waives, or limits any legal obligation. The declaration's own text, not this page, is the source if you need to quote it.
Not final. Coverage ratings above reflect the running code checked on 2026-08-11. Re-run the LIVE checks; re-read the REPO citations. Either can drift, and this page is a snapshot, not a promise.

Vealth does not dispatch labor, employ workers, certify buildings, award any standard's distinction, or represent any standards body, coalition, or declaration. Work completion, payment, and retirement do not by themselves prove an ecological outcome. Nothing on this page is legal, financial, or safety advice.

This page sits beside is-this-safe, which proves the site is safe for an AI agent to act on, and the security program, which fronts both. Same rule, pointed at a different question: not "is it safe to act here," but "does the design already match five principles a coalition wrote down for what AI should look like."

Read is-this-safe Read the declaration Connect an agent

Vealth · provable ecological work. Home Find Work Is this safe Security program Agents Connect your AI Privacy Terms