Proof of Authority for agentic Regen: automate on bounded, revocable, on-chain grants — never a master key

[brawlaphant]

Proof of Authority for Regen automation

As Regen leans further into automation and agents — keepers, agent voters, and the M013 burn keeper in the thread next door — one question decides whether that automation is safe:

What key does the automation run on?

A keeper or agent running on a treasury key or master key is a standing risk: unbounded power, no expiry, and no public accountability.

There is a better answer already available on-chain, using only stock Cosmos modules, verified live on regen-1 with SDK v0.53.4:

Grant automation a bounded, revocable, on-chain-provable authority — a “Proof of Authority” — instead of handing it a raw key.

The pattern

Using stock x/authz + x/feegrant:

• SendAuthorization(spend_limit, allow_list=[sink])
  An executor can move REGEN only up to a defined cap, and only to a specific address, such as a burn sink. Nowhere else.

• GenericAuthorization(/cosmos.gov.v1.MsgVote)
  Vote-only delegation. The executor can vote on the grantor’s behalf and nothing else.

• feegrant BasicAllowance(spend_limit)
  The grantor pays the executor’s gas, capped. The executor holds no funds at all.

• Expiry + revocation + public queryability
  Every grant carries an expiry, is instantly revocable through MsgRevoke / MsgRevokeAllowance, and is queryable by anyone at:

/cosmos/authz/v1beta1/grants

The grant itself becomes the public proof of exactly what the automation is allowed to do.

Why it matters

This is the safety layer that lets Regen adopt automation and agents without trusting raw keys.

Every automated action becomes:

• bounded

• dated

• provable

• revocable

• killable by governance

That is what makes the M013 burn keeper safe to run.

It also generalizes. Any agent acting on-chain for a DAO, treasury, working group, or community process should run on a scoped grant, not a raw key.

x/authz even enforces the correct posture: granter ≠ grantee.

The key that decides stays separate from the key that acts.

No chain change

This is not a request for a new module or chain upgrade.

It is a pattern and reference implementation using modules the chain already has.

EcoWealth has built and audited the flow:

• scoped burn authority

• vote authority

• feegrant support

• EIP-712 manifest for the Base side

Happy to share the code and help standardize it.

The movement

Let’s make bounded, provable, revocable authority the norm for automation and agents on Regen.

As the ecosystem gets more agentic, it should become safer at the same time — not riskier.

Feedback welcome.

[brawlaphant]

The numbers sharpen the case for Plan A

I ran the actual numbers, and they sharpen the case for Plan A — especially one design choice in particular.

Where the network is today

REGEN is around $0.00105, with a market cap near $155k and 24h volume under $100.

The whole validator set is securing the chain with roughly $96k of bonded stake.

The community pool is effectively empty.

That means under zero emissions, validator income goes to roughly $0. A validator seat becomes pure cost.

What that cost actually is

A lean validator node costs about $40–100/month.

A more redundant setup costs around $150–300/month.

So funding a small active set of roughly 7 validators is on the order of $8k–25k/year total.

That is small in absolute terms, but there is currently no meaningful on-chain source for it: the community pool is empty, and zero emissions means no validator income.

The honest design point

You cannot have all three at once:

1. decentralized validator operations

2. zero emissions

3. unfunded validators

Pick two.

The coherent answer is that the Foundation, or another designated entity, curates the active authority set and covers its cost through a modest, explicit funding line.

That means roughly $8k–25k/year to keep the validator set reliable, instead of expecting validators to keep absorbing losses on goodwill.

Zero emissions only holds if the small validator set is funded. Otherwise, it accelerates the same attrition we are already seeing.

EcoWealth’s position

We are in either way.

Our marginal cost to run a Regen validator is near zero because we already operate reliable infrastructure. We are glad to be one of the roughly 7 validators, whether or not there is a funding line.

We will also stand up the on-chain proposals needed to execute this directly:

• set emissions to 0

• set max_validators to 7

This should execute in days, not quarters.

We are not going anywhere.

[Gregory_Regen (Gregory Landua)]

Brandon — this is exactly the right security instinct, and thank you for making it concrete. “The key that decides stays separate from the key that acts” is a powerful minimum rule for agentic infrastructure.

I want to add an institutional layer to the technical one. A bounded x/authz grant is necessary, but it is not by itself accountability. It proves what an agent may do; we also need durable public context for why the grant exists, who authorized it, what evidence the agent must produce, who reviews the result, how a decision can be challenged, and what happens when the work is poor even if it stayed technically within scope. Put differently: capability security constrains action; institutions make action answerable over time.

I would love to see a standard public “authority record” accompany these grants:

• a human-readable mandate linked to the on-chain authorization;
• explicit scope, budget or rate limits, expiry, and revocation path;
• required receipts and evidence for every material action;
• a named human or governance review boundary;
• challenge and remediation rules; and
• a cumulative performance history that future delegators can inspect.

That would connect this pattern directly to the agentic-tokenomics repository, where bounded automation, governance layers, contribution records, and reputation can be developed as one system. KOI can keep the rationale and evidence discoverable without pretending that retrieval itself is governance.

It is also worth keeping two uses of “Proof of Authority” distinct. This thread is about scoped authority for agents using modules already on Regen. The older consensus PoA RFC concerns who validates the chain. They share a concern with accountable authority, but progress on one does not mean the other has been adopted. Regen remains a sovereign PoS chain today, while PoA/bioregional-validator designs remain a governance direction to evaluate carefully.

This is where my federated-work-harness review also points: local autonomy, shared interfaces, machine-enforced limits where rules are binary, and explicit human gates where judgment is real.

Thank you again for leading with both urgency and revocability. If we make the receipts and review history as first-class as the grant, Regen could offer a credible model for agentic institutions rather than merely agentic transactions.