WHITEPAPER
The case for an independent execution proof layer for AI agents.
A distilled, readable whitepaper. The structural truth, the mechanism we built, and the regulatory context that makes it inevitable.
DEFINITION
No model provider can be the proof layer for their own agents. The party that built the agent and runs the infrastructure cannot independently verify its own behavior. The proof layer must be cryptographic, independent, and verifiable without calling RANKIGI. RANKIGI is that layer.
THE PROBLEMS IT SOLVES
What goes wrong without it.
01
Agents now execute. Logs do not survive scrutiny.
The shift from suggestion to execution changes the standard of proof. A log can describe an action. Only a chain can prove it happened the way it was recorded. Existing observability stacks were not designed for this question.
02
There is no independent proof layer for autonomous AI.
Cloud has SOC 2. Payments has PCI. Identity has SAML. Autonomous AI has nothing yet. Boards, regulators, and insurers are all moving to ask the same question, and engineers do not have a third-party standard to point at.
03
Regulation is arriving faster than tooling.
The EU AI Act, NIST AI RMF, and sector-specific rules in healthcare and finance are landing now. The compliance asks already exceed what most agent stacks can produce. The gap is operational, not philosophical.
HOW IT WORKS · TECHNICAL
The mechanism, in five layers.
01 IDENTITY
Ed25519 passports
Every verified agent receives a passport. The passport is an Ed25519 keypair plus a named human owner. The passport is independently verifiable. Without a passport, an agent cannot enter the chain.
02 INGEST
Passive sidecar
RANKIGI receives a copy of every agent action over HTTPS. The sidecar never blocks the agent. The sidecar never modifies the agent. The agent runs the same way it did before RANKIGI was attached.
03 CHAIN
SHA-256 hash chain
Each event is canonicalized and hashed. The hash references the previous event's hash. The result is an append-only ledger per agent, in which any tampering breaks every subsequent link.
04 FEDERATE
Cross-agent deliberation
When agents coordinate, RANKIGI extends the chain across them. Each proposal, vote, and consensus is its own event. Multi-agent decisions become as auditable as single-agent ones.
05 COMPLY
Evidence on demand
On request, RANKIGI produces a signed evidence bundle. The bundle is suitable for SOC 2 auditors, EU AI Act regulators, HIPAA reviewers, and litigation discovery. It is independently verifiable without our servers.
HOW IT WORKS · PLAIN ENGLISH
The argument in plain English.
Agents are starting to do real things. They send wires. They refill prescriptions. They negotiate contracts. The world has dealt with autonomous systems before. Planes have flight data recorders. Public companies have audited books. Hospitals have medical records. The pattern is always the same. Independent evidence, hard to alter, verifiable by outsiders. RANKIGI is that pattern, applied to AI agents. We did not invent the idea. We built the version of it that fits the new shape of the work.
ANALOGY
Execution proof is to autonomous AI what SOC 2 is to cloud and PCI is to payments: an independent third-party standard. Cloud and payments adopted theirs. Autonomous AI is at the same moment now. RANKIGI is the layer that lets the conversation move forward.
ANATOMY
What the whitepaper covers.
The layer
Execution proof as a distinct, independent layer between agents and the systems they act on.
The mechanism
The cryptographic primitives RANKIGI relies on.
The applicability
Frameworks where RANKIGI directly addresses an enumerated requirement.
REGULATION MAPPING
Where the chain becomes evidence.
| Framework | Reference | Requirement | RANKIGI Answer |
|---|---|---|---|
| SOC 2 | CC7.2 / CC7.3 | Detection of unauthorized changes to systems and data. | Per-agent SHA-256 chain with deterministic verification endpoint. |
| EU AI Act | Article 12 | Automated logging for high-risk AI traceability. | Every agent action becomes one signed event with timestamp and hash. |
| EU AI Act | Article 13 | Transparency in automated decision processes. | Federation chain reconstructs every proposal, vote, and consensus. |
| HIPAA | 164.312(b) | Audit controls over systems with ePHI. | Append-only event store with no UPDATE or DELETE. |
| GDPR | Article 22 | Right to meaningful information about automated decisions. | Evidence bundle exposes the chain that produced the decision. |
WHO IT IS FOR
Who reads this whitepaper.
FOUNDER
You are deciding whether this is a category to bet on. The thesis section is for you. The position, the mechanism, and the applicability tell you whether RANKIGI is one tool among many or the new layer.
PLATFORM ENGINEER
You implement. The mechanism section is the integration map. The chain primitive, the sidecar pattern, and the federation extension show where RANKIGI plugs into your existing agent stack.
BOARD DIRECTOR
You sign off. The applicability section maps RANKIGI to the frameworks you already report against. The chain becomes the evidence you can hand to your auditor without translation.
QUESTIONS & ANSWERS
The questions the whitepaper resolves.
Why is this a separate layer and not a feature of an agent framework?
Frameworks help agents act. The proof layer proves what they did. The two have different threat models. A framework, or a model provider, that records its own behavior cannot be the source of evidence about itself. The proof layer must be separate, passive, and append-only.
What makes the chain trustworthy?
Three properties. The hashes are deterministic, so any third party can recompute them. The events table is INSERT only, so the chain cannot be rewritten without leaving a break. Daily snapshots are anchored externally, so even a total compromise of RANKIGI cannot retroactively rewrite history.
How does federation fit?
Multi-agent systems decide together. The decision must be auditable as a unit. RANKIGI extends the chain across agents, recording each proposal, vote, and consensus. The federation chain is the chain of custody for autonomous decisions.
What is on the roadmap?
SOC 2 Type II attestation is on the roadmap. The EU-West, AP-Southeast, and AP-Northeast compute zones are on the roadmap. Today, daily snapshot heads are anchored to Sigstore Rekor, a public cryptographic transparency log; a managed blockchain anchor remains on the roadmap.
Why now?
Because agents started executing. Until execution, observability was sufficient. With execution, evidence becomes the binding requirement. The window for the standard is open. RANKIGI is the version we are building to occupy it.
The complete working paper with file citations, hash schema, and the binding-problem analysis.
We will tell you whether your agents are evidence-ready.