Verification and Custody Launches Leave Agent Trading Costs Unanswered

Two products launched four days apart in early November 2026: Bitget's Open Custody Network (Nov 5) and Space and Time Enterprise (Nov 9). Each addresses one piece of the trust problem, but neither is an agent dispute-settlement or execution-proof system. The sources reviewed do not give a cost per verified trade or a dispute mechanism for autonomous trading agents; our search was limited.

Executive Summary

Two products launched four days apart in early November 2026. Space and Time Enterprise is covered in an AI-agent context. OCN is not, and the evidence does not support treating them as one "agent infrastructure stack." Space and Time Enterprise (November 9) sells cryptographic proofs that SQL calculations over private data were computed correctly. Bitget's Open Custody Network (OCN) (November 5) lets eligible users and institutions keep assets with external custodians while trading on Bitget. Neither is an agent dispute-settlement or execution-proof system, and Bitget's release does not mention AI agents.

The defensible reading is narrower. Verification and custody products aimed at enterprises and exchanges are being launched. Space and Time's four named launch partners include an onchain asset manager, a vehicle maker, an Oracle consultancy and a data nonprofit; none is described as an autonomous trading desk. OCN is framed as a custody-choice product for market makers, asset managers and other institutions. Each addresses a piece of the agent-trust problem: proving a calculation on private data, and keeping assets outside an exchange.

The open question is the one that decides whether autonomous trading agents can be trusted with real capital. What does it cost to verify an agent's trade, and how is a disputed trade settled? The supplied evidence contains no public cost figure, no proof-generation benchmark, no named OCN custodian and no dispute mechanism. We treat that as the central finding. The sources reviewed show products that verify pieces of the workflow and publish no price for verifying it.

Market Context

On November 10, 2026, Bitcoin was $74,369, Ether $2,100, Solana $85.32 and BNB $689.45 (DefiLlama/CoinGecko). We have no before-and-after price data for either token, so we cannot say whether the launches moved prices.

The settlement base that any agent trading system would use is large. Total stablecoin supply was $293.62 billion on November 10, 2026 (DefiLlama). USDT accounts for $176.21 billion (60.0%) and USDC for $70.29 billion (23.9%). Stablecoins are the settlement asset most agent-payment designs rely on, so the concentration matters for anyone designing agent settlement.

Execution venues for agents are also deep. Chain TVL on the same date was $41.74 billion on Ethereum, $5.72 billion on Base, $5.10 billion on BSC, $5.07 billion on Solana and $2.35 billion on Hyperliquid L1 (DefiLlama). Capital and venues exist; the evidence on the trust layer is thin.

The tokens most directly tied to the two launches are small relative to the incumbents. SXT, the token that powers Space and Time's blockchain, had a market cap of $41.85 million at about $0.009 on November 10 (DefiLlama/CoinGecko). Bitget's exchange token BGB stood at $1.73 with a $1.21 billion market cap on November 10 (CoinGecko). Token value is not a measure of technical merit or adoption, and we do not treat it as one.

The demand side is untested. CoinDesk's March 11, 2026 headline on the x402 agentic-payments protocol read: "Coinbase-backed AI payments protocol wants to fix micropayment but demand is just not there yet." We confirmed only the headline and date, not the article's figures.

Two Launches, Not One Stack

What Space and Time Enterprise is

Space and Time is a Microsoft-backed blockchain data project. It launched Space and Time Enterprise on Monday, November 9, 2026. Decrypt describes it as a platform that lets businesses use private data in AI tools without handing over the underlying records.

The mechanism is a proof. Companies can prove that calculations made on private data are correct. AI applications, auditors and other stakeholders can then check results without accessing the database. The product is built on Proof of SQL, which Decrypt describes as Space and Time's patented technology for proving that database queries were computed correctly.

The company has institutional backing. Space and Time raised $20 million in 2022 in a strategic round led by M12, Microsoft's venture fund. TechBullion reports that the SXT Chain network currently delivers verifiable data to DeFi protocols, stablecoin issuers and tokenized-asset platforms.

Four launch partners are named: onchain asset manager Maple Finance, Indonesian vehicle maker and distributor Indomobil, Oracle software consultancy Trenez, and The Data Foundation, a Washington, D.C. nonprofit focused on government data. Only one of the four, Maple Finance, is an onchain asset manager. The others are a vehicle maker, a software consultancy and a nonprofit; the sources do not say how each will use the product.

What Bitget's Open Custody Network is

Bitget announced OCN on November 5, 2026. Bitget describes it as a network connecting banks and institutional-grade custodians with Bitget's trading infrastructure for eligible users. Users can keep assets with participating external providers while trading on Bitget. The assets are not held directly in exchange custody.

OCN connects external custody to Bitget's spot, margin and futures markets through "asset mirroring." Bitget positions it for market makers, asset managers and other institutions. The stated purpose is to manage custody-concentration risk while deploying externally held assets across trading strategies. The release quotes Bitget's CEO framing it as "Assets in Bank, Tradings on Bitget."

TokenPost's November 7 report describes OCN in the same terms: external custody connected to Bitget's spot, margin and futures markets. It frames OCN as a custody-choice product and does not mention AI agents.

Why the "agent infrastructure stack" framing fails

We read the full body of the November 5 release. It does not mention AI agents. It does not name any participating custodian or bank. It does not describe how asset mirroring settles or how disputes are handled. This check covers only that release, not every Bitget announcement, so we do not claim Bitget has no agent plans. We claim only that the release does not support an agent framing.

The two products also do not form a joint offering. The only tie between Bitget and Space and Time that we found dates to March 27, 2023. On that date Bitget announced a collaboration for exchange financial transparency, and its release called Bitget the first centralized exchange to offer financial transparency through Space and Time. That work concerned verifiable data on exchange assets and liabilities, not agents. Bitget is not among the four launch partners named for Space and Time Enterprise.

A stack implies components designed to interoperate. These two products were announced separately, by different companies, for different users, and they solve different problems. They can be combined in a design, but nothing in the evidence says anyone has done so.

Four Trust Problems, and Which Ones Were Addressed

An autonomous trading agent raises four separate trust questions. The launches touch two of them. A reader who treats "verification" as one thing will overestimate what has been delivered.

Computation correctness. Did the agent's calculation, such as a risk figure, a portfolio valuation or a signal, follow correctly from its inputs? Space and Time Enterprise targets this directly through Proof of SQL.
Data authenticity. Were the inputs real? A price feed, a balance or an order-book snapshot can be wrong or manipulated before any calculation starts. The sources reviewed do not claim Proof of SQL addresses this.
Asset custody. Where do the funds sit, and who can move them? OCN targets this by keeping assets with external custodians outside exchange custody.
Mandate compliance and dispute settlement. Did the agent act within its authority, and who decides when a principal and a counterparty disagree about a trade? Neither launch addresses this, and no source we reviewed describes a mechanism for it.

The table below maps each launch against the four problems. "Addressed" reflects only what the cited sources state.

Trust problem	Space and Time Enterprise (Nov 9)	Bitget OCN (Nov 5)
Computation correctness	Addressed: proof that SQL queries over private data were computed correctly	Not addressed
Data authenticity	Not claimed in the sources reviewed	Not addressed
Asset custody	Not addressed	Addressed for eligible users: assets held with external custodians
Mandate compliance and dispute settlement	Not described	Release does not describe how disputes are handled

The gaps in the table matter more than the filled cells. A trading agent that computes correctly on authentic data, holds assets in safe custody and acts within its mandate needs all four. The evidence shows progress on two of the four problems (computation correctness and asset custody), nothing claimed on data authenticity, and nothing described for mandate compliance and dispute settlement.

What Proof of SQL Proves, and What It Does Not

Proof of SQL is a proof of correct query computation. Given a dataset and a query, it lets a verifier confirm that the stated result is what the query produces over that data. The sources reviewed do not claim it proves the underlying data is true.

That distinction is easy to lose in marketing language. A "verified" output can still rest on bad inputs. If a price feed was wrong, a proof of correct computation over that feed produces a correctly computed wrong answer. This is our inference from the description of Proof of SQL, not a sourced claim, and we flag it as such.

The gap is not a flaw in the technology. A proof of computation is a different object from a proof of origin. Closing the gap requires pairing computation proofs with some mechanism for authenticating inputs. Oracle networks such as Chainlink and Pyth are commonly used to deliver market data onchain; this is background context, not a sourced finding of this report.

For trading agents the gap is practical. Agent decisions depend on live market data and on account balances. Both come from outside the proven database unless an authenticity layer is added. A verifier can confirm that the agent computed its position size correctly from the numbers it was given. The verifier cannot confirm from the proof alone that those numbers matched the market.

The launch partners illustrate where the product fits today. An auditor checking a calculation over a company's private records is a natural use case, though the sources do not say which partners will do this. The data sits in a database the company controls, and the question is whether the reported result follows from it. Even there, the proof says nothing about whether the records were honestly entered. It narrows the audit question to data integrity. It does not remove it.

Nate Holiday, CEO of MakeInfinite Labs, the original developer of Space and Time, told TechBullion that businesses want to put their data to work in AI and other applications, and that they need to protect proprietary information, verify their data and audit their agents. The product as described lets companies prove calculations over private data without exposing it; the sources do not say it authenticates the underlying data.

Custody Separation Is Not Mandate Enforcement

OCN is aimed at one specific risk. Bitget positions the structure as a way to manage custody-concentration risk. For institutions deploying capital across strategies, reducing concentration is a real benefit.

It is a different benefit from the one autonomous agents need. Custody separation answers "where are the funds?" An agent's principal also asks "did the agent do what I authorized?" Moving funds to a bank custodian does not make an agent's trades conform to a mandate. An agent with trading access to mirrored assets can still trade badly or outside its limits, and a custody arrangement alone does not prove otherwise.

The release does not frame OCN as an agent product, so any agent use would be an extrapolation. The extrapolation is not absurd. A design where an agent trades on an exchange while the principal's assets remain at a bank has an obvious appeal. But the release does not describe agent permissions, trading limits or any control over what a connected strategy can do. We cannot assess a design the source does not describe.

Several operational questions are also unanswered, and the release does not address them. It does not say how asset mirroring settles, who the participating custodians and banks are, which jurisdictions and eligibility rules apply, or what happens in a dispute or a custodian failure. These are the questions a risk committee would ask first. Until they are answered, OCN is an announced network with unnamed members.

The concentration point cuts both ways. OCN moves trust from the exchange to the custodian and to the mirroring mechanism. Mirroring is the new component. If the mirrored balance on the exchange diverges from the custodied asset, someone must reconcile it. The release gives no account of who that is or how it is done.

The Unit-Economics Question

The break-even test

Verification is a cost, and it pays only when it costs less than the loss it prevents. In simple terms, verifying a trade is rational when the cost of verification is below the probability of an undetected error multiplied by the loss it would cause. This is our analytical framing, not a sourced claim.

The test has practical consequences. For a large, infrequent, high-value trade, even an expensive proof can be worth it. For a high-frequency strategy making many small trades, per-trade verification may exceed the value at risk. The economics favor selective verification: proving the calculations that govern position limits and risk, and sampling the rest.

The test also applies to disputes. A settlement mechanism has its own costs, including evidence, time and arbitration. If resolving a disputed agent trade costs more than the trade, principals will not use it. They will instead limit agent authority so that no single error is large. That is a plausible equilibrium, and it would shape how much capital agents are allowed to control.

What has not been published

We found no verified numbers for the quantities the break-even test needs. The sources reviewed give no proof-generation time, no GPU or hardware requirement, no per-query cost and no onchain verification gas for Proof of SQL. They give no audit-overhead figure for any agent verification approach. They give no pricing for Space and Time Enterprise.

On the custody side, the release gives no fee schedule, no settlement time for asset mirroring and no description of custodian charges. We cannot say whether OCN is cheap or expensive relative to holding assets on exchange.

We also have no verified data on the risk side. We found no sourced incident data on agent trading losses, exploits or hacks, and no verified measure of AI-agent trading volume. Without a loss rate or a volume, both sides of the break-even test are unknown. This is the central gap, and it is why the question stays open.

Demand signals

The one sourced demand signal is cautionary. CoinDesk's March 2026 headline said demand for the x402 micropayment protocol was "just not there yet." We read only the headline, so we draw no figures from it.

The launch partners are mostly non-trading firms, which is consistent with enterprise use, though the sources do not say what any partner will verify. Maple Finance, an onchain asset manager, is the closest to the trading world. Even so, the sources do not say what Maple will verify or at what scale.

That is a reasonable place for a verification product to start. Enterprises with audit needs are a plausible early market, though no source we reviewed shows what they will pay. Autonomous agents controlling their own capital are a more speculative market. The partner mix is consistent with the first and offers no support for the second.

Market Data: Verification Tokens and the Settlement Base

Specialist verifiable-data tokens are small next to the incumbent oracle networks. The table compares them using DefiLlama and CoinGecko data as of November 10, 2026.

Asset	Role	Price	Market cap
SXT	Space and Time token	about $0.009	$41.85 million
LINK	Chainlink oracle network token	about $10.50	about $8.1 billion
PYTH	Pyth oracle network token	$0.0656	$520.29 million
BGB	Bitget exchange token	$1.73	$1.21 billion

LINK's market cap is roughly 190 times SXT's ($8.08 billion against $41.85 million). That gap reflects the scale of the incumbent oracle business, not a verdict on Space and Time's technology. SXT has a different function: proof of computation rather than price delivery.

DefiLlama lists Chainlink's TVL at $1.84 billion and Pyth's at $0. DefiLlama's TVL for oracle protocols is not a measure of total value secured, so these figures cannot be compared as a security metric. We include them to prevent a common misreading. An oracle's TVL tells you little about how much value depends on its data.

Oracle networks deliver market data onchain, which is relevant to the data-authenticity problem that Proof of SQL does not claim to solve. A complete verified-agent design would likely draw on both a computation proof and an authenticated data source. The evidence does not show that any such combination has been built or priced.

BGB's market cap should not be read as a measure of OCN's prospects. It reflects Bitget's exchange business broadly. We have no evidence that the OCN announcement moved it.

Evidence Ledger: What Is Known and What Is Not

The ledger below separates what the sources establish from what remains unpublished. A reader deciding whether to build on these products should treat the right-hand column as the due-diligence list.

Topic	Established by sources	Not published in sources reviewed
Space and Time Enterprise launch	Nov 9, 2026; built on Proof of SQL; four named partners	Pricing, proof time, hardware needs, per-query cost, onchain verification gas
What the proof covers	Correct computation of SQL queries over private data	Any claim that the underlying data is authentic
Bitget OCN launch	Nov 5, 2026; external custody linked to spot, margin and futures via asset mirroring	Named custodians, eligibility rules, jurisdictions, fees
OCN settlement and disputes	Not described in the release	How mirroring settles; what happens in a dispute or custodian failure
AI agent link	None in the OCN release or TokenPost report	Any Bitget agent product (not searched beyond the release)
Agent trade verification cost	Nothing sourced	Cost per verified trade; audit overhead
Agent dispute mechanism	Nothing sourced	Any standard or process for autonomous trading agents
Agent trading losses or volume	Nothing sourced	Incident data; trading volume

The ledger shows that the missing information is concentrated where the thesis matters most. The launches are documented. The economics and the dispute process are not. That asymmetry is the finding.

Risks

Narrative conflation (High). The market may price OCN or SXT as agent-infrastructure plays when the primary sources do not describe them that way. The trigger is promotional framing that merges the two launches. The consequence is disappointment when neither delivers agent-specific features, and mispriced expectations for builders who assume a joined-up stack exists.

Verified outputs on unverified inputs (High). A correct proof over bad data yields a trusted wrong answer. The trigger is any deployment that proves computations over price feeds or balances without authenticating them. This risk is inferred from the description of Proof of SQL and is most acute where "verified" labels encourage users to relax other checks.

Verification cost exceeds risk removed (Medium). If per-trade proof or audit costs are high relative to trade size, high-frequency agent strategies will not verify. The trigger is a published cost figure that is large compared with typical position risk. Today the risk is unquantifiable, because no cost figure exists.

Custody opacity (Medium). OCN names no custodians and describes no settlement or failure process. The trigger is a custodian problem or a mirroring mismatch before the process is documented. Bitget positions custody separation as reducing custody concentration, but the new dependency on custodians and the mirroring mechanism is not yet disclosed.

Thin token liquidity and attention (Medium). SXT is a small-cap token, and we have no liquidity data. Price may be sensitive to headlines. The trigger is launch-driven speculation that fades. Holders should not read price action as adoption evidence.

Thin sourcing (Medium). Our account of Space and Time Enterprise rests on Decrypt and TechBullion. Our account of OCN rests on Bitget's press release and TokenPost. We reviewed no technical documentation for either product. The trigger is any technical disclosure that contradicts the press descriptions, which would require revising this report.

Weak agent demand (Medium). The March 2026 x402 headline said demand was not there yet. The trigger is continued low usage of agent-payment and verification products. This would stretch the timeline for any agent-specific market to mature.

Outlook

Over the next three to six months we expect incremental enterprise adoption and little resolution of the agent questions. The probabilities below are our estimates, based on the thinness of the published evidence and the enterprise orientation of both launches. They are not forecasts derived from sourced data.

Base case (55%). Both products continue as described. Space and Time announces additional enterprise or auditor users. Bitget may add detail on OCN. No public cost-per-proof benchmark and no agent dispute-settlement standard emerges. The agent-trust question remains open, and "verification" stays an enterprise-audit story. This is the most likely outcome because both launches are aimed at institutional or enterprise users, and because neither company has yet published the data that would change the picture.

Upside case (20%). One of the vendors publishes concrete numbers: proof-generation cost or time for Proof of SQL, or named custodians and a settlement description for OCN. Or a third party documents a working combination of computation proofs, authenticated data and external custody. That would move the discussion from concept to pricing. We give it a smaller weight because cost data was absent at launch.

Downside case (25%). Marketing conflation outruns evidence. The agent framing spreads, a public failure or a clear gap between "verified" labels and real data integrity emerges, and sentiment turns against the category. Alternatively, weak agent demand, as the x402 headline suggested, delays any agent-specific use. We weight this above the upside because the High-rated risks could materialize without any new disclosure.

The three cases total 100%. The base case, with no resolution of the open questions, is the most likely.

For traders. Do not treat SXT or BGB moves as evidence of agent adoption. The sourced facts tie neither token to an agent product. SXT is small relative to the oracle incumbents, and we have no liquidity data. Watch for primary disclosures, not secondary framing.

For builders. Treat the products as components. If you build an agent that needs verifiable computation, Proof of SQL is a candidate for the calculation step. You still need an authenticated source for inputs, a custody design you can inspect and a mandate-enforcement layer of your own. Ask both vendors for cost and settlement documentation before committing, because the public record does not contain it.

For investors and allocators. The distinction between enterprise-audit demand and autonomous-agent demand matters for sizing. The launch partners support the first. Diligence questions are specific: What is the cost per proof? Who are OCN's custodians? What happens in a disputed trade or a custodian failure? Until those have answers, assign no value to the agent narrative.

What to watch. Three disclosures would change our view. First, any published proof-cost or latency figure for Proof of SQL. Second, named OCN custodians with a description of how asset mirroring settles and how disputes are handled. Third, any agent-specific dispute or settlement standard from a credible source. We will update this analysis when any of the three appears.