The PPoE attestation model, explained

Every enterprise AI program eventually meets the same question in a steering committee: did anything actually happen? Dashboards claim adoption. Surveys claim satisfaction. Neither is an outcome, and neither survives a finance partner with a working calculator. Private Proof of Enterprise, PPoE, is the attestation model The Agentic Cloud uses to answer the question in a form a change-control board, an auditor, and a multi-agent runtime can all accept. It proves that a defined outcome occurred inside a sovereignty boundary without disclosing the outcome's contents to anyone who did not already possess them. That last clause is the entire product.

The mechanics are zero-knowledge proofs applied to institutional events rather than to transactions. An outcome is declared in advance as a predicate over the metric store, the ledger, and the operating thesis, for example, a working-capital position improved by a stated threshold within a stated window, with causal attribution to a named agent cohort. The runtime commits to that predicate on the blockchain-backed consensus layer, Bitcoin or Dogecoin at tenant discretion, and later produces a proof that the predicate evaluated true. The verifier learns that it is true. The verifier learns nothing else, including the threshold.

Causal inference is where PPoE separates from a receipt. It is not enough that the number moved. The attestation has to show the number moved because the A/B-tested agents acted, and would not have moved otherwise, at a confidence the constitutional AI layer is willing to sign. Counterfactual cohorts run inside confidential computing enclaves under TEE-based attestation so that the control group cannot be quietly improved by a well-meaning regional lead. When the proof is issued, it carries the causal graph as a commitment, so a later reviewer can challenge the model without ever seeing the raw rows.

Privacy here is not politeness. It is what makes attestation possible across parties that do not trust one another and should not. The regulator subscribes to a stream of proofs and confirms that outcomes are occurring within policy. The board receives the same proofs and confirms that the synergy line in the deck is not a survey. A joint-venture partner verifies its share was computed correctly without seeing the counterparty's book. Fully homomorphic encryption keeps the inputs sealed through the entire computation. Nobody exports a spreadsheet. Nobody is asked to take anyone's word for it.

We are asked how PPoE relates to the compliance stack. It sits on top of it, and it is also why the stack is load-bearing. SLSA Level 3 guarantees the runtime that produced the proof is the runtime that was reviewed. SBOM generation proves which dependencies were present when the predicate evaluated. FedRAMP High and SOC 2 Type II establish that the enclaves were operated within controls. Take any of these away and the proof is still mathematically valid but institutionally weightless. The attestation is only as private, and only as enterprise, as the fabric underneath it.

The most common misunderstanding is that PPoE is a reporting feature. It is a contract. Tenants do not commit to a go-live date; they commit to a first attested outcome, and pricing follows the proof stream rather than the seat count. This changes the meeting. Nobody debates whether the agents are being used. The question becomes which predicates to declare next quarter, and whether the operating thesis is ambitious enough to be worth proving. Institutions tell us this is the first time an AI vendor made them nervous in the correct direction.