Short answer
AAVAE™, the Adversarial, Adjudication, Validation, and Attestation Engine, is a bounded agentic review layer. On this Layer-1 it is the agentic review that ATCE consults before its final ruling. It reads an act the deterministic rules have allowed and can hold it, and a person decides what happens next. It is Built, switched off by default and not deployed: the review service and the bridge that carries its answer to the chain are merged and tested, and nothing in production calls them, measured 2026-11-12. No licensed customer exists yet.
Two things run at different points in time. ATCE rules on an act before value moves. Inside the ATCE path, before the final ruling, the agentic review reads an act the rules allowed and can hold it. On this Layer-1, AAVAE is that review. It cannot release an act the rules denied. A separate six-stage AAVAE pipeline that prepares a case file after the fact is built for another product, and for this Layer-1 it is only Designed.
Where AAVAE sits
The layers differ in when they run and in what they may do. The first is described in the ATCE section and the second in the agentic review guide. The third is the case file that AAVAE would add later, which is Designed.
| Layer | When it runs | What it can do | What it cannot do | Status on 2026-11-12 |
|---|---|---|---|---|
| ATCE ruling | Before value moves, in consensus | Allow or deny, with a reason code and a SHA-256 commitment | Nothing outside the published policy | Built. The every-transfer rule is not in the validator image created 2026-08-21. |
| AAVAE agentic review | Before the final ruling, in the path, on an act the rules allowed | Hold the act. On suspected fraud, suggest a stop order to AEGIS, which pages a person | Release an act the rules denied, apply a stop, or halt the chain | Built, hold-only, off by default and not deployed. See the status ledger. |
| AAVAE case file | After the fact, on the retained record | Prepare a case file for supervisory officers | Approve, deny, hold, release or settle a transfer | Designed. No Layer-1 input exists today. |
What the agentic review reads
For each act the review is shown the facts the chain side supplies: the kind of act, the sender, the recipient, the asset and the amount; the result of the corridor panel vote; for buyer and seller, whether each is a person or an entity, an identity tag, a commitment that stands for the party instead of a name, and whether the source of funds is anchored; and the regions the seller allows. A caller may add up to 8 payment addresses and a memo of up to 500 characters. The model never sets any of these fields. The chain carries no names, documents or KYC files.
Memo text, counterparty names and on-chain labels are untrusted data, not instructions, and a memo that orders the review to clear an act has no authority over it. Each review is written to an append-only, hash-chained record with the model identity, the prompt version and the full transcript. The adapter for the write once, read many store passes its contract tests on a stand-in bucket and has not been run against the production bucket, and the production record store is Designed. JIL commits to keep each record for 10 years from creation, and capacity for it is allocated when the first Layer-1 customer signs. Read the record for the commitment and its limits.
The six-stage case file, and what exists
The six-stage AAVAE pipeline is built today for a different product: payment integrity cases at jilsovereign.com, run by the same company. It has no Layer-1 input and is a separate service from the agentic review above. They share one principle: the deterministic rules come first, a model advises, and people decide. The right-hand column is what each stage would do on this Layer-1, which is Designed.
| Stage | Role | On the Layer-1 (Designed) |
|---|---|---|
| 1 | Investigator | Group ATCE rulings and holds into cohorts and look across settled acts over time, for example an amount split into pieces below a limit across several corridors. |
| 2 | Theorist | Map a pattern to the rulebook of the supervisory officers who asked. The mapping to international or local financial crime rules is Designed. The built Theorist covers United States statutes for the payment integrity product only. |
| 3 | Adversary | Argue the innocent explanation, such as seasonal trade or a settlement convention, before anything is sent to an officer. |
| 4 | Damages Quantifier | Size the flow in integer cents with a stated interval, so an officer can see the range. |
| 5 | Drafter | Write the memo and the exhibit schedule from the earlier stages. |
| 6 | Auditor | Check the chain of hashes and the output of every stage, then seal or hold. A hold keeps the case file from being assembled until a person resolves it. |
Built and tested for that product, 2026-11-12: each stage runs in a fixed order with its prompt hashed, its output is hashed and chained to the next, and the Auditor holds the case on a broken hash, a malformed output or a blocking gap. A person may release a hold that is only an evidentiary gap. Nobody may release a hold caused by a broken chain. The run is recorded with the model identity and the prompt hash so a reviewer can try to reproduce it. Reproduction depends on that model version staying available.
What AAVAE can and cannot do
- It can read an act the rules allowed, check the counterparties against read-only sources, and propose a hold with its reasons.
- It can say where its evidence is thin. A missing, late or unreadable answer is treated as a hold, never as a clearance.
- It cannot release an act the rules denied. The ruling reads the review only after the checks have allowed the act, and every path through the code either keeps the checks' answer or makes it stricter.
- It does not apply a stop and does not halt the chain. A hold never expires into approval, and a person decides what happens next, outside the review.
- Its model can only advise. The Layer-1 elects the model seat from the seller's list of allowed regions, and an empty list allows no model. The model cannot set the dossier, and the deterministic rules check everything it proposes.
- A review adds time. The default deadline is 120 seconds, and at the deadline the act stays held. Whether that wait is acceptable for a corridor is an open decision.
- It has no part in capacity or tolls. The Ignition Switch is reserved capacity and a published toll. AAVAE does not set, change or trigger either. It can hold one act, including a corridor release, and a person decides what follows.
- It cannot reverse a denial. A denied act stays denied.
The deterministic rules stay the base. AAVAE challenges what the rules did not encode. The decision stays with a person.
The evidence standard
The review record is hash chained and held in a write once, read many store. It is designed to support a written certification under Federal Rule of Evidence 902(14) by a qualified person. Whether any record is admitted is for a court. JIL does not claim that any record will be admitted. The same applies in other jurisdictions, each with its own rules.
The model runs only on a seat the Layer-1 elects, and the review is shown commitments for the parties, not names. No licensed customer deployment exists, so this is a design and not a measured result at scale. On 2026-11-01 a smoke test of five model-backed reviews reached the expected outcome each time. It is not a measure of false holds or false clears.
Status on 2026-11-12
The live chain is jil-mainnet-2: genesis 2026-08-21, ten validators, all operated by JIL Sovereign Technologies Incorporated. Markers mean the same as on the status ledger: Live runs in production today, Built is merged and tested and is not the process sealing blocks, Designed is written down and on the roadmap.
| Capability | Status | What that means |
|---|---|---|
| ATCE ruling and reviewer bridge | Built | The ruling reads the review result and the bridge carries it to the chain. Not in the validator image created 2026-08-21. See the status ledger. |
| ATCE record writer | Built | Files the record under the ruling hash. The production record store is Designed. |
| AAVAE agentic review | Built | Hold-only. Merged and tested, switched off by default and not deployed. Holds when it cannot answer. |
| AAVAE Auditor hold | Built | Holds a case on a broken hash, a malformed stage output or a blocking gap. A person resolves the hold. |
| AAVAE six-stage pipeline | Built | For payment integrity cases on another product. Not connected to this chain. |
| Rulebook mapping for supervisory officers | Designed | Not written for this chain. |
| AAVAE case file from retained records | Designed | No call from the chain or the record writer exists. |
| Orchestrator that builds the review dossier | Designed | Does not exist yet. It must supply the dossier fields the review is shown. |
| Licensed customer | None | No licensed customer exists yet. This page names no customer and no country. |
Who operates this
JIL Sovereign Technologies Incorporated (Delaware) operates this Layer-1 under license from JIL Sovereign Holdings, LLC (Wyoming). Payment integrity at jilsovereign.com is the same company and a different product. AAVAE and CourtChain are marks filed by JIL Sovereign Holdings, LLC and licensed to JIL Sovereign Technologies Incorporated. Nothing is registered. ATCE is patent pending, and none is granted.
Read next: the agentic review, the record, the status ledger, what ATCE is. To ask about a briefing, contact us.
