Who could ask for funding? Who would decide? Who would check the work? Here is one proposed answer, with a budget you can follow.
A proposal to test, not rules already in force. No constitution has been adopted. There are no verified constituents, binding elections, or common treasury operating through this site. AI would help people do the work; people would hold the authority.
One year. One $10 million budget.
Imagine a future institution with 25,000 verified constituents and a lawful $10 million common treasury. A constitution sets spending limits before anyone picks projects. Donors cannot buy votes or control the budget.
Set the budget before picking winners
Proposal v0.1
Use five defined spending envelopes in one annual cycle. Changes between envelopes need their own public reasons, review, and lawful approval.
Purpose
Keep urgent or popular projects from quietly draining reserves or core services.
Tradeoff
Fixed envelopes can block a useful project. Administration and oversight cost money that could otherwise fund work.
Open question
What reserve level, operating cost, and rules for moving money would people accept?
Illustrative annual envelopes — total $10,000,000
Envelope
Amount
Use
Operations
$1,000,000
Staff, access, administration, and independent oversight.
Reserves
$2,000,000
Held for risks and future needs; no routine project spending.
Research
$2,000,000
Test methods, publish evidence, and check results.
Grants
$2,000,000
Support lawful work by outside groups.
Public-benefit projects
$3,000,000
Build useful tools and services.
Total
$10,000,000
$7 million for projects; $3 million for operations and reserves.
Five proposals compete for $7 million.
The project requests total $7.5 million. Only $7 million is available in their envelopes. The public-benefit requests alone exceed their $3 million limit by $500,000. A popular idea still needs a workable budget.
Make every request easy to inspect
Proposal v0.1
Use the same public form for every project: sponsor, beneficiaries, full cost, milestones, risks, conflicts, and success measures. Publish corrections and earlier versions.
Purpose
Make competing requests comparable and expose who stands to benefit.
Tradeoff
Small groups may struggle with forms. Help with applications must be available without favoring insiders.
Open question
How can we require useful evidence without making the process too costly for new or local groups?
Fictional requests and later example allocations
Project
Envelope
Requested
Later allocation
A. Independent AI audits
Research
$1,200,000
$1,200,000
B. Language and access research
Research
$800,000
$800,000
C. Local skills grants
Grants
$2,000,000
$2,000,000
D. Public resource tools
Public-benefit projects
$2,000,000
$2,000,000
E. Plain-language public guides
Public-benefit projects
$1,500,000
$1,000,000 after revision
Project total
Across three envelopes
$7,500,000
$7,000,000
A. Independent AI audits: inspect the request
Sponsor
An independent research team
Beneficiaries
People affected by proposed AI tools
Cost
$1,200,000; example allocation: $1,200,000.
Milestones
Publish test plan; complete outside tests; publish findings.
Risks and conflicts
Tests may miss harm. The team must disclose model-provider funding.
Success measures
Coverage of agreed tests; outside replication; published failures.
B. Language and access research: inspect the request
Sponsor
A language-access research group
Beneficiaries
People facing language or access barriers
Cost
$800,000; example allocation: $800,000.
Milestones
Recruit varied testers; test access; publish fixes and limits.
Risks and conflicts
A sample may miss whole communities. Disclose ties to translation vendors.
Success measures
Task completion across languages and access needs; unresolved gaps.
C. Local skills grants: inspect the request
Fictional project request. No grants are available through this site.
Sponsor
A coalition of local learning groups
Beneficiaries
Adults seeking practical skills
Cost
$2,000,000; example allocation: $2,000,000.
Milestones
Publish grant rules; select recipients; report completed programs.
Risks and conflicts
Well-connected groups may win. Reviewers must disclose ties and step aside.
Success measures
Fair access to grants; completion; independent feedback from learners.
D. Public resource tools: inspect the request
Sponsor
A public-interest software team
Beneficiaries
People trying to find useful local services
Cost
$2,000,000; example allocation: $2,000,000.
Milestones
Test a prototype; complete access review; publish a limited release.
Risks and conflicts
Costs may grow; data may be wrong. Disclose suppliers and data-sharing ties.
Success measures
Successful service searches; accessibility; upkeep cost; reported errors.
E. Plain-language public guides: inspect the request
Sponsor
A community publishing group
Beneficiaries
People who need clear explanations of services and rights
Cost
$1,500,000; example allocation: $1,000,000 after revision.
Milestones
Publish smaller scope; test drafts with readers; release reviewed guides.
Risks and conflicts
Guides may be wrong or go stale. Disclose sponsor and publisher interests.
Success measures
Reader understanding; correction time; cost per completed guide.
Check the plans before asking for priorities.
Independent review, with evidence people can challenge
Proposal v0.1
Separate reviewers check legal authority, feasibility, full cost, and risk. AI can compare the plans and show different scenarios, with links to source evidence. Human reviewers sign the findings and hear challenges.
Purpose
Stop an attractive pitch from hiding an unlawful activity, weak evidence, or missing costs.
Tradeoff
Review takes time and money. Reviewers can be biased, and useful evidence may contain private information.
Open question
Who selects and replaces reviewers, and what can be published without exposing private records?
Reviewers publish their conflicts and step aside where needed. A second team tests major claims. AI summaries show disagreements and missing evidence. AI cannot certify its own comparison, choose a winner, or treat a prediction as a fact.
Reject: a request that is outside lawful powers or violates protected rights. Publish reasons and an appeal path.
Revise: a useful plan with fixable costs, access gaps, or conflicts. Publish the changed scope and review it again.
Defer: a plan that needs more evidence or cannot fit this cycle. Give a reason, a review date, and no promise of future funding.
Listen widely, then require a recorded decision
Proposal v0.1
Give each verified constituent the same 100 nonpurchasable, nontransferable priority credits for this example cycle. A stratified citizen panel reviews the evidence and publishes majority and minority reports. The Assembly approves a lawful portfolio in a public recorded vote.
Purpose
Let people shape priorities while checking costs, rights, and the interests of less vocal groups.
Tradeoff
Priority credits show preferences, not consent to a binding spend. Participation can be unequal. A panel cannot speak for everyone, and the Assembly may choose differently.
Open question
How should panel selection, access support, participation gaps, and departures from public priorities be judged?
The credits are a proposed way to rank needs; they are not money and cannot be sold or given away. The Assembly must explain any departure from the panel reports or public priorities. Neither a donor nor an AI system can authorize spending.
In this fictional result, the panel supports a smaller version of project E. Its minority report asks for stronger local access testing on C and D. The Assembly accepts those conditions and approves A–D at their requested costs, plus E at $1 million. The final project total is $7 million. Operations stay at $1 million and reserves at $2 million.
Approval is the start of accountability.
Humans control custody and each release
Proposal v0.1
Accountable humans and lawful entities sign contracts and execute approved work. Treasury custody is separate from project selection. Each payment needs multiple authorized people, documented checks, and verified milestones.
Purpose
Prevent one person, supplier, or AI system from moving funds without review.
Tradeoff
More checks slow payment. Signers can collude, and a completed milestone does not prove a project helped people.
Open question
How many independent signers, what custody rules, and what stop-payment powers would be workable and lawful?
For project E, the illustrative $1 million cap is released in three stages: $200,000 after the revised scope and contract are approved, $300,000 after tested drafts pass independent review, and $500,000 after accepted delivery. These releases total $1 million. Failed checks pause the next release; a change in scope returns for approval.
AI may flag missing receipts or compare reported progress with a contract. It cannot hold signing authority or autonomously transfer funds. A human review records why a payment was made, delayed, or refused.
Publish outcomes and make challenge possible
Proposal v0.1
Publish contracts, spending, milestones, changes, and measured outcomes, with narrow justified privacy redactions. Provide formal challenges, protected whistleblowing channels, an ombuds route, and independent audit.
Purpose
Give people a way to find waste, favoritism, broken promises, and poor results after approval.
Tradeoff
Reports create work and can miss real harm. Public detail can expose private people; blanket secrecy can hide abuse.
Open question
What protections and remedies make it safe to report wrongdoing, and who checks whether those remedies work?
Auditors must be separate from the people managing the project. A challenge may lead to correction, a spending pause, a new review, or lawful recovery of funds. The parties get reasons and an appeal. Contracts may limit what can be reversed; money already spent cannot simply be voted back into the treasury.
A fair process can still fund a poor project. Compare results with the original measures, report failures, and change the next cycle based on evidence.
A useful precedent, not a safety guarantee.
Decidim’s official documentation describes public results linked to proposals, with milestones and a history of changes. That is a useful reporting pattern to study. It does not establish that this proposed treasury is safe, and this site does not run Decidim or a financial system.