For ICE Data Services · Private
Where ICE could take this next
Each one is a suggested use ICE could sell, across mortgage, clearing, environmental markets, pricing and regulatory records. None is running on an ICE system yet. Each one is built as a reference check you can run here on fictional records, with the Hive protocol under it alongside. Below them are 19 more, one per ICE product.
Start with evaluated pricingA record ICE keeps about its own customers reads as ICE's record. Clients won't put an ICE tool inside their own systems, and other vendors won't send their usage to a meter ICE runs. Hive sits outside all of them. It holds no ICE data and isn't in the path, so if Hive stops, nothing at ICE stops. That is what lets ICE sell the uses below to clients and to other vendors.
1 · Evaluated pricing
For buy-side firms that use ICE evaluated pricing for illiquid bonds and private credit.
Asking for prices on a specific basket tells the pricing source what a firm holds or plans to trade. ICE could offer a record where the firm commits its weights first, ICE signs the prices it delivered, and the result can be checked later by the firm's auditors. ICE never sees the weights, and the firm can't swap in a better price after the fact.
A fund will not hand its positions to the company that prices them, however good the privacy promise. The record has to be kept by someone with no stake in the price.
What runs today: Builds on the zero-knowledge passport, which already runs on ICE gateway deliveries. See the passport. Under it here: a commitment that opens only the records an auditor asks for.
What it doesn't do: It doesn't check whether a price is right. It shows which ICE prices went into a figure and that nothing changed afterwards.
2 · Data from many vendors
For licensing teams at ICE and for clients that blend ICE data with other vendors' feeds.
A bank's model mixes ICE curves with feeds from other data vendors, and each vendor's licence claims the derived result. ICE could offer a month-end record of which inputs fed which outputs. Each vendor sees its own line and the split the parties agreed, and nothing else.
Other data vendors will not send their usage to a meter ICE runs. A record kept outside every vendor is one they can each accept.
What runs today: Extends the two-vendor credit that already runs on ICE and U.S. Treasury data. See it. Under it here: the rule that works out what a blended product may be used for.
What it doesn't do: Hive doesn't set the split. The parties agree the rule, and Hive records the inputs it is applied to.
3 · Licensed data in AI sessions
For ICE data sold into AI assistants, research tools and client retrieval systems.
ICE's worry with AI buyers is data that stays behind after the licence ends. ICE could offer a lease record: what was delivered, to which session, when the right was withdrawn, and the client environment's signed statement that its copies and caches were cleared. If the client runs a hardware enclave, its attestation goes in the same record.
Clients will not let ICE software inside their own computing environments. They can run a published, open checker that signs what happened and shares none of their prompts.
What runs today: Under it here: a rights change followed to every listed consumer and cache, with the ones that haven't confirmed it shown by name.
What it doesn't do: It can't prove a model kept nothing. It records who confirmed the deletion, when, and who didn't.
4 · Mortgage technology
For lenders, servicers and loan buyers using ICE Mortgage Technology.
When a loan goes bad, the buyer goes through the file looking for a reason to send it back. ICE could offer a record of each automated check at closing: the guideline version it used, a fingerprint of each document it read, and the person who signed off. A buyer can check that record without seeing the borrower's personal data. If a guideline is corrected later, the loans that relied on the old version are listed.
ICE builds Encompass. If a software version is ever part of a repurchase dispute, a record ICE keeps about its own software won't settle it. A record kept outside ICE can.
What runs today: Under it here: which rule and source versions a decision actually read, so a stale check can be told apart from a fresh one. Corrections are P07.
What it doesn't do: It doesn't decide whether a loan is defective or who pays. It shows what was checked, with which version, and when.
5 · Clearing
For ICE's clearing houses and the members that clear across several venues.
In a volatile market, each clearing house sees only its own side of a member's book. ICE could accept a member's signed statement that a set of collateral is free, checked against the other clearing house's own signed record, without seeing the size or the counterparty behind it.
Another clearing house won't share its members' positions with ICE, and members won't show ICE their whole book. Both can sign a record that sits outside every venue.
What runs today: Under it here: one shared allowance kept by several systems at once, where no unit can be used twice.
What it doesn't do: It needs the other clearing house to sign its side. Hive doesn't set margin or release collateral.
6 · Environmental markets
For ICE's environmental markets and the companies that retire credits bought there.
Buyers of environmental credits are asked to show that what they retired was issued once and retired once. ICE could offer one record linking a retirement back through the ICE contract to the registry's issuance. A second retirement of the same serial number, anywhere in the record, shows up.
ICE runs the exchange and the clearing. It doesn't run the registries or the buyers' reporting systems, so the chain has to be kept by someone who can take records from all of them.
What runs today: Under it here: an append-only log with a witness that cosigns each checkpoint, so a rewritten history or a split view is caught.
What it doesn't do: It catches double retirement only across the records it holds. It doesn't check what happens outside them.
7 · Systems that came with acquisitions
For ICE teams moving work from systems that came with acquisitions onto newer ones.
During a migration, the old system and the new one both keep running, and the hard part is showing they agree. Hive could take each system's record of the same item as it happens and match them, so a gap shows the hour it opens. Nothing is rewritten in either system.
Each team trusts its own system. A record kept outside both is one both teams can accept.
What runs today: Under it here: two records of the same holding checked against each other through a weekend night.
What it doesn't do: Hive doesn't move or change data in either system. It shows where the two disagree.
8 · Regulatory records
For ICE's own regulatory teams and for clients answering requests under DORA, MiFID II and SEC Rule 17a-4.
Retention rules run for years, longer than any one signing key. ICE could offer records that are renewed before a key or algorithm retires, so an examiner can still check an old record against what was there at the time.
An examiner gives more weight to a record that the firm being examined couldn't quietly change. A copy kept outside the firm gives them that.
What runs today: Under it here: a renewal sealed with both Ed25519 and ML-DSA-65, the NIST post-quantum signature. Both must verify.
What it doesn't do: It doesn't make a firm compliant with any rule. It keeps the record checkable for as long as the rule asks.
Across ICE's products
Each card names a real ICE product, links to ICE's own page for it, and says who would buy. Every one of the 19 now runs here. Press Run it and your browser checks a genuine signed file and the bent copies it has to refuse, then shows how long that took. Four use zero-knowledge proofs, so a figure can be checked without anyone seeing the numbers behind it. These are reference builds on fictional records, not products running on an ICE system. Pick an area to narrow the list.
ICE Fair Value Leveling Service and evaluated pricing · Fund boards, valuation committees and compliance officers
The problem. Under SEC Rule 2a-5, auditors sample by hand whether a fund's valuations used the ICE evaluation and level it says it did.
The record. Each evaluation and level assignment is signed when ICE delivers it and rolled into one quarterly pack the auditor checks in a browser. A later correction lists the valuations that used the old figure.
ICE data APIs and the MCP channel · Heads of market data at banks and funds
The problem. Vendor usage audits take months of legal and IT time on both sides.
The record. A month-end declaration signed by ICE and the firm with the shared key. Retries count once, and pulls made around an agent sit on their own line. ICE could choose to accept it in place of an on-site audit.
ICE FRTB service · Heads of market risk and regulatory capital at banks
The problem. Supervisors question observation counts, and the trades behind them can't be shown with counterparty names.
The record. A signed count of distinct observations per risk factor. An examiner opens only the observations they ask about, and the rest stay closed.
ICE Initial Margin (ISDA SIMM) · Dealers and firms under the uncleared margin rules
The problem. Two counterparties get different margin numbers, and neither wants to show its book to find out why.
The record. Each side's inputs are fingerprinted. The record shows whether the gap is in the trade list, the sensitivities or the model version, without either full book.
ICE private credit reference data · Direct lenders, BDCs and their investors
The problem. Investors and lenders rely on a manager's quarterly word that borrowers stayed inside their covenants.
The record. The borrower's figures are committed privately, then tested against ICE's deal terms. The result can be checked with the passport, and the financials stay closed.
ICE Fund Names Rule, Rule 18f-4 and SFDR services · Asset managers and ETF issuers
The problem. Allocators want to see that a fund met its names test every day without seeing the holdings.
The record. Holdings are committed each day and checked against ICE's classifications. The result is passed with a passport, and the positions stay closed.
ICE Solvency II service (Tripartite Template) · Insurers and the managers they invest with
The problem. Insurers want full look-through. Managers won't hand over every position.
The record. The manager's holdings are committed over ICE's Solvency II data and totalled by risk bucket. The insurer can open single lines only where it needs them.
ICE FATCA, SEC Rule 15c2-11 and entity data · Custodians, broker-dealers and withholding agents
The problem. When a status changes near a trade, the firm has to show what the data said at the time.
The record. Each pre-trade check is tied to the exact ICE reference version it read, so a stale read can be told apart from a fresh one.
ICE Data Indices on Snowflake · Quant funds and research teams
The problem. Bulk history is priced high because copies outlive the licence. Funds also want to show investors their tests ran on unaltered data.
The record. Each query against the share, what it read and what the share reports leaving it, all on one record, with each result tied to the ICE data version.
ICE Chat and Desktop, with Carnac and Live Ink · Banks and trading desks in the EU and UK
The problem. Firms have to show what an assistant was asked, what grounded the answer and who approved the action.
The record. One record of the question, the entitlement that allowed it, the ICE values behind the answer and the person who approved it.
Encompass Investor Connect and MSP · Correspondent aggregators, investors and servicing buyers
The problem. Buyers re-check delivered loan files by hand.
The record. Each delivered loan carries the signed record of its checks and servicing events, and a buyer can check it without the borrower's personal data.
Simplifile Collaboration and MERS eRegistry · Title companies, settlement agents and lenders
The problem. Altered payoff instructions are a common route for wire fraud.
The record. The instructions are fingerprinted when issued and at each handoff. A changed digit shows as a mismatch in the agent's browser before anyone sends money. The record links to the eNote on MERS.
ICE Aurora and AllRegs Ask Regi · Lenders' compliance and underwriting teams
The problem. Lenders want to use AI answers and need to show later what each one rested on.
The record. Each answer carries the document fingerprint, the citation it drew on, the rule version and the person who signed off.
ICE ETF Hub · Authorised participants and ETF issuers
The problem. Showing a proposed basket early leaks trading intent.
The record. A check that the basket meets the agreed bounds against ICE prices, before the bonds in it are revealed.
ICE Bonds risk matching auctions in ICE FI Select · Dealer desks
The problem. Dealers hesitate to upload their whole inventory.
The record. Inventory is fingerprinted at upload. Afterwards, each dealer can check its matched lines against the session's reference price, and unmatched lines are never published.
ICE Best Execution, Liquidity Indicators and ICE Compass · Buy-side execution desks
The problem. Months later, a desk has to rebuild what it knew when it traded.
The record. At execution, the ICE reference price, liquidity score and quotes are sealed into one file and renewed for as long as it is kept.
ICE Benchmark Administration, including the LBMA Gold and Silver Prices · Banks that contribute to ICE benchmarks
The problem. Contributors spend heavily showing their submissions came from their market data.
The record. Each submission is linked to the inputs it drew on and to the published fixing, on a tamper-evident record their compliance team can check.
ICE Global Network timing services · Trading firms in ICE colocation
The problem. Rules require trading clocks to stay traceable to UTC within set limits.
The record. Signed clock-offset readings from ICE's timing service and from the firm's own servers, in one log both can check.
ICE Climate data and Hazard Watch for municipal debt · Underwriters and bond insurers
The problem. Readers ask later which climate figures were used when the bond came to market.
The record. The ICE climate figures in use on the closing date, sealed so anyone reading later can check them.
Tested and timed
Each product has a genuine signed file and two to four bent copies: a figure changed after signing, a record logged in the wrong order, a proof made against the wrong limit, a step that was skipped. The genuine file has to pass and every bent copy has to be refused for the right reason. All 91 cases do. The numbers below are measured, not estimated.
One row for each of the 27, in the order they appear on this page. Click a name to jump to it. The four marked Zero-knowledge check a figure without anyone seeing the numbers behind it.
| # | What it is | Cases as expected | Median check, Node | In your browser |
|---|---|---|---|---|
| The eight suggested uses, from the top of this page | ||||
| 1 | Portfolio value, holdings private Zero-knowledgeValue a portfolio from ICE evaluated pricing, without ICE seeing the holdings. | 5 of 5 | 12.6 ms | Not run yet |
| 2 | Credit across three vendorsCredit every data vendor behind one figure, without any vendor seeing another's data. | 4 of 4 | 1.66 ms | Not run yet |
| 3 | AI licence that ends on recordA licence for AI use that ends when the session ends, with a record of it ending. | 4 of 4 | 1.51 ms | Not run yet |
| 4 | Mortgage loan fileOne record of a loan file from Encompass to MSP and MERS. | 4 of 4 | 1.26 ms | Not run yet |
| 5 | Collateral not pledged twiceShow one clearing house that collateral isn't already pledged at another. | 4 of 4 | 0.61 ms | Not run yet |
| 6 | Certificates, issuance to retirementFollow a carbon credit from the registry to an ICE futures contract to retirement. | 4 of 4 | 0.96 ms | Not run yet |
| 7 | Old and new system matchedTwo systems' records of the same item, matched line by line during a migration. | 3 of 3 | 1.56 ms | Not run yet |
| 8 | 2019 record checked in 2033Records kept for the length of a retention rule, still checkable at the end. | 4 of 4 | 3.06 ms | Not run yet |
| 19 more products ICE could sell, from the cards above | ||||
| 9 | Fair value evidence for fund boardsICE Fair Value Leveling Service and evaluated pricing · Fund boards, valuation committees and compliance officers | 3 of 3 | 1.66 ms | Not run yet |
| 10 | A usage declaration in place of an on-site auditICE data APIs and the MCP channel · Heads of market data at banks and funds | 4 of 4 | 0.94 ms | Not run yet |
| 11 | Real price observations for FRTBICE FRTB service · Heads of market risk and regulatory capital at banks | 4 of 4 | 0.50 ms | Not run yet |
| 12 | Initial margin differences, locatedICE Initial Margin (ISDA SIMM) · Dealers and firms under the uncleared margin rules | 3 of 3 | 0.78 ms | Not run yet |
| 13 | Private credit covenants, checked without the financials Zero-knowledgeICE private credit reference data · Direct lenders, BDCs and their investors | 3 of 3 | 11.8 ms | Not run yet |
| 14 | Fund names and derivatives limits, shown daily Zero-knowledgeICE Fund Names Rule, Rule 18f-4 and SFDR services · Asset managers and ETF issuers | 3 of 3 | 34.3 ms | Not run yet |
| 15 | Solvency II look-through without line-by-line holdingsICE Solvency II service (Tripartite Template) · Insurers and the managers they invest with | 3 of 3 | 0.56 ms | Not run yet |
| 16 | What the reference data said at the moment of a tradeICE FATCA, SEC Rule 15c2-11 and entity data · Custodians, broker-dealers and withholding agents | 3 of 3 | 0.74 ms | Not run yet |
| 17 | Research on ICE history, meteredICE Data Indices on Snowflake · Quant funds and research teams | 3 of 3 | 1.16 ms | Not run yet |
| 18 | AI assistants under DORA and the EU AI ActICE Chat and Desktop, with Carnac and Live Ink · Banks and trading desks in the EU and UK | 3 of 3 | 1.07 ms | Not run yet |
| 19 | Loan delivery and servicing transferEncompass Investor Connect and MSP · Correspondent aggregators, investors and servicing buyers | 3 of 3 | 1.41 ms | Not run yet |
| 20 | Payoff instructions that can't be quietly changedSimplifile Collaboration and MERS eRegistry · Title companies, settlement agents and lenders | 3 of 3 | 0.75 ms | Not run yet |
| 21 | AI answers in underwriting and complianceICE Aurora and AllRegs Ask Regi · Lenders' compliance and underwriting teams | 3 of 3 | 0.81 ms | Not run yet |
| 22 | Custom ETF baskets, checked before they are shown Zero-knowledgeICE ETF Hub · Authorised participants and ETF issuers | 3 of 3 | 12.2 ms | Not run yet |
| 23 | Dealer sweeps without exposing what didn't matchICE Bonds risk matching auctions in ICE FI Select · Dealer desks | 3 of 3 | 1.11 ms | Not run yet |
| 24 | A best execution file for every tradeICE Best Execution, Liquidity Indicators and ICE Compass · Buy-side execution desks | 3 of 3 | 1.43 ms | Not run yet |
| 25 | Benchmark submissions tied to their inputsICE Benchmark Administration, including the LBMA Gold and Silver Prices · Banks that contribute to ICE benchmarks | 3 of 3 | 0.94 ms | Not run yet |
| 26 | Clock readings in colocationICE Global Network timing services · Trading firms in ICE colocation | 3 of 3 | 0.84 ms | Not run yet |
| 27 | Climate figures at municipal issuanceICE Climate data and Hazard Watch for municipal debt · Underwriters and bond insurers | 3 of 3 | 0.63 ms | Not run yet |
Nothing you check leaves this page.
Measured 2026-10-14 on v20.20.1, one core of an Intel Xeon at 2.90 GHz in a shared cloud sandbox: 300 runs per product after warm-up, 40 for the zero-knowledge ones. A check covers every signature, the log checkpoint and its witness, and the product's own rule. The zero-knowledge checks take about 12 ms per proof because PLONK verification needs pairing arithmetic; the fund names check verifies three days, so three proofs.
Making a proof is slower than checking one, and it happens once, where the private numbers are. On the same machine the median was 1.7 s for the covenant proof and 5.9 s, 5.8 s and 5.7 s for the eight-position portfolio, names and basket proofs. 13 proofs were made for these files. The build also tried to make a covenant proof for a borrower over its limit, and the prover refused, as it has to.
The same runs, made over the network from this page in a cloud Chrome browser on 2026-10-14, with the browser cache cleared before each cold run.
| What | Cold cache | Warm |
|---|---|---|
| This page, first byte | 183 to 238 ms | |
| Run all 27: fetch every file in parallel | 98, 169 and 665 ms | 0.2 ms |
| Run all 27: the checks | 357 to 416 ms | 212 to 215 ms |
| Run all 27: total | 456, 530 and 1,084 ms | 214 to 216 ms |
| First zero-knowledge card: fetch, including the proof checker | 122 ms | |
| First zero-knowledge card: check, including the proof library's one-time setup | 150 ms | 18 ms |
| A card without proofs, such as payoff instructions | 29 ms fetch, 2.5 ms check | 2.7 ms |
What was slow, and what we fixed. The first version of Run all fetched the 27 files one after another, about 170 ms each, and took 6.3 s cold. It now fetches them all at once. In the protocol code, every hash went through the browser's asynchronous digest call, which came to about 27 microseconds per hash in a proof check. Hashing now runs synchronously, about 7 microseconds per hash, which made an inclusion proof check four times faster (0.47 ms to 0.11 ms) and building a 100,000-record log 3.5 times faster (5.3 s to 1.5 s). The slowest step left is the proof library's first setup in a new tab, about 130 ms (150 ms for the first proof check against 18 ms for the next), paid once per visit.
Run them yourself from the kit: node tests/products-smoke.mjs and node tests/products-bench.mjs. The fixtures are rebuilt with node tools/build-products.mjs, which makes fresh demonstration keys and fresh proofs each time. The four circuits are in zk/. Protocol timings, with the speed fixes made today, are on the protocols page.
These are suggested uses ICE could sell. None is built for ICE or running on an ICE system. Every run uses fictional records and keys held by Hive, in the same code your team can download and run offline. Hive never decides, blocks or sits in the path, and holds no ICE data. A record shows what existed, when, and that it didn't change. People decide what it means. Nothing you check leaves this page.