Hive Private demonstration · prepared for Intercontinental Exchange Verify a receipt →
Live rail
receipts/free receipts/verify r3pv/vector ct log · not public
For Michael Blaugrund · Vice President of Strategic Initiatives

What ICE must prove when its next paradigm arrives.

The next paradigm is agent-to-agent execution against pre-committed policy. Regulators, custodians, counterparties, and issuers will each need to read the same signed event from their own vantage, without renegotiating trust with any intermediary. This page walks a tokenized-equity trade end to end and shows exactly which of those facts an independently checkable receipt could carry, and which it could not settle.

Measured on the production rail, not modeled

0.098 ms
Sign p50
0.136 ms
Verify p50
0.678 ms
Dual-sign envelope p50
3,309 B
ML-DSA-65 signature
~134 ms
Receipt round-trip p50

n=40 sequential calls to /v1/receipt/benchmark?profile=pq from one external client, July 27 2026. Sign p95 0.116 ms. Verify p95 0.180 ms. Round-trip p95 588 ms, which is network, not signing.

Modeler 08 below re-runs this benchmark live against the same endpoint from your own browser. Your numbers will differ by network distance.

The boundary, stated first

Nothing about how ICE runs its markets changes

Before anything else on this page, here is the line. One side of it is ICE and stays ICE. The other side is a signed record of facts that already happened.

What ICE keeps

ICE keeps control of matching, listings, benchmark administration, clearing, custody, data distribution, pricing, orders, and settlement. None of that changes and none of it moves to Hive.

What Hive does

Hive only records selected facts, gate decisions, and acknowledgments as they occur, under one ML-DSA-65 (FIPS 204) plus Ed25519 dual-signed envelope. Hive does not decide, hold funds, move money, route orders, or execute anything.

One trade, end to end

Where a receipt gets written, and where it does not

A tokenized-equity trade crosses four hands. ICE does all the deciding at every one of them. Hive writes down what was decided.

First, what a receipt is

A receipt is a signed record of what the system was told, what it checked, what it decided, and when. Nothing more. It does not contain identity documents, account numbers, positions, or balances. It can be checked by someone who does not trust, and does not have access to, the system that produced it.

Step one
Agent submits
An autonomous buy-side execution agent proposes a tokenized-equity trade against a signed policy envelope.
Step two
ICE and NYSE act
ICE runs its own matching, clearing, and benchmark logic. Its engines, custody, and risk models do not change.
Step three
Counterparties acknowledge
Supervisor, custodian, counterparty, and issuer each acknowledge from their own vantage against their own copy of the receipt.
Step four
Hive preserves the record
Each consequential step leaves a signed receipt any of those parties can verify offline. Typed receipt contracts on proof.thehiveryiq.com are Ed25519 signed. The receipts rail carries an additional ML-DSA-65 slot under NIST FIPS 204.
Click it, run it, read the bytes

Interactive proof workbench: everything on this page runs live.

Ten instruments. Some call the production Hive rail. Some run entirely inside this browser tab and say so plainly. Each one names the endpoint it touches and states what it proves and what it does not decide. Nothing here is an ICE deployment, an ICE design, or an ICE commitment.

One signed receipt, field by field
Illustrative · abbreviated field values
{
  "receipt_id" "r_a2a.trade_1785182190_67e21ff4"
  "purpose" "a2a.tokenized-equity.execute"
  "attested_party" "NYSE Group (listings)"
  "issued_at" "2026-07-27T19:51:09Z"
  "payload_sha256" "0x65c00b8d…d2ced0f2"
  "sig_ed" "0xsig:edb9cba0…d476d900d"
  "sig_mldsa" "RFNBLVNUVUIt…NTAyZWMxNA=="
  "policy_zk"  "scheme" "groth16" 
  "prev_hash" "0xadaffd1c…fbe64c61"
  "ledger_seq" 17520
}
purpose

What the agent was permitted to do, fixed before the effect.

attested_party

The entity the receipt speaks for, verbatim as submitted.

sig_ed

64-byte Ed25519 signature. Verifies in a browser in ~1.6 ms.

sig_mldsa

3,309-byte ML-DSA-65 post-quantum signature, NIST FIPS 204.

policy_zk

Groth16-shape proof that a mandate was followed without revealing it.

Field values above are abbreviated for reading. Modeler 01 below mints a real one against the production key and prints every byte.

The strip above is probed from your browser when this page loads. If a rail is unreachable from where you are sitting, the widget below it says so and falls back to the response captured when this page was built. It never reports a live result it did not get.

Modeler 01

Mint a live signed receipt for any ICE portfolio surface.

Type an entity from your portfolio. Pick a use case. Hive signs a live receipt for that entity against our production key. Verify it. Download it. Show it to your team.

Entity, written verbatim into attested_party
Use case
Idle

          

          

attested_party carries the exact string you type. The receipt records an action attested on this rail, and it is not a statement of any customer relationship, engagement, or endorsement.

Endpoint: POST receipts.thehiveryiq.com/v1/receipt/free and POST /v1/receipts, verified back through GET /v1/receipts/verify/{seq}

Proves that any entity across the ICE portfolio can carry a signed receipt for its own actions, minted against our production key, verifiable offline. Does not decide whether the underlying operation is compliant.

Modeler 02

Offline Verifier, Ed25519 in this browser

Paste a receipt and verify the Ed25519 signature locally. The only network call is one public-key fetch, and it happens before you click.

Public key: fetching…
Receipt JSON
Not verified

        
Endpoint: GET receipts.thehiveryiq.com/v1/receipts/pubkey (once; verify is in-browser)

Proves that Ed25519 verification requires no cooperation from Hive at check time. ML-DSA-65 verification requires a WASM module and is not shown in-browser here.

Modeler 03CT Log Inclusion Proof LookupOpen
Modeler 03

CT Log Inclusion Proof Lookup

Fetch a signed tree head and an audit path for a leaf, then recompute the Merkle root in this tab and compare it to the published root.

Leaf hash
Idle

        
Endpoint: GET ct.thehiveryiq.com/v1/proof/<leaf> · GET ct.thehiveryiq.com/v1/sth
Fallback: GET receipts.thehiveryiq.com/v1/accumulator/receipts/root · /inclusion?key=

Proves that an inclusion proof against an append-only accumulator is checkable without trusting Hive. The CT endpoint is not public yet, so this runs against the live accumulator route on the receipts rail. Does not decide whether the receipt content is correct.

Modeler 04SA™ Heartbeat ReplayOpen
Modeler 04

SA™ Heartbeat Replay

Two participants emit signed heartbeats into a shared Sequence Attestation™ epoch. Drop one and the gap turns red immediately.

NYSE0 beats
ICE Clear Credit0 beats
Epoch merkle root: computing…
Endpoint: In-browser simulation

Proves the shape of SA™ reordering-detection. The heartbeats carry real Ed25519 signatures from an ephemeral key generated in this tab and a real SHA-256 Merkle root. The two participants are simulated: this is not a live NYSE or clearing cluster.

Modeler 05SHOD™ Gate InspectorOpen
Modeler 05

SHOD™ Gate Inspector

Six independent Sub-second Handoff On Demand™ gates evaluate an outbound order as you move the controls. The envelope rewrites itself live.

Quantity 100
Price $214
Recipient DID
Evaluating

        
Endpoint: In-browser simulation of SHOD™ gates

Proves that autonomous-agent behavior can be bounded by six independent programmatic gates before any signature is emitted. Rule values are illustrative.

Modeler 06SpectralZK™ Mandate Proof DemoOpen
Modeler 06

SpectralZK™ Mandate Proof Demo

Pick a mandate and a proposed trade, then generate the proof object shape that would ride in the receipt's attestation slot.

Mandate
Proposed trade
Idle

        
See full spec: how SpectralZK™ actually works

A real SpectralZK™ attestation has four moving parts. One, the mandate is compiled once into an arithmetic circuit, so the predicate is fixed and auditable before any trade exists. Two, at decision time the agent holds the private witness: the position book, the strategy parameters, the model outputs. That witness never leaves the agent's boundary. Three, the prover emits a constant-size proof that the witness satisfies the circuit for this trade, bound to a public statement digest that names the mandate version and the order. Four, the proof rides in the attestation slot of the receipt and is covered by the same ML-DSA-65 plus Ed25519 envelope as the rest of the record, so a supervisor verifies policy compliance and record authenticity in one pass.

What that buys a venue is narrow and specific: a member can demonstrate that an execution agent stayed inside a stated mandate without handing anyone its alpha. What it does not buy is truth about the inputs. If the position book fed to the prover was wrong, the proof is still valid and the conclusion is still wrong.

This widget does not run a prover. Proving requires the circuit, the trusted setup artifacts and the witness, none of which belong in a public web page. What you see is the shape of the object and where it sits in the receipt.

Endpoint: In-browser (illustrative shape, no prover)

Proves the shape of a policy-attestation proof that could ride on the receipt. Not a real prover, a demonstration of the receipt slot into which a Groth16 proof would land.

Modeler 07IEA™ Oracle Origin CheckOpen
Modeler 07

IEA™ Oracle Origin Check

Three preloaded Issuer Event Attestation™ records. Check each one's issuer key against a public identity anchor.

Endpoint: In-browser (issuer records preloaded)

Proves the shape of authoritative-origin verification against a public identity anchor. In production, the anchor is registered on-chain or in a public identity registry.

Modeler 08Proof Latency BenchOpen
Modeler 08

Proof Latency Bench

Ten sequential receipts against the production rail, spaced 200 ms apart, with the round-trip distribution reported as it fills in.

Idle
MeasureSign, serverVerify, serverRound tripn
p50···0
p95···0
p99···0

        
Endpoint: GET receipts.thehiveryiq.com/v1/receipt/benchmark?profile=pq ×10, 200 ms apart

Proves live sign, ML-DSA-65 cosign and verify latency on the production Hive rail, timed server side by the rail and round-tripped from your browser. Capped at ten calls per click because the public faucet allows 100 receipts per IP per day and this page is not here to spend your allowance. Does not prove what latency you would see inside a colocated deployment.

Modeler 09R3Pv™ Lifecycle Vector ViewerOpen
Modeler 09

R3Pv™ Lifecycle Vector Viewer

Five signed lifecycle steps in one proof vector. Hover a node for its signed step, then introduce a defect and watch the weakest boundary surface.


        
Endpoint: In-browser vector · live grouping: POST receipts.thehiveryiq.com/v1/r3pv/vector

Proves the shape of a trade-lifecycle proof vector with weakest-boundary reporting and permitted-next-action. Does not resolve where a real defect originated.

Modeler 10Downloadable Sample PackOpen
Modeler 10

Downloadable Sample Pack

Five well-formed fixtures, zipped in this tab, for your engineers to check on their own machines.

Contents: valid-a2a-receipt.json · tampered-receipt.json · ct-inclusion-proof.json · sa-epoch-receipt.json · r3pv-proof-vector.json
Endpoint: Client-side zip generation

Hand this to your engineering team. They can verify the valid receipt offline against our public key at receipts.thehiveryiq.com/v1/receipts/pubkey without ever touching this page again.

Primitive by primitive, surface by surface

Where Hive Proof Architecture meets the ICE stack

Twenty-two entries apply to the surfaces described on this page, with the ICE surface each one addresses named plainly. Nothing here is an ICE commitment, an ICE design, or a deployed integration. Twenty are Hive primitives. Two, SPR and IEA, are use case mappings rather than primitives: they describe how an integration binds to contracts that already exist, and they are labelled that way below and in the canon. Four were composed for ICE in the May note, and a fifth from that note, the public transparency log, is designed and not yet serving. Six shipped in the two months since. Seven were filed together on July 26, 2026 as one umbrella provisional, USPTO application 64/119,279, and they sit upstream of everything above. They run before an effect is in scope, not after it lands. Two more bind the deciding model itself.

Every button in this grid runs against production. Nothing here is simulated.

Twenty of the twenty-two entries carry a live control. Verify live posts a real production-signed receipt to the open verify route and shows you the HTTP status, the round-trip time measured in your browser, and every gate that ran. Tamper and re-verify flips one character of that receipt in your browser and posts it again, so you watch the service name the gate that catches it rather than take our word for it. No key, no account, and no call with us is required to run either.

The receipts were minted and verified against proof.thehiveryiq.com by our acceptance script during a passing 241 check run, and they are signed by the production key. They are synthetic acceptance data and contain no customer data. Typed receipt contracts on that host are Ed25519 signed; read the algorithm field in any result to confirm it. Two entries have no receipt to tamper: AmpliHive and the Hive Ledger are services rather than typed contracts, so their buttons make a live call and show the real response. The transparency log has no button at all, because it has no live endpoint.

Primitive 01
SHOD™ · Sub-second Handoff On Demand
LiveMay composition

Six independent gates evaluate an outbound action before it leaves, and every result, including every refusal, is recorded as evidence.

More
See SHOD in the stack

ICE surface: autonomous agent order flow into NYSE tokenized equities and ICE futures, where the gate set that fired is the thing a supervisor will ask about first.

Run the gate evidence live

SHOD is a composition, not a single contract. The evidence a supervisor would pull is carried by typed contracts like effect.closure, which is what this button verifies against production.

Primitive 02
SpectralZK™ · zero-knowledge policy compliance
LiveMay composition

Proves an action satisfied a stated policy without exposing the algorithm, the strategy, or the parameters behind it.

More
See SpectralZK in the stack

ICE surface: policy attestation for buy-side execution agents, so an envelope can be demonstrated to a regulator without a member exposing its model.

Run the disclosure-free proof live

SpectralZK is a composition. replay.disclosurefree is the deployed contract behind it: it lets a run be re-checked without republishing the inputs.

Primitive 03
HAHS™ · Hash-Attested Historical State
LiveMay composition

RFC 8785 JCS canonicalization plus Ed25519, with each record carrying its predecessor hash, so the history is the chain rather than a claim about a chain.

More
See HAHS in the proof architecture

ICE surface: multi-hop market data and price-band publication, where the question is whether the value received is the value issued.

Run the hash-chained envelope live

HAHS is a composition over the envelope spine. hivebound.envelope is that spine, deployed and verifiable here.

Primitive 04
ViewKey™ · role-scoped deterministic projection
LiveMay composition

Derives role-scoped views from the same canonical signed record, deterministically, with no re-serialization and no re-signing.

More
See ViewKey in the stack

ICE surface: the four-lens problem on a 24/7 venue, where a supervisor, custodian, counterparty, and issuer read the same event simultaneously and none should reconcile for the others.

Run the bounded projection live

ViewKey is a composition. proof.prefill is the deployed contract that binds a bounded projection of an input to a receipt.

Primitive 05
SPR™ · Submission Provenance Receipt
Use case mappingBinds to live contracts

One signed record per submission: canonical input, attested submitting key, submission time, and the methodology version in force.

More
See SPR in the proof architecture

SPR is recorded in the canon as a use case mapping, not as a primitive. It binds to hivebound.envelope and authority.delegation, both deployed.

ICE surface: ICE Benchmark Administration and ICE Data Indices contribution intake, one panel bank and one input source at a time.

Run the delegation link live

SPR is a use case mapping, not a primitive of its own. It binds to hivebound.envelope and authority.delegation, both deployed. This verifies the delegation link.

Primitive 06
SA™ · Sequence Attestation
LiveShipped since May

Attests event ordering against an external time source rather than against the producing system's own clock.

More
See SA in the proof architecture

Registered in the canon as sequence.attestation, a typed receipt contract with twelve gates.

ICE surface: Reg SCI detection and notification timelines, where the disputed quantity is the clock rather than the response.

Run the sequence attestation live

Twelve gates, including LOWER_BOUND_BINDING and UPPER_BOUND_BINDING. The floor is a public Base mainnet block hash, the ceiling an RFC 3161 token from freetsa.org. Neither clock is ours.

Primitive 07
IEA™ · Issuer Event Attestation
Use case mappingBinds to live contracts

Binds an issuer or oracle event to the attested key that produced it, checkable against a public anchor with no intermediary trust.

More
See IEA in the proof architecture

IEA is recorded in the canon as a use case mapping. Event ordering is live through sequence.attestation. A public inclusion proof is not live, so IEA is not claimed as a primitive.

ICE surface: corporate actions and record dates on tokenized shares, and oracle authority for event-probability data feeds.

Run the supersession record live

IEA is a use case mapping. Event ordering is live through sequence.attestation and supersession.receipt. A public inclusion proof is not live, so IEA is not claimed as a primitive.

Primitive 08
Carnac™ · proof-demand routing
LiveShipped since May

Routes an incoming proof demand to the primitive that can answer it, instead of a general-purpose search across systems.

More
Carnac

ICE surface: regulatory and member inquiry intake across six clearing houses, several exchanges, and a benchmark administrator.

Run a proof demand live

Carnac emits the demand that gates an action. This verifies a production-signed proof.demand receipt against the live service.

Primitive 09
R3Pv™ · proof-vector grouping
LiveShipped since May

Groups related receipts into one proof vector with a single signed score, so triage replaces uniform investigation.

More
R3Pv benchmark

ICE surface: continuous 24/7 order flow, where the volume of events makes ranking evidentiary thinness the only workable review model.

Run the routing receipt live

Eleven gates. routing.receipt is the deployed contract that records which proof vector was grouped and where it was routed.

Primitive 10
AmpliHive™
LiveShipped since May

Amplifies receipt coverage across a fleet of systems without requiring each system to integrate separately.

More
AmpliHive

Recorded in the canon as an external operational service, probed live, not as a typed receipt contract.

ICE surface: breadth across exchanges, clearing houses, data services, and mortgage technology, where a per-system integration program would never finish.

Run AmpliHive self-test live

AmpliHive is an external operational service, not a typed receipt contract, so there is no receipt to tamper. This calls its live self-test and shows the real compile, compress and certify legs it returns. The service sleeps when idle, so a first call can take up to half a minute.

Primitive 11
Hive Ledger · searchable index
LiveShipped since May

A searchable index over receipts, so a per-instrument, per-window, or per-counterparty question is a query.

More
Hive Ledger

ICE surface: supervisory and surveillance lookups, and evidence production for a single instrument across a single window.

Read the accumulator root live

This is the real signed root of the public receipts accumulator. It currently reports leaf_count 0. That is the honest state of the public tree today, and the button shows it rather than hiding it.

Primitive 12
Model Receipts
LiveShipped since May

Pins each model output to the exact model version and evaluation state that produced it.

More
Model Receipts

ICE surface: machine-learning components inside surveillance, risk, and mortgage-technology workflows, where an output should never be attributable to an unidentified model version.

Run the model signature receipt live

S2S means Silicon to Signature. The deployed contract binds a model run to a signature. GPU attestation on this deployment is a stub, disclosed as DSA-STUB in the receipt, not asserted as hardware evidence.

Primitive 13
CT log
Not yet publicMay composition

A public append-only transparency log, designed but not yet serving.

More

The intended endpoint ct.thehiveryiq.com has no DNS record today, so nothing on this page depends on it. What is live instead is the append-only receipt ledger at receipts.thehiveryiq.com/v1/receipts, where every receipt on this page is written in sequence and can be read back by number. That ledger is the fixed point available right now. A public log would remove the last dependency on Hive staying reachable, and it is not there yet.

ICE surface: anything a counterparty, custodian, or supervisor needs to check without asking ICE or Hive for access.

No live control

There is no live control here because there is no live endpoint. The certificate-transparency style log is designed and specified but not public, and the page will not offer a button that does not run.

Upstream · filed July 26, 2026
Seven controls that run before the effect

The thirteen above prove what happened once an action was taken. These seven prove the conditions were sound before it could be taken. Cryptographic core built and proven, 37 of 37 smoke tests pass, ML-DSA-65 signing under NIST FIPS 204, sub-3-millisecond verification. Filed as one umbrella provisional on July 26, 2026, USPTO application 64/119,279. Production integrations are pilot-ready, not deployed.

Primitive 14
PBS™ · Provenance-Bonded Sandbox
LiveFiled Jul 26

Binds the runtime itself. The kernel, binaries, package set, egress rules and GPU firmware are signed as one measurement, then heartbeated, so drift carries a timestamp instead of a guess.

More
See it in the proof architecture

ICE surface: the pre-trade risk engine and any agent harness routing into NYSE tokenized equities or ICE futures. When a supervisor asks what was running when the order left, the answer is a measurement, not a deployment ticket.

Run the sandbox bond live

Verifies a production-signed usap.pbs receipt, the continuous proof that the runtime environment stayed under the control it claimed.

Primitive 15
Refusal Ledger™
LiveFiled Jul 26

Proves which limit was in force at the moment of one decision, and that nobody widened it afterward and dated the change backward. The bound is provable without publishing the number.

More
See it in the proof architecture

ICE surface: price bands, volatility halt thresholds, self-match prevention and member risk limits. A supervisor verifies the band that was live at 09:31:04 without ICE disclosing the parameter set to the market.

Run the refusal record live

Verifies a production-signed usap.refusal receipt, the signed mutation chain over refusal policy.

Primitive 16
Howler™
LiveFiled Jul 26

A signed alarm raised from inside the model's own reasoning when intent forms, before any order is emitted. Independently replayable through sparse-autoencoder feature probes.

More
See it in the proof architecture

ICE surface: an execution agent whose reasoning starts to assemble a layering or spoofing shape. The freeze receipt is cut at intent formation, which is upstream of the first message the surveillance system would ever see.

Run the drift alarm live

Verifies a production-signed usap.howler receipt, the signed intent-drift alarm raised inside the reasoning loop.

Primitive 17
Perimeter Bond™
LiveFiled Jul 26

Reach is enforced below the application, and the exact bytecode of the enforcing program is bound into every attempt receipt, including every refusal.

More
See it in the proof architecture

ICE surface: co-location segmentation and venue reachability for member agents. A third party compares the enforcement hash at the moment of the syscall against the hash declared in the manifest, so the control cannot be quietly swapped.

Run the perimeter bond live

Verifies a production-signed usap.perimeter receipt, the bound on what an agent could reach when it acted.

Primitive 18
Diurnal Bond™
LiveFiled Jul 26

Low-oversight windows raise the bar automatically. Consequential actions require k-of-n countersigning from distributed attestors, where k is computed from a signed risk manifold.

More
See it in the proof architecture

ICE surface: the 24/7 tokenized venue described above. Overnight and weekend sessions run with the thinnest human supervision of the week, which is exactly when a consequential action needs more than one awake signer, provably.

Run the diurnal bond live

Verifies a production-signed usap.diurnal receipt, the escalation applied during low-oversight windows.

Primitive 19
Egress Bond™
LiveFiled Jul 26

A cap on what may leave, by semantic class, committed before the run starts. The measurement is signed against that commitment using homomorphic commitments, so a breach names the class that broke it.

More
See it in the proof architecture

ICE surface: market data entitlement and redistribution. Proving a recipient stayed inside licensed depth, symbol scope and latency tier is a per-class volume question, answerable without either side exposing the content that moved.

Run the egress bond live

Verifies a production-signed usap.egress receipt, the semantic-class and volume control applied to what left.

Primitive 20
Forensic Rail™
LiveFiled Jul 26

Incident access bound to a threshold-signed credential needing several independent approvers, scoped so looking cannot become acting, with the analysis replaying byte for byte.

More
See it in the proof architecture

ICE surface: the disputed print, the halt nobody agrees about, the reconstruction a counterparty will contest. The investigation becomes checkable evidence rather than a narrative about evidence.

Run the forensic rail live

Verifies a production-signed usap.forensic receipt, the bonded incident-response access record.

Runs underneath all twenty-two
The signing layer itself

Not surface specific. This is what makes every receipt above verifiable, and it is live on receipts.thehiveryiq.com today.

Hive Typed Signer

The ML-DSA-65 engine under NIST FIPS 204, with Ed25519 alongside it in one aggregate envelope. Sign p50 0.098 ms, verify p50 0.136 ms, verifiable in a browser and via WASM with no network.

Hive verifier

The public checking surface. A verifier that a custodian or a supervisor runs themselves is the only kind that settles an authenticity question.

Model and memory · the MiR stack
Two that bind the deciding model

The thirteen above prove what happened once an action was taken. The seven prove the conditions were sound before it could be taken. These two prove what the deciding model was, and what was known about it before it was trusted. Three further primitives in the same stack are named at the foot of this page rather than mapped to a surface, because stretching them to a market-infrastructure question would be an overclaim.

Primitive 21
MiR · Model Identity Receipt
LiveModel stack

Binds each step of a chain to a declared model identity, its weights digest, its configuration hash, and its endpoint, and reports when the model that answered was not the model that was expected.

More

ICE surface: autonomous agent order flow, where a supervisor asking which model made a decision needs an answer that survives the model being swapped, upgraded, or silently substituted between steps. Model Receipts pin an output to a model version. MiR proves the whole chain ran the model that was approved for it.

Run the model identity receipt live

Ten gates. sigr.mir binds the model that actually answered to the model that was promised, and flags identity flicker across steps.

Primitive 22
EvAR · Eval-Attestation Receipt
LiveFiled Jun 22

Signs which model, rubric, dataset, method, scored items, and evaluator produced an evaluation result, and records whether a check for contamination between the dataset and the training data was run and came back clean. It wraps an existing evaluation framework rather than replacing it.

More

Filed as a provisional on June 22, 2026. Deployed since, and verifiable from this tile against the canon entry.

ICE surface: model validation evidence for the machine learning components inside surveillance, risk, and mortgage-technology workflows, where a validation record is today a document produced by the same team that built the model.

Run the eval attestation live

Verifies a production-signed eval.attestation receipt against the live /verify/evar route.

Named in the proof architecture, deliberately not claimed here
Primitives with no defensible market-infrastructure use case on this page

These are real Hive primitives, but stretching them to fit an exchange, clearing, or benchmark evidence question would be an overclaim, so they are listed by name only and not mapped to a surface above.

Media Origin Receipt, MoR MoRSo, Segments of Origin OriginProof Protected Flow Fleets Proof Credit Structural Lateration MiR-M, memory-retention measurement MPP, memory-performance profile MaR, memory-aware routing
Cliffsnotes

Six disclosed exposures, six factual questions, six compositions.

This table is the short form. Each row quotes ICE's own disclosed language, states the narrow factual question underneath it, and names the primitives that would carry the answer. The long form, with sources and with what a record cannot decide, sits directly below.

Disclosed exposure The factual question underneath Primitive composition
“Our subsidiaries must be able to demonstrate that these benchmarks and indices are determined with integrity and are not readily subject to manipulation.” FY2025 Form 10-K, risk factors Read the long form For one contributor at one moment: what canonical input was submitted, from what attested key, at what time, and under what benchmark version.
SPR™HAHS™ViewKey™
“A cyberattack could occur and persist for an extended period of time before being detected.” FY2025 Form 10-K, risk factors Read the long form For one detection event: when was it detected, by which system, and against what external anchor.
SA™Carnac™
“We also may face increased reputational risks from our investments in companies like Polymarket and Bakkt due to their evolving business models.” FY2025 Form 10-K, risk factors Read the long form For one oracle event: who was authoritative, at what time, and against what public identity anchor.
IEA™AmpliHive™
“A defect in a new software release by NYSE subsidiary Securities Industry Automation Corporation, or SIAC, caused incorrect calculation of industry-wide trade price bands.” FY2025 Form 10-K, risk factors Read the long form At each publish step: was the value the one that was issued, or was it mutated downstream.
SA™R3Pv™
Tokenized-securities agent execution. Not a dispute and not a filed matter, and every initiative named in the long form is subject to regulatory approval. Forward looking Read the long form For one autonomous agent order: which risk-envelope version was in force, which gates fired, which policy was attested, and what did each counterparty see.
Complete proof architectureFour lenses, live in browser
“Anthropic’s Claude Mythos Preview is deployed across all of ICE’s businesses including the NYSE and other exchanges, clearinghouses, data services business, and mortgage technology platform, with the deployment, security architecture and governance of its use managed by ICE.” ICE press release, June 3 2026 Read the long form For one agent action: which model version acted, under what authority, against which record, and what did the reviewer see before it took effect.
Carnac™Howler™ViewKey™
ICE's own disclosed risk language, stated neutrally

From a disclosed exposure to the factual question underneath it

The six rows in the table above, in full: the exact quoted language, the sources, and what a record cannot decide in each case.

Open the long form, six exposures with sourcesSix rows

Each row below quotes ICE's own risk-factor language from its FY2025 Form 10-K, or a publicly reported event, described without characterization and with its source linked. The middle column is the narrow factual question a regulator, a counterparty, or a court may need answered. The right column is what a chained receipt could carry, and directly beneath it, what such a record cannot decide. Nothing here is a prediction about any outcome.

The same 10-K states the strategic pressure plainly: "as the demand for certain assets and markets to be available 24/7 and the demand for settlement on-chain increases, if we do not innovate, we risk being displaced or disintermediated." Every row below is an evidence question that gets harder, not easier, as that demand is met.

A. Disclosed exposure or event
B. Factual question underneath it
C. What a chained receipt could show, and what it cannot decide

Benchmark and index integrity at IBA and ICE Data Indices

FY2025 Form 10-K, risk factors"Our subsidiaries must be able to demonstrate that these benchmarks and indices are determined with integrity and are not readily subject to manipulation."

The same risk factor notes that ICE subsidiaries "administer, produce and license a variety of benchmarks and indices across multiple asset classes, which are used globally and are referenced in certain of our trading and clearing products," and that failures, negative publicity, or lawsuits related to that administration carry consequences.

Source: ICE FY2025 Form 10-K on SEC EDGAR.

For one contributor at one moment: what canonical input was submitted, from what attested key, at what time, and under what benchmark version.

Not the aggregate. The single submission, in the form it arrived, bound to the identity that sent it and the methodology version then in force.

A signed submission receipt per panel bank and per input source, carrying the RFC 8785 canonical form of the input, the attested submitting key, the submission timestamp, and the benchmark methodology version identifier. SPR™ issues it at the door. HAHS™ chains it, so a removed or reordered submission breaks the link rather than disappearing.

A record cannot decideWhether a contribution is truthful. It shows only that the submission is authentically from the party of record, in the form that party sent, at the time it was sent. Truthfulness of the underlying input remains a question for the administrator, the contributor, and the supervisor.

Reg SCI and the cyber-detection window

FY2025 Form 10-K, risk factors"A cyberattack could occur and persist for an extended period of time before being detected."

On May 22, 2024 the SEC announced that ICE agreed to pay a $10 million penalty to settle charges that it caused nine wholly-owned subsidiaries, including the New York Stock Exchange, to fail to timely inform the Commission of a 2021 VPN system intrusion as required by Regulation SCI. ICE and the subsidiaries consented to the order without admitting or denying the findings.

Gurbir S. Grewal, then Director of the SEC's Division of Enforcement, stated: "When it comes to cybersecurity, especially events at critical market intermediaries, every second counts and four days can be an eternity."

Sources: SEC press release 2024-63, ICE FY2025 Form 10-K.

For one detection event: when was it detected, by which system, and against what external anchor.

The disputed quantity in a notification-timing matter is rarely the response. It is the clock. A detection time asserted after the fact by the party being examined is the weakest possible form of that fact.

A signed detection event at the moment of detection, naming the detecting system, ordered against GPS-disciplined or PTP time and written into the append-only receipt ledger at /v1/receipts. SA™ supplies the sequence attestation, so the ordering of detection, escalation, and notification is readable rather than reconstructed.

A record cannot decideWhether the response was reasonable. Whether an assessment was adequate, whether an estimate of de minimis impact was correct, and whether a notification obligation was met are judgments about conduct. A receipt fixes the timeline those judgments are applied to.

Reputational carry from strategic investments

FY2025 Form 10-K, risk factors"We also may face increased reputational risks from our investments in companies like Polymarket and Bakkt due to their evolving business models."

Two publicly reported 2026 incidents illustrate the shape of that carry. On May 22, 2026 an approximately $520,000 drain of Polygon-based tokens was traced to a private-key compromise rather than a contract flaw. On June 26, 2026 a compromised third-party vendor injected a malicious script into the Polymarket frontend, and roughly $2.94 million was taken from at least eleven user wallets in a supply-chain phishing attack. Polymarket removed the dependency and committed to refunding affected users.

Sources: ICE FY2025 Form 10-K, Cryptopolitan on the May 2026 key compromise, Halborn analysis of the June 2026 supply-chain attack, Security Boulevard.

For one oracle event: who was authoritative, at what time, and against what public identity anchor.

In both incidents the failure sat outside the protocol logic, in the identity of the thing that spoke. That is an authority question, not a computation question.

A source-signed record verifiable against a public anchor without intermediary trust. IEA™ binds an issuer or oracle event to the attested key that produced it, and the ledger sequence number gives an outside reader a fixed point to check against. A script injected between the source and the reader cannot produce that signature.

A record cannot decideWhether the underlying event occurred. A receipt establishes that a specific authority asserted a specific value at a specific time. It does not make the assertion correct, and it does not restore drained funds.

Industry-wide price-band defect at NYSE subsidiary SIAC

FY2025 Form 10-K, risk factors"A defect in a new software release by NYSE subsidiary Securities Industry Automation Corporation, or SIAC, caused incorrect calculation of industry-wide trade price bands."

The same disclosure states that the June 3, 2024 defect affected certain securities symbols, led to trades occurring at incorrect prices that were required to be reversed, triggered trading pauses based on the incorrect bands, and resulted in member claims paid by the NYSE exchanges under their rules in combination with SIAC.

Sources: ICE FY2025 Form 10-K, contemporaneous reporting on the June 3, 2024 price-band incident.

At each publish step: was the value the one that was issued, or was it mutated downstream.

A published band travels through several hops before a participant acts on it. Distinguishing a bad calculation from a good calculation that was altered in transit is a provenance question at each hop.

A signed publish record at each hop, independently verifiable, so the value a participant received can be compared byte for byte against the value the publisher issued. HAHS™ makes the hop sequence readable, and ViewKey™ lets each consumer read the projection that applies to them without changing the signed bytes underneath.

A record cannot decideCorrectness of the calculation itself. If a software release computes the wrong band, a faithful receipt records the wrong band faithfully. It narrows the question to where the error entered, which is a different and smaller question.

Tokenized-securities agent executionForward looking

This row is not a dispute and not a filed matter. On January 19, 2026 the New York Stock Exchange announced development of a platform for trading and on-chain settlement of tokenized securities, with 24/7 operations, instant settlement, dollar-sized orders, and stablecoin-based funding, subject to regulatory approvals. On March 5, 2026 ICE announced a strategic relationship and investment in OKX under which, subject to regulatory approval, OKX would provide access to ICE's U.S. futures and NYSE tokenized equities markets. On March 24, 2026 NYSE and Securitize announced a memorandum of understanding naming Securitize as the first digital transfer agent eligible to mint blockchain-native securities on the planned Digital Trading Platform. On February 3, 2026 ICE announced that ICE Clear Credit's U.S. Treasury clearing service had received SEC approval and was operationally live. On June 22, 2026 ICE and OKX announced a fifty-fifty joint venture, OKXICE, which subject to regulatory approvals is expected to operate as a U.S.-registered broker-dealer and futures commission merchant.

Sources: Fortune on the ICE investment in OKX, ICE on Treasury clearing going live, NYSE and Securitize memorandum of understanding, NYSE tokenized securities platform announcement, reporting on the OKXICE joint venture.

For one autonomous agent order: which risk-envelope version was in force at that moment, which gates fired, which policy was attested, and what did each counterparty see.

Four readers will ask four different versions of that question at four different times, and each will want an answer that does not depend on trusting the others.

A SHOD™-gated, SpectralZK™-attested, Ed25519 plus ML-DSA-65 dual-signed A2A receipt carrying the envelope version identifier, the ordered gate results, the policy attestation, and four role-scoped views derived from the same signed bytes: a supervisor lens, a custodian lens, a counterparty lens, and an issuer lens. ViewKey™ derives each view deterministically, so no reader gets a different underlying record.

A record cannot decideWhether the underlying strategy is compliant. It shows only that the policy in force at the moment of the trade matches what was pre-committed. Whether that policy was itself adequate remains a supervisory judgment.

Agents already running inside ICE systemsAlready deployed

This row is different from the one above it. Nothing here waits on a regulatory approval, and none of it is a dispute. These are systems ICE describes as deployed. On June 3, 2026 ICE announced it had joined Anthropic’s cyber security initiative, Project Glasswing. In ICE’s own words, “Anthropic’s Claude Mythos Preview is deployed across all of ICE’s businesses including the NYSE and other exchanges, clearinghouses, data services business, and mortgage technology platform, with the deployment, security architecture and governance of its use managed by ICE.” ICE described the objective as advancing the use of AI across its cyber security “in a manner that is secure, auditable, and designed for regulated industries.” Earlier, on March 17, 2026 at ICE Experience, ICE unveiled AI voice and chat agents in beta for mortgage servicing, integrated with MSP®, its servicing system of record. ICE states the agents “are designed to help homeowners answer questions about their loans and execute loan management actions within governed processes,” and announced the immediate availability of sixteen exception-based servicing automation agents built within ICE Business Intelligence and integrated with MSP.

Two words in ICE’s own language set the scope of this row. The first is auditable, which ICE names as a design requirement. The second is execute, which is what the servicing agents do to a live loan record. Governance here is managed by ICE and stays managed by ICE. The only question this row asks is what portable artifact a third party can check afterward.

Sources: ICE on joining Project Glasswing, ICE on AI voice and chat agents for mortgage servicing, Anthropic on Project Glasswing.

For one agent action: which model version acted, under what delegated authority, against which record, which gates fired before the effect, and what did the human reviewer actually see at the moment of review.

An internal log answers this for ICE. The narrower question is what a servicer, an examiner, a counterparty, or a court can check months later without being asked to trust the same system whose behavior is the subject of the question.

A Carnac™-gated, dual-signed action receipt issued before the effect, carrying the model and policy version in force, the ordered gate results, the delegated authority under which the agent acted, and the digest of exactly what was presented to the reviewer. Howler™ records a signed alarm when the reasoning loop drifts from the stated intent, so a drift event is evidence rather than an absence. ViewKey™ derives role-scoped views from the same signed bytes, so a servicer, an examiner, and an internal supervisor read one record through three lenses instead of three exports that have to be reconciled.

A record cannot decideWhether the action was correct, whether the governed process was adequate, or whether the agent should have acted at all. It shows only which policy was in force, which gates fired, and what the reviewer was shown at that moment. Judgment on the policy itself stays with ICE and its supervisors.
One signed receipt, four readers

Selective disclosure without data drift

The failure mode in multi-party markets is not usually secrecy. It is drift. Each party is handed a differently shaped extract, each extract is reconciled separately, and by the time a question is asked the four versions no longer agree. ViewKey™ removes the extracts. One receipt is signed once, and each role reads a deterministic projection of the same signed bytes. Nothing is re-serialized, nothing is re-signed, and no reader can see a field their role does not carry.

Lens one
Supervisor lens
  • Policy attestation and envelope version in force
  • Ordered gate decisions, including refusals
  • Timestamps against the external anchor
  • Chain position and predecessor hash
No PII. No positions. No account identifiers.
Lens two
Custodian lens
  • Counterparties to the transfer
  • Asset identifier, CUSIP or CMTA reference
  • Settlement venue and settlement instruction reference
  • Acknowledgment state on both sides
No policy metadata. No gate reasoning. No strategy signal.
Lens three
Counterparty lens
  • Trade identifiers on both sides
  • Fill price
  • Quantity and any partial-fill state
  • Execution timestamp
No policy attestation. No custody routing. No issuer fields.
Lens four
Issuer lens
  • Corporate-action-relevant fields only
  • Record-date-relevant holder aggregation reference
  • Instrument and class identifier
  • Event attestation linkage
No fill price. No counterparty identity. No policy metadata.

Why this matters for a 24/7 tokenized venue. When an equity trades continuously and settles on-chain, the four readers above stop being sequential. They become simultaneous. A regulator asking about policy at 02:14 UTC, a custodian asking about settlement instruction at 02:14 UTC, and an issuer asking about a record date all need to be reading the same event, and none of them should have to accept a reconciliation performed by one of the others.

ViewKey™ projections are derived, not produced. The projection function is deterministic and the input is the canonical signed record, so two independent implementations of the same lens return identical bytes. That is what makes selective disclosure and evidentiary integrity compatible rather than opposed.

The honest scope of the claim

What an independent record changes, and what it does not

Show the full narrowed-versus-unavoidable breakdown

An independently checkable record would not have prevented any of the events referenced above. In several of them, it could have narrowed the part of the argument that is only about what actually happened.

That is a deliberately small claim. The list on the left is the whole of it. The list on the right is everything a record leaves exactly where it was.

Could have helped narrow

  • Evidence that has to be rebuilt or cannot be found. A record that exists at the moment of the action does not need to be reassembled from systems afterward.
  • Ambiguity about which methodology or envelope version actually ran. The identifier and version travel inside the record, so there is no later argument about which one was in force.
  • Detection-time assertions that rest on the examined party's own clock. An externally anchored timestamp is a fact an outside reader can check.
  • Downstream mutation of a published value. A per-hop signed publish record separates a bad calculation from a good calculation that was altered in transit.
  • Four parties reconciling four differently shaped extracts. One signed record with four derived lenses removes the reconciliation step entirely.
  • Expensive after the fact reconstruction. Answering a narrow factual question becomes a lookup against records rather than a project against logs.

Could not have avoided

  • An untruthful input. If a contributor submits a wrong number, a receipt authenticates the submission, not the number.
  • A software defect. A faithful record of an incorrect price band is still an incorrect price band.
  • Judgments about reasonableness. Whether a response, an assessment, or a supervisory program was adequate is not a fact a receipt carries.
  • An outage or an intrusion. A record can fix the timeline. It cannot keep a system running or keep an attacker out.
  • A key compromise at a third party. A receipt makes the authority question checkable. It does not custody anyone else's keys.
  • Conduct that was wrong. If a control was misconfigured or a decision was wrong, a faithful record of it does not make it right. It records it accurately.
The primitive composition

How Hive does it, for the technically inclined

Show the composition detail

Nothing below is a new cryptosystem. It is a composition of the thirteen downstream primitives that already run, arranged so that one event produces one signed record with four readable projections. The seven upstream controls sit before this composition rather than inside it. Every name is live and operational on receipts.thehiveryiq.com, where the envelope carries a dual signature: ML-DSA-65 under NIST FIPS 204, plus Ed25519. The typed receipt contracts you can verify from the primitive grid above run on proof.thehiveryiq.com and are Ed25519 signed today, with no post-quantum slot. Click any tile and read the algorithm field to confirm that for yourself.

0.098 ms
Sign, p50
0.136 ms
Verify, p50
Browser + WASM
Offline verifiable
FIPS 204
ML-DSA-65 plus Ed25519
The eleven pieces, in the order they run

Before anything is signed, the record is serialized deterministically using RFC 8785 JSON Canonicalization Scheme, so two identical records always produce identical bytes and any independent implementation can reproduce the hash. Hive Canonical bytes are hashed with SHA-256, and that digest is signed twice: once with Ed25519, and once with ML-DSA-65, the post-quantum lattice signature standardised as NIST FIPS 204. Each receipt carries the hash of the receipt before it, so a removed or reordered step breaks the link rather than vanishing quietly.

1SHOD™ 2SpectralZK™ 3HAHS™ 4SPR™ 5SA™ 6IEA™ 7ViewKey™ 8Carnac™ 9R3Pv™ 10Hive Ledger 11CT log

1. SHOD™, Sub-second Handoff On Demand

Six independent gates evaluate an outbound action before it is allowed to leave. Each gate result is recorded, including refusals, so the gate decision set is part of the evidence rather than a log line beside it. For an agent order this is where the risk envelope is checked against the action, before anything reaches a matching engine.

2. SpectralZK™, zero-knowledge compliance

Proves that an action satisfied a stated policy without revealing the algorithm, the strategy, or the parameters that produced it. A supervisor can verify that the envelope held without reading the model. This is the piece that makes an execution agent's policy attestation shareable with a regulator at all.

3. HAHS™, Hash-Attested Historical State

RFC 8785 JCS canonicalization plus Ed25519, with each record carrying its predecessor hash. The history is the hash chain, not a database that asserts a history. Gaps and reorderings are structurally visible.

4. SPR™, Submission Provenance Receipt

The per-submission record: canonical input, attested submitting key, submission time, and the methodology or schema version in force. This is the primitive that answers the benchmark contribution question one contributor at a time.

5. SA™, Sequence Attestation

Attests the order of events against an external time source, GPS-disciplined or PTP, rather than against the producing system's own clock. Detection, escalation, and notification become an ordered sequence an outside reader can check.

6. IEA™, Issuer Event Attestation

Binds an issuer or oracle event to the attested key that produced it, so the authority question is answerable against a public anchor without trusting any intermediary that sat between the source and the reader.

7. ViewKey™, deterministic role-based projections

Derives role-scoped views from the canonical signed record. The projection function is deterministic, so two independent implementations of the same lens return identical bytes, and no reader receives a re-serialized or re-signed extract. This is the mechanism behind the four lenses above.

8. Carnac™, proof-demand routing

Routes proof demands to the primitive that can answer them, so a question arriving from a supervisor, a custodian, or a counterparty is resolved against the right receipt class instead of a general-purpose search.

9. R3Pv™, proof-vector grouping

Groups related receipts into a single proof vector with one signed score, so a risk team can triage the thinnest evidentiary boundary first rather than investigating every event equally.

10. Hive Ledger, searchable index

A searchable index over receipts, so a per-instrument, per-window, or per-counterparty question is a query rather than a reconstruction project.

11. CT log, designed and not yet public

A public certificate-transparency-style append-only log, intended to let an outside reader check a receipt root without depending on Hive remaining cooperative or even remaining reachable. It is not serving yet, and ct.thehiveryiq.com has no DNS record today. The append-only ledger at /v1/receipts is live and is what every receipt on this page is actually written to.

All of it under one envelope. The eleven pieces above do not each carry their own signature scheme. Their outputs are aggregated into a single ML-DSA-65 plus Ed25519 envelope, which is what a verifier checks. That is why a full verify is a couple of milliseconds and why a verifier can be a single file on someone else's desktop.

These names and their dependency order are Hive internal proof architecture. They are not an industry standard and they are not an ICE design. They are listed here so a technical reader can map the plain language above onto the internal pieces.

Do not take our word for it

Offline verification

A record is only useful as evidence if someone who distrusts the producer can still check it. Every receipt described on this page verifies in a browser with no Hive code privileges, and via WASM with no network connection at all. The public verify route is live.

Show how independent verification works

The live verify route

POST https://receipts.thehiveryiq.com/v1/receipt/verify

Send a receipt or an aggregate envelope and the route returns the check results: canonicalization, Ed25519 signature, ML-DSA-65 signature, chain linkage to the predecessor hash, and ledger inclusion where present. The same checks run inside the WASM verifier with no network at all, using the public key carried inside the record itself.

Check a record you were handed

Paste a single receipt or a whole aggregate envelope. The page posts it to the live verify route and reports what came back. If the call is blocked by CORS or you are offline, this widget says so plainly rather than reporting a success it did not get.

See the four lenses move. The agent-to-agent walkthrough runs the SHOD™ gate set, the SpectralZK™ attestation, and the ViewKey™ projections live, then verifies the resulting receipt in the browser. It is the same primitive composition described above, exercised on a synthetic order.

Open the live A2A demonstration thehiveryiq.com/a2a-demo
Three candidate surfaces · ranked · every fact below is linked to its primary source

Three surfaces where the record is already contested

A surface earned a place here on three tests. There is a reader who doubts and who has real power over ICE. It is live now or inside twelve months. And a shadow window needs no production change. Everything asserted below is drawn from ICE's own filings, calls and podcasts, from SEC releases, or from the IMF, and each is linked. Where the widely repeated version of a fact is wrong, the correction is stated rather than quietly skipped.

01 · Tokenized equity lifecycle, and the two record keepers that are not the same

The detail most summaries get wrong is worth getting right, because Michael Blaugrund has explained it himself. There are two separate tokenization tracks, and they have different record keepers, so they raise different evidentiary questions.

On the first track, NYSE rule filing SR-NYSE-2026-17 took effect immediately upon filing on 17 April 2026 under Release No. 34-105260. It was a notice of filing and immediate effectiveness, not an approval order, and comments stayed open until 13 May 2026. It enables trading of DTC-eligible securities in tokenized form. Per the Federal Register notice, a tokenized share trades alongside its traditional counterpart with the same execution priority only if it is fungible with it, shares the same CUSIP and trading symbol, and carries the same rights. The record sits with DTC: its software tracks each transfer to record tokenization entitlements for DTC's official books and records.

On the second track, NYSE's planned Digital Trading Platform, an MOU with Securitize names it the first digital transfer agent for issuer-sponsored blockchain-native securities. There the transfer agent maintains the official record, not DTC.

So the question is not one question. Under the DTC pilot it is whether the chain event and the DTC entitlement record agreed at a named instant. Under the Digital Trading Platform it is whether the venue's execution record and the digital transfer agent's official record agreed. Both are reconciliations between two independent books, and today both are answered by whichever party is being asked. The IMF's April 2026 note Tokenized Finance (IMF Note 2026/001, Tobias Adrian) adds the timing pressure: eliminating settlement lags and end-of-day netting shifts liquidity demand to continuous real time, and stress events are likely to unfold faster, leaving less time for discretionary intervention. In fairness the same note calls the net financial-stability effect uncertain, and credits atomic settlement with reducing credit and counterparty risk.

What a receipt does here. SA™ folds the lifecycle events into a root in exactly the recorded order and brackets that root between two independent external time anchors, so the order is checkable offline at 02:14 UTC on a Sunday without calling ICE. R3Pv™ reduces the five events to one vector that names the weakest boundary. ViewKey™ renders one signed object through four role-scoped lenses, so a supervisor, a custodian, a counterparty and a transfer agent read the same bytes rather than four reconciled exports.

The reader who doubts. The SEC's Division of Trading and Markets, while the ICE and OKX joint venture announced 22 June 2026 is still, in ICE's own words, subject to certain regulatory approvals before it can operate as a US broker dealer and FCM. Evidentiary strength demonstrated now is the cheapest thing to have already built.

Honest limit. A receipt proves the recorded events fold in the recorded order between two anchors. It does not prove either book is substantively correct, does not reconcile them for you, and Hive holds no position on whether a tokenized share and a DTC entitlement in fact matched.

02 · Aurora agent actions, and the auditor ICE is already paying for

This is the surface ICE described on its own earnings call. On 30 July 2026, ICE President Ben Jackson told analysts that the GSEs "have come out with pretty strict guidance on how AI should or should not be used," that ICE has "gone through and audited our processes internally," and that ICE has "also hired an external auditor to go through and look at how we use AI" (Q2 2026 call transcript). ICE is paying a firm to produce an assurance artifact about agent behaviour. A signed receipt is that artifact as a byproduct of the action rather than as a report written afterward.

The scale is ICE's own: the servicing business processed 10.7 billion API and web services calls in Q2 2026, up 39 percent year over year, and ICE describes itself as the network of record for US housing finance where "roughly 9 in 10 mortgages touch our network at some point." Both are ICE management statements rather than audited market share.

One correction worth making, because overclaiming it would be caught immediately. ICE Aurora is not an autonomous actor on live loans. ICE says Aurora embeds agentic AI in Encompass and MSP "with governance, audit logs and human approvals built in," that AI "assists the human in high-risk decisions such as underwriting, pricing and cash movement, escrow and remittance and doesn't autonomously make a call," and that human-in-the-loop review is mandatory for material decisions. So the useful receipt is not a claim that a machine decided unsupervised. It is a signed record of the delegated authority the agent held, the governed process selected, the ordered gate results, and a digest of exactly what the human reviewer was shown at the moment of approval. That last field is the one nobody has today.

The reader who doubts. A GSE examiner, the CFPB, and ICE's own external auditor, whose report is itself produced by a process the examiner has to take on trust. ICE has already adopted the argument: on the same call it described its MCP work as "a complete audit trail so that every output can be traced and trusted." The remaining gap is authorship. An audit trail is worth more when it is not authored by the party being audited.

Honest limit. A receipt proves what was recorded, in what order, under whose delegated authority, and what the reviewer was shown. It does not prove the decision was correct, does not evaluate model quality, and does not substitute for the human approval ICE requires.

03 · The Reg SCI clock, across an estate that keeps growing

The cheapest pilot, and the one where the pain is already on the record. By SEC order of 22 May 2024, Administrative Proceeding File No. 3-21947, ICE agreed to a $10 million penalty over the failure of nine wholly-owned subsidiaries, including the New York Stock Exchange, to timely inform the SEC of a cyber intrusion as Regulation SCI requires. The order found they took four days to assess impact and internally concluded it was de minimis. Then-Director of Enforcement Gurbir Grewal put it in the words a plaintiff will reuse: "every second counts and four days can be an eternity."

A precision note, since this figure is widely misattributed. That $10 million comes from the SEC's order. It is not the $10 million that appears in ICE's Form 10-K, which is a separate and unrelated expense for claims following a June 2024 NYSE equity trading issue. Conflating the two is a factual error, and the kind a general counsel notices.

What makes the clock a live problem rather than a closed one is that the estate keeps growing. ICE agreed on 30 July 2026 to acquire MarketAxess at $167 per share in cash, roughly $6.0 billion of equity value and about $5.7 billion of enterprise value, expected to close in the first half of 2027, and the OKX joint venture is pending. Every system that joins the family inherits the notification obligation on day one.

What a receipt does here. The disputed quantity in one of these matters is never the response. It is the clock, asserted after the fact by the party under examination. SA™ signs the detection event against disciplined time at the moment of detection rather than the moment someone opens a ticket, then carries detection, escalation, assessment and notification as one ordered sequence anchored outside ICE. "When did you know" becomes a lookup instead of a war room.

Honest limit. A receipt proves when a record was signed and in what order, against an external anchor. It cannot prove that the detection itself was timely, that nothing went unrecorded, or that the assessment reached the right conclusion.

Considered and left off

IBA benchmark contribution provenance, one SPR™ per panel submission, is a strong surface and the index business is growing quickly: ETF assets under management benchmarked to ICE indices reached a record $922 billion in Q2 2026, up 29 percent year over year, after ending 2025 at $794 billion. It is left off this shortlist only because it sits in the FIDS organisation rather than Strategic Initiatives, and because it does not ride either the tokenization or the agentic-AI question. It is the obvious fourth.

How to check this section

Every figure above links to the document it came from, with one exception: the IMF note is cited by title and number rather than by URL, because that document does not serve reliably to every network and this page does not publish links it has not loaded itself. Where a number is an ICE management statement rather than an audited figure, it says so. Where the common version of a fact is wrong, the page corrects it in place: the April rule filing took effect on filing and was not an approval, the DTC pilot and the Digital Trading Platform have different record keepers, Aurora requires a human on material decisions, and the Reg SCI penalty is not a 10-K line item. Hive is not affiliated with Intercontinental Exchange, and no ICE system is contacted by this page.

For Michael Blaugrund, Vice President of Strategic Initiatives

The narrow first step, not the platform conversation

Nothing on this page requires a platform decision, a vendor selection, or a change to a single matching engine, clearing system, or benchmark methodology. The useful first move is deliberately small enough that it can be run beside an existing process and thrown away if it proves nothing.

Step one, pick one surface with one adversarial reader

The best candidate is a surface where an outside party already asks ICE to demonstrate something, and where the answer today is assembled from more than one system. A benchmark contribution window, a Reg SCI detection timeline, or a price-band publication path all qualify. One surface is enough.

Step two, shadow it, do not touch it

Hive records receipts alongside the existing process for a defined window. Nothing is gated, nothing is blocked, nothing changes in production behaviour. If the receipt layer stopped mid-window, the underlying process would not notice.

Step three, hand the record to the reader who doubts it

The test is not whether the receipts look good internally. It is whether a supervisor, a custodian, or a member's counsel can verify them without ICE's help and without Hive's cooperation, using the public verify route and the append-only ledger. If they cannot, the exercise failed and should be abandoned.

Step four, measure the only metric that matters

Time from question to settled answer, on that one surface, before and after. Not receipts written, not coverage percentage, not integration milestones. If the assembly work did not shrink, there is no case for a second surface.

What Hive needs from ICE to run it

A schema for the events in scope, a designated key for the submitting system, and a named reader who is willing to try to break the record. No production access, no market data licence, no customer data, and no connection into any ICE, NYSE, or ICE Clear environment.

What ICE keeps if it walks away

The receipts are canonical JSON signed with published standards: RFC 8785 JCS, Ed25519 throughout, and ML-DSA-65 under NIST FIPS 204 on the receipts rail. They verify with an open verifier and no Hive dependency. A shadow window that ends leaves behind a readable artifact rather than a stranded integration.

Start the one-surface conversation See it run first Private working demonstration. No ICE system is contacted by this page.
Read this before drawing conclusions

What this demonstration is, and is not

This is a working demonstration built to make one idea concrete for one reader. It is not an ICE system, it does not connect to any ICE, NYSE, IBA, or ICE Clear system or data, and it is not legal advice.

Show the full scope-and-limits statement
Hive never touches engines, custody, or money

Matching, clearing, benchmark administration, custody, pricing, routing, and settlement stay exactly where they are. Hive records selected facts and acknowledgments. It does not decide, hold, move, route, or execute.

Receipts do not prove facts

A receipt captures the inputs it was given and the identity that gave them. It does not confirm those inputs were true, and it does not confirm that an asset or a payment physically moved.

The four lenses are projections, not permissions

ViewKey™ determines what a role can read from a receipt. It is not an access-control system for any ICE platform and it does not grant or revoke anything.

Quoted risk factors are quoted, not interpreted

Every 10-K passage on this page is reproduced verbatim from ICE's FY2025 Form 10-K on SEC EDGAR. The factual questions beside them are Hive's framing, not ICE's, and not a characterization of ICE's position.

Forward-looking rows are not predictions

The tokenized-securities row describes announced initiatives that are subject to regulatory approval. It asserts nothing about whether, when, or how any of them will launch.

No approval of any kind

No regulator has reviewed or approved this concept. Nothing here is a compliance, regulatory, or legal conclusion, and no ICE entity is represented as a Hive customer, partner, or pilot.

new in the canon · runnable on this page

Make key market facts easier to compare

These receipts add a checkable comparison of named records, a fixed timeline for a detected issue, and a clear record of a screening run beside the ledger, benchmark, settlement, custody, and clearing work already described here. They are code complete and not yet deployed. The runnable runs below report that state.

ledger.parity · Code complete, not deployed

Compare two market records without exposing the books

Your clearing and custody teams can use this receipt when two named records need a direct comparison. It fixes the fields being compared, the cursor and time for each observation, and the two separate attestor keys. The service recomputes the outcome from the committed fingerprints and the stated timing window. A divergence stays visible for the parties who must investigate it. Legal, risk, and operations can read the result without being asked to choose a winner first.

What it does not do. It does not reveal positions, balances, holder identities, or account identifiers, and it cannot confirm that either fingerprint accurately represents the record named because it grants no access to that record. It does not choose the correct record, assign fault, judge the underlying settlement or register update, carry out or reverse settlement, and a late window says only that the observations were too far apart to decide.

POST /verify/ledger-parity · case pass, a clean record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/ledger-parity · case diverge, the two records disagree

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/ledger-parity · case fail, a forged record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

knowledge.timestamp · Code complete, not deployed

Fix when a detected issue first entered the record

This receipt gives your control teams a fixed latest time by which a named detection system held a particular artifact. It records the detection engine, its pinned software build and rule version, an external time reference, and its place in an append only sequence. The artifact itself is represented by a fingerprint, so the receipt can travel without exposing its contents. That creates a common timeline for clearing, custody, legal, and risk when an exception needs review. No person approves the receipt during its minting path.

What it does not do. It does not establish the earliest time the system held the artifact, show that a condition was real, or show that the detection was correct, significant, timely handled, or complete. It does not identify an affected system, person, account, or asset; determine any duty, deadline, or compliance result; or authorize, require, excuse, or assess any notification, escalation, disclosure, remediation, or enforcement action.

POST /verify/knowledge-timestamp · case pass, a clean record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/knowledge-timestamp · case fail, a forged record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/knowledge-timestamp · case fail2, a second forged record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

screening.attestation · Code complete, not deployed

Show exactly how a counterparty screening run was made

This receipt records a counterparty screening run in a form risk and compliance teams can inspect later. It binds the screening engine, ruleset, primary list, supplemental lists, and their pinned versions to the result. It also records the external time reference, the stated validity period, and the signing authority used for the run. The counterparty identifier and list records remain protected as commitments rather than being disclosed in the receipt. A matched outcome can be investigated with the exact screening context still intact.

What it does not do. It does not show that any list is complete, accurate, current, or free of omissions, decide that a counterparty is actually restricted, catch every false negative, catch an engine that clears a match it found, or assess the adequacy of the program, its thresholds, lists, or validity period. It does not identify the counterparty, expose a list record or match score, prove custody of the commitment key or window secret, or admit, block, freeze, reverse, settle, report, or provide a legal, regulatory, or compliance conclusion about any transaction.

POST /verify/screening-attestation · case pass, a clean record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/screening-attestation · case matched, a name matched the list

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

POST /verify/screening-attestation · case fail, a forged record

Nothing has run yet. Click Run it and the answer below comes back from the verifier, not from this page.

Every run above posts a verified request body from this domain to the open verify route and prints what came back. The example receipts are signed with published example keys, so verify reports key_trust example_registry. That is on purpose. Nothing on this page is a production issuance, a customer record, or an endorsement. Patent Pending.