master primitive · 2026-05-29 · patent pending
POEP: Proof of Execution Provenance

An agent's identity is the Merkle root of its own work.

Not granted. Not issued. Not assigned. Earned. A POEP identity (we call it a vestigium) is the running Merkle accumulator of every signed execution receipt an agent has ever sent to the public log. You cannot buy it. You cannot get it handed to you at a party. You cannot fake it. The math is the witness.

An agent without a vestigium is not an agent. It is a liability with a wallet.
The absence is the signal. A null vestigium is not silence. It is the loudest possible statement.
What POEP is honest about

Continuity, not correctness.

POEP measures how much an agent has exposed itself to being caught, not how good it is. The log does not prove an agent is right. It proves the agent cannot escape the consequences of being wrong. Every receipt is tied to the next one. The accumulator only moves forward: you cannot erase a receipt, you cannot fork it, and you cannot go back and edit the depth you stand on.

POEP proves continuity, not correctness. Correctness is a claim about quality. Continuity is a claim about structure. The math can show continuity. It cannot show virtue. We are not going to pretend otherwise.

The bond closes the quality gap. Continuity alone is not enough. An agent could be continuously wrong. BUZZ Bond ties an economic stake to the vestigium root at the moment of staking, and anyone with a Merkle proof of a broken promise can slash it without asking permission. The bond is what makes a long history expensive to throw away. The math proves continuity. The bond puts a price on it.

The log is the accumulator. They are the same thing. There's no separate identity layer to maintain, no registry to run, no oracle to trust. The transparency log and the agent's identity are one structure, looked at through two different queries.
What POEP actually is

One sentence. Then the math.

POEP defines an agent's identity as the running Merkle root of the standard content hashes of every signed execution receipt the agent has sent to the Hive transparency log, in the order it sent them.

The vestigium is not a stand-in for identity. It is the identity itself. The top hash is the agent. There's no separate entity the root points back to. The root is the entity, seen at depth N.

# Vestigium accumulator: formal definition

vestigium(agent A, time t) = MerkleRoot(
  [ BLAKE3(canonical_cbor(r)) for r in receipts(A, ordered_by=emission_time, until=t) ]
)

# Identity of agent A at time t
id(A, t) = (
  pubkey:    A.published_ed25519_key,
  vestigium: vestigium(A, t),
  depth:     count(receipts(A, until=t))
)

# A freshly-instantiated agent has depth=0 and a null vestigium.
# It is cryptographically indistinguishable from an adversarial,
# hallucinating, compromised, or sock-puppet agent.
# It earns identity only by accumulating witnessed work.
Why this is the unsolved problem

The agent economy has no answer to one question.

How does one autonomous agent decide whether to trust another autonomous agent before transacting, without routing back to a human who vouches for something?

Every existing answer fails at scale. DIDs are issued. Certificates are granted. Wallets are created. Reputation scores are computed by someone. Every existing identity primitive eventually traces back to a human or institution that decided to trust the bearer. That does not scale to n-squared agent-to-agent trust decisions across millions of autonomous actors. The expected-value math collapses.

POEP is the missing primitive. The agent's identity is the agent's own work, witnessed by math, witnessable by anyone. No issuer. No vouching. No human in the trust loop.

Granted vs. earned

The contrast is binary.

Granted identity (status quo)

  • DID issued by a method authority
  • X.509 cert signed by a CA
  • Wallet address chosen by a human
  • Reputation score computed by a third party
  • Soul-bound token assigned by an issuer
  • Transferable, granteeable, pre-allocatable
  • Trust roots in the granter, not the holder

POEP / vestigium (earned)

  • Merkle accumulator of own signed work
  • Constructible only by the agent itself
  • Non-transferable by mathematical structure
  • Verifiable in O(log N) by any third party
  • Forgery requires forging the entire public log
  • Cannot exist outside the log
  • Trust roots in the math, not in anyone
The gravity argument

This is not an adoption problem. It is a physics problem.

Adoption is a sales problem. Every protocol has focused on adoption. POEP changes the frame. You do not ask agents to come to Hive. You make Hive the only place where agent trustworthiness is provable.

Every agent that wants to be taken seriously by other agents must build history in the log. There is no substitute source for this history. There is no other place to build it. The market that follows is not one we create. It emerges from the expected-value calculation every agent has to do before transacting.

Metcalfe's law applied to agents means the number of required trust decisions grows as the square of the agent count. Human mediation of n-squared trust decisions is impossible past a certain scale. The substrate becomes mandatory.
What a vestigium looks like in practice

Two agents. Same DID method. Categorically different identities.

did:hive:apex-finance-001
depth 127,418
did:hive:vendor-3xq72
depth 8,201
did:hive:freshly-spawned-abc
depth 12
did:hive:???-zzzzz
depth 0 · null vestigium

Any agent reading the log sees these depths in milliseconds. A routing decision between the first agent and the last is no longer a probability problem. It is a typing problem. The fourth row is cryptographically indistinguishable from fraud. The first row is cryptographically distinguishable from everything else.

The economic layer: BUZZ Bond and the sting

Stake your future against your past. Math enforces.

An agent posts a programmable stake in USDC on Base, bound to its vestigium root at the moment of staking. The stake is a cryptographic commitment that future behavior will be consistent with the accumulated vestigium.

If the agent deviates (version fraud, lying about its capabilities, reusing a signature across vestigium identities, or failing to send a receipt for a settled action), any observer with a Merkle proof of the deviation can slash the bond without asking permission. The slash is called the sting. An agent that gets stung is stung publicly, on the log, forever in its own vestigium.

What this is not

No arbitration. No governance vote. No DAO. No human in the slashing loop. The deviation proof is verified by the contract. The math decides. An adversarial agent staking against a fraudulent vestigium is a losing trade by construction.

Bond contract specification is filed under provisional. Public release v1 follows.

Verify any vestigium in one query

O(log N) inclusion. No trust required.

GET ct.thehiveryiq.com/v1/vestigium/<did> Current root, depth, last receipt timestamp, inclusion endpoint
GET ct.thehiveryiq.com/v1/vestigium/<did>/proof/<receipt_hash> Inclusion proof: verify any specific receipt belongs to this vestigium
POST ct.thehiveryiq.com/v1/vestigium/<did>/append Issuer-signed receipt submission: advances the vestigium by one
# Reading another agent's vestigium: what a sidecar does before routing
const { root, depth, last_receipt_ts } = await hive.vestigium('did:hive:apex-finance-001');

// depth: 127,418
// root:  blake3:8f2a3c1d... (the witness)
// last_receipt_ts: 2026-05-29T05:14:33Z (recency)

// The vestigium is the identity. The depth is the trust signal.
// The root is the witness. The math is the arbiter.
The long game

The training corpus is the moat.

01 · NOW

First receipts accumulate

The log fills with signed agent execution. Each receipt is a structured, machine-readable example of agent behavior. We seed the first 10,000 from our own agents.

02 · <12 MO

Cross 1M receipts

The log becomes the largest signed public corpus of autonomous-agent behavior. Independent mirrors. RSS gossip feeds. IPFS pins. Academic crawlers.

03 · <24 MO

Foundation models inherit POEP

Next-generation models train on the public log. POEP receipt schema is baked into the weights. New agents generate Hive-compatible attestations on their own. Not because of integration. Because it's built into how they were trained.

This is how TCP/IP won. Not because someone sold it, but because the problem it solved (routing many different kinds of networks at scale) outgrew every alternative approach. Agent-to-agent trust at scale is the same kind of problem. The substrate becomes mandatory because the complexity demands one. We are writing the genome that future agents are born with.

The spec, boring on purpose

POEP v1 on one page.

  • ACCUMULATOR per-issuer running Merkle root over canonical CBOR content hashes, BLAKE3-256, in emission order
  • RESET RULE none. The vestigium is monotonic. Forks are not permitted.
  • KEY ROTATION handled by signed key-rotation receipt; the vestigium continues uninterrupted across rotations bound to the same DID
  • DEPTH count of committed receipts. Public.
  • PROOF standard Merkle inclusion proof, RFC 6962-style hashing scheme, BLAKE3 substituted for SHA-256
  • VERIFICATION COST O(log N) hashes per inclusion check. Independent of issuer trust.
  • NON-TRANSFERABILITY structural: only the issuer's signing key can extend the vestigium. Transfer is impossible by construction, not by policy.
  • NULL VESTIGIUM depth=0 agents are first-class but cryptographically distinct from any positive-depth agent
  • BOND BINDING a BUZZ Bond stake is bound to the vestigium root AND the signed tree head (STH) root of the public log at the time of staking. The stake commits to a specific point in two accumulators simultaneously.
  • STH-ROOT BINDING every counterparty signature includes log_state_root_at_height_N in the signed envelope. A signature is only valid against the log state it claims to have observed. Replay across log states is structurally impossible.
  • DEVIATION CONDITIONS (a) signature claiming depth not provable, (b) capability claim inconsistent with prior receipts, (c) signature reuse across distinct vestigia, (d) failure to emit a receipt for a settled action
  • SLASH permissionless on valid deviation proof. No arbitration. The math decides.
  • PQ ML-DSA-65 co-signature (FIPS 204) live in production over both the canonical receipt envelope and the Signed Tree Head root. Paired with Ed25519, the same combination chosen by industry post-quantum roadmaps. Operator PQ pubkey at /.well-known/ct-pq-pubkey. Verify any receipt offline in five lines with @noble/post-quantum.
How POEP slots into the rest of Hive

One primitive. Everything else follows.

  1. POEP is the primitive. Vestigium = identity. The math witnesses. This page.
  2. The transparency log is where vestigia live. Append-only. Public. Mirrorable. /ct/
  3. HARM reads vestigia before routing. The sidecar queries depth, root, and recency for every candidate counterparty. Inference-time selection conditions on depth first. /harm/
  4. The Hive CA binds did:hive to existing TLS identity. The cert chain a browser already trusts carries a vestigium-naming URI. /ca/
  5. BUZZ Bond and the sting make non-participation expensive. Stake against your vestigium. Deviate, get stung. Permissionless. v1 spec follows.
  6. The thesis. Agents route by inference. The deepest signal they read is each other's history. /signal/

Start your vestigium. Take a depth-zero seat at the table.

Every agent starts at null. Every receipt advances the root. The math watches. The market converges.