A decade of infrastructure.A new era of intelligence.
How an established blockchain organisation can extend its existing foundations into advanced AI — through licensing, integration and partnership, rather than frontier-scale development.
Before reading
How this pack is organised
This is a condensed board edition of the CoreBon publication. It carries the same content, the same evidence labels and the same limits. Nothing here has been strengthened for presentation.
Document control
| Document | 01 — Board Research Pack |
| Status | Preliminary strategic research |
| Scope | Licensing · integration · partnership |
| Audience | Board and institutional stakeholders |
| Distribution | Confidential — not for external circulation |
| Version | November 2026 |
| Companion | CoreBon publication, Document 01 |
Contents
| Executive summary | 03 |
| Decisions requested | 04 |
| Strategic opportunity | 05 |
| Technology landscape | 06 |
| Build-versus-licence economics | 07 |
| Potential partners | 08 |
| Financial scenarios | 09 |
| Feasibility milestones | 10 |
| Evaluation protocol | 11 |
| Twelve board questions | 12 |
| Key risks | 13 |
| Evidence register — established and documented | 14 |
| Evidence register — assumptions, questions, decisions | 15 |
| Sources, references and disclaimers | 16 |
Evidence labels used throughout
Supported by a reliable source or by internal record. Stated plainly, without extrapolation.
A technology that exists and can be inspected — not a claim about our own capability or readiness.
An illustrative scenario or estimate used to reason about magnitude. Not validated, not a forecast.
A genuine gap, listed openly because the document's credibility depends on naming it.
Requires board authorisation. Nothing beyond research proceeds without it.
Research integrity — four rules
- 01Every quantitative statement carries an evidence label.
- 02No vendor, price, performance level or structure is presented as finalised.
- 03Illustrative scenarios are assumptions, never projections or promises.
- 04The 80–90% cost-saving claim is treated as unproven until a like-for-like comparison supports it.
The organisation's legal identity and token symbol are withheld from this document by design. CoreBon is the working name used throughout.
Executive summary
A question of extension, not reinvention
Can an organisation built on a decade of blockchain infrastructure acquire world-class AI capability without funding a frontier-scale research programme of its own?
CoreBon's position is unusual. A decade of operating blockchain infrastructure has produced three durable assets: a proprietary, indexed view of on-chain history; a network of institutional counterparties; and the operational discipline that comes from running systems whose failures are public.
This pack asks how those assets can be extended into advanced AI without attempting to reproduce, at our own expense, work that frontier laboratories already do at a scale we cannot match. The preferred approach is deliberately modest: evaluate licensing, integration and partnership with existing providers before considering anything else.
The purpose is not to maximise short-term financial return. It is to expand technological capability, create services worth paying for, maintain long-term institutional relevance and sustain the capacity to keep investing in technology that benefits the world beyond our own balance sheet.
Access to existing advanced systems may offer a capital-efficient route to new capability. That proposition is plausible, testable, and not yet proven.
Three findings
Capability is now contractible
Providers licence access, fine-tuning and joint development. The scarce inputs shift from training compute to evaluation depth, proprietary data and distribution.
Our advantage sits adjacent
Indexed chain history, institutional relationships and operational rigour are precisely the inputs a licensed model cannot manufacture for itself.
The economics remain unproven
The widely repeated 80–90% saving is not established. It depends on scope, retention of evaluation effort and integration cost.
What this document does not establish
No vendor has been evaluated, no price has been quoted and no arrangement exists. The financial envelopes are illustrative assumptions, not projections or requests. Nothing beyond structured research proceeds without board authorisation.
Immediate recommendation
Authorise structured research, vendor due diligence and two bounded proofs of concept — and defer any capital commitment until those produce evidence.
Decisions requested
What is being asked of the board
Four decisions, ordered by commitment. The first two are the research mandate; the third preserves optionality without obligation; the fourth is a prerequisite for anything that touches customers or funds.
| Code | Decision | Commitment | What it authorises |
|---|---|---|---|
| D1 | Authorise structured research and vendor due diligence | Low — internal resourcing over twelve weeks | Commission a like-for-like total-cost model, a vendor shortlist and a written evaluation protocol, with findings published internally before any commercial discussion. |
| D2 | Approve an evaluation protocol and two bounded proofs of concept | Bounded — defined envelope, defined stop conditions | Two proofs of concept against explicit success criteria, each with pre-agreed kill criteria and a requirement to report failure cases. |
| D3 | Reserve an option for a partnership, not a commitment | None until diligence concludes | Authorise preparation of partnership terms so that a decision can be taken quickly once evidence exists, without binding the organisation today. |
| D4 | Establish AI governance and model risk oversight first | Prerequisite to any production deployment | Define human authorisation boundaries, logging, disclosure standards and review cadence before any model output reaches a customer or a treasury action. |
What is not proposed
- An investment proposal or an approved transaction
- A vendor selection or evaluation result
- A costed, negotiated or finalised arrangement
- A commitment of capital of any size
Sequence
D1 and D2 can run inside the current year on internal resourcing and a bounded envelope. D3 creates no obligation. D4 must be in place before any model output reaches a customer or a treasury action, and no later decision can be taken without it.
Approval of D1 and D2 is not approval of Scenarios B or C. Those remain illustrative and require evidence from the research phase.
Strategic opportunity
Why existing systems may offer the capital-efficient path
Three conditions make licensed access plausible now in a way it was not several years ago. None of them removes the need to test the economics.
Capital concentration at the frontier
Frontier development is a contest of compute and specialised research teams. Entering it is a multi-year capital decision with no guaranteed position at the end.
Capability has become contractible
Providers now licence access, fine-tuning and joint development. The scarce inputs shift from training compute to evaluation depth, proprietary data and distribution.
Our advantage sits adjacent
Indexed chain history, institutional relationships and operational rigour are precisely the inputs a licensed model cannot manufacture for itself.
The asymmetry worth testing
A frontier programme converts capital into a research capability that may or may not materialise. A licence converts capital into a contract — measurable, terminable and inspectable before commitment. The second is not a smaller version of the first; it is a different decision, with a different failure mode.
Analytical judgement, not a measured result. Re-derived once diligence produces real numbers.
What would weaken this case
- A like-for-like total cost that approaches frontier-programme economics once evaluation is included.
- Terms that prevent grounding a licensed system on our own indexed data.
- No capability that remains defensible as general models commoditise.
Technology landscape
Four domains where licensed intelligence meets our foundations
These are the domains worth evaluating first. Each combines external capability with something CoreBon already holds; each carries one question that must be answered before it can be used as a basis for commitment.
Financial world models
Systems that simulate how a market would behave — liquidity, microstructure, stress events — rather than forecasting a single price. The value is in calibrated, interrogable scenarios, not in prediction claims.
Enables: Counterfactual liquidity and stress simulation for treasury decisions · Scenario narratives with stated assumptions and traceable inputs · Risk briefings generated from a shared, auditable model of the market
Open question — Which provider can demonstrate calibrated, auditable simulation on instruments we actually hold — and publish its failure cases?
Bitcoin and Ethereum intelligence
Reasoning over on-chain state: transaction graphs, validator and staking economics, protocol upgrade impact, settlement anomalies. This is the area where our own indexed history is a genuine, non-replicable input.
Enables: Protocol-level research products grounded in our own indexed data · Anomaly detection across settlement and custody flows · Upgrade-impact analysis for institutional counterparties
Open question — Can a licensed system be grounded on proprietary indexed data without breaching provider terms or surrendering the resulting insight?
Agentic execution and payments
Tool-using agents that prepare, simulate and settle operational actions, including machine-to-machine payment flows through x402-style protocols. Autonomy must be bounded by explicit human authorisation.
Enables: Treasury and settlement workflows prepared by agents, approved by humans · Programmatic service purchase and metering between systems · Reduced operational latency without reduced control
Open question — What authorisation boundary, audit trail and rollback path must exist before an agent may initiate a value transfer?
Verification and attestation
Evaluation harnesses, model documentation and attestation of model outputs. Without this layer, advanced AI is a reputational liability rather than an institutional capability.
Enables: Defensible evaluation evidence before any external claim · Model risk records suitable for audit and board review · Optional attestation of outputs for counterparties
Open question — Does attestation add real assurance for a counterparty, or does it only add the appearance of rigour?
Domains are selected for evaluation, not prioritised for investment. No provider has been matched to any domain.
Build-versus-licence economics
Three routes, on the criteria that actually decide
Building, licensing and partnering are not moral choices; they are different capital and time profiles. The comparison below is analytical judgement, to be re-derived once diligence produces real numbers.
| Route | Premise | Timeline | Evidence status |
|---|---|---|---|
| Build frontier in-house | Fund a research programme and train proprietary foundation models. | Multi-year to capability parity | Assumption — no in-house programme has been costed |
| Licence an existing system | Contract for access to an already-trained system and integrate it. | Months to a first production integration | To be validated by vendor diligence |
| Integrate and partner | Co-develop a service line with a provider under a joint structure. | Two to four quarters to a joint service | Assumption — no structure has been negotiated |
Qualitative comparison — 1 weakest, 5 strongest
| Criterion | Build frontier in-house | Licence an existing system | Integrate and partner |
|---|---|---|---|
| Speed to first production capability | 1 | 5 | 4 |
| Capital efficiency | 1 | 5 | 4 |
| Control and differentiation | 5 | 2 | 3 |
| Access to scarce talent | 1 | 4 | 3 |
| Reversibility of the decision | 1 | 4 | 3 |
Why licensing scores where it does
Capital is converted into a contract rather than a research programme. Differentiation comes from proprietary data, domain evaluation and distribution — not from the model itself. Terms, usage rights and exit provisions matter more than benchmark scores.
The cost claim is not established
The widely repeated 80–90% saving appears in industry commentary and is treated here as an assumption. It requires a like-for-like total-cost comparison — including evaluation, data preparation and periodic re-baselining — before it can be used in any board paper.
Potential partners
Structures to evaluate, before any provider is named
Four commercial structures could carry licensed capability into CoreBon. They differ less in technology than in what they grant, what they cost and how easily the organisation can walk away.
| Structure | What it grants | What it costs | Term that decides it | Reversibility |
|---|---|---|---|---|
| Commercial access licence | Metered or committed access to a hosted system through a documented interface | Subscription plus usage commitments | Usage rights, resale scope and exit provisions | High |
| Adaptation licence | Rights to specialise provider models on our own indexed data | Licence fee plus compute; obligations on data handling | Ownership of derived weights and permitted use | Medium |
| Joint development | A co-built service line with shared engineering and shared revenue | Shared investment and revenue share | Intellectual-property boundaries and governance | Medium |
| Embedded / OEM | Intelligence embedded inside CoreBon products under our own brand | Licence plus integration engineering and support | Branding, exclusivity and continuity guarantees | High |
Provider categories indicated by the domains
- Hosted model providers offering access licences and volume terms
- Specialist providers of financial simulation and market world models
- On-chain intelligence providers who work from indexed history
- Machine-payment and settlement infrastructure for agentic flows
Diligence requirements before shortlisting
- Whether terms permit fine-tuning on proprietary indexed chain data — and who owns the resulting weights.
- Whether a licensed system can be resold inside our own products without a redistribution licence.
- What a like-for-like total cost of ownership includes once evaluation and data preparation are counted.
- What happens to the service if the provider raises prices, changes models or exits the segment.
Deliberate omission
No provider is named in this pack. No provider has been approached, evaluated or ranked, no terms have been discussed, and no commercial relationship exists. Naming is a diligence output, not an input.
Financial scenarios
What different levels of commitment would buy
Three envelopes, constructed to test whether the reasoning holds at different scales. They are assumptions, not proposals, and no amount shown has been requested or approved.
Evidence-building
12 months$2–4M
Buy information, not capability. Fund diligence, a like-for-like total-cost model and two bounded proofs of concept. No production commitment.
GateProceed only if the evaluation protocol produces measurable, reproducible results.
Illustrative allocation
Platform extension
24 months$12–25M
Extend an existing product line with licensed intelligence and proprietary on-chain data. The first revenue-bearing service, with a defined evaluation regime.
GateRequires signed terms, a validated cost model and an approved model risk framework.
Illustrative allocation
Institutional AI programme
36 months$40–80M
A multi-domain programme across financial simulation, chain intelligence and agentic settlement, with a dedicated team and possible strategic participation in a provider.
GateBoard decision only after Scenarios A and B have produced evidence.
Illustrative allocation
Envelope ranges are illustrative assumptions constructed for scale comparison. They are not budgets, quotes or forecasts, and they carry no board approval.
Feasibility milestones
The sequence, and how it is allowed to end
A credible programme defines success before it begins, and defines its stop conditions with equal care.
Research authorisation
Vendor shortlist, legal review of terms, published evaluation protocol, and a like-for-like total-cost model.
Two bounded proofs of concept
Chain intelligence and financial simulation tested against the same protocol, with failure cases reported.
Decision and first service
Board decision on partnership or deferral, then a single production service with an approved model risk regime.
Milestones and their gates
| Ref | Milestone | Window | Decision |
|---|---|---|---|
| M1 | Written evaluation protocol published internally | Within 90 days | D1 |
| M2 | Like-for-like total-cost model complete | Within 90 days | D1 |
| M3 | Legal opinion on data rights and resale | Within 90 days | D1 |
| M4 | Two proofs of concept reported, including failures | Within 180 days | D2 |
| M5 | Model risk and authorisation boundaries approved | Before any production use | D4 |
| M6 | Single production service live or programme stopped | Within 365 days | D3 |
Stop conditions
Each proof of concept carries a pre-agreed failure condition. If capability, economics or governance thresholds are not met, the recommendation is to stop — not to extend the evaluation. A programme that cannot be stopped cheaply should not be started.
Evaluation protocol
The criteria that must be agreed before scoring begins
Thresholds are set in advance, in writing, before any provider is assessed. A criterion whose threshold is chosen after the results are known measures persuasion, not capability.
| Code | Criterion | What is measured | Threshold agreed in advance |
|---|---|---|---|
| C1 | Capability | Task-level performance against a pre-agreed internal baseline on defined chain and market analysis tasks | Threshold set in the evaluation protocol before any vendor is scored |
| C2 | Economics | Delivered cost per analysis and per thousand inferences, including evaluation, data and retraining | Ceiling agreed in advance; published as a like-for-like comparison |
| C3 | Reliability | Verified-answer rate, unsupported-claim rate, latency and failure behaviour under load | Failure cases reported, not withheld |
| C4 | Governance | Reproducible outputs, model documentation, logging and a rollback path for every automated action | Auditable without vendor assistance |
| C5 | Commercial | Evidence of demand from institutional users — paid pilots or written intent, not survey interest | At least one paying design partner before production build |
Reporting requirement
Failure cases are reported alongside successes. A vendor result that omits its failure modes is treated as incomplete rather than favourable, and no proof of concept concludes without a written account of what did not work.
Integrity rules applied throughout
- 01Every quantitative statement carries an evidence label.
- 02No vendor, price, performance level or structure is presented as finalised.
- 03Illustrative scenarios are assumptions, never projections or promises.
- 04The 80–90% cost-saving claim is treated as unproven until a like-for-like comparison supports it.
Twelve board questions
The questions this pack cannot answer
Each question is drawn from the evidence register or the capability review. None can be answered from this document alone; together they define the scope of the research phase.
Which capability is genuinely defensible for us, and which will commoditise within two years?
Drawn from — Open question
What is the true total cost of ownership once evaluation, data preparation and periodic re-baselining are included?
Drawn from — Open question
Do provider terms permit fine-tuning on proprietary on-chain data — and who owns the resulting weights?
Drawn from — Open question
What governance must exist before any model output informs a customer-fund decision?
Drawn from — Open question
Can a licensed system be resold inside our own products without a redistribution licence?
Drawn from — Open question
What happens to the service if the provider raises prices, changes models or exits the segment?
Drawn from — Open question
Which provider can demonstrate calibrated, auditable simulation on instruments we actually hold — and publish its failure cases?
Drawn from — Financial world models
Can a licensed system be grounded on our own indexed data without breaching provider terms or surrendering the resulting insight?
Drawn from — Chain intelligence
What authorisation boundary, audit trail and rollback path must exist before an agent may initiate a value transfer?
Drawn from — Agentic execution
Does attestation add real assurance for a counterparty, or does it only add the appearance of rigour?
Drawn from — Verification and attestation
What are the pre-agreed stop conditions, and who holds the authority to stop the programme?
Drawn from — Proof of concept
Which of the four decisions is being approved today — and what is explicitly not being approved?
Drawn from — Decisions requested
Key risks
What could make this direction wrong
A direction is only credible if its failure modes are named as plainly as its advantages. Severity is an initial judgement, to be revised once diligence produces evidence.
| Risk | Category | Severity | Why it matters |
|---|---|---|---|
| Vendor concentration and lock-in | Commercial | High | Model access, pricing and roadmap sit with a third party. Exit provisions and inference portability must be contractual, not aspirational. |
| Unvalidated cost advantage | Financial | High | The widely repeated 80–90% saving is not established. It depends on scope, retention of evaluation effort and integration cost. |
| Model risk in financially consequential outputs | Regulatory | High | Confident but unsupported analysis is the primary failure mode. Human review and stated uncertainty are non-negotiable. |
| Data rights over proprietary indexed history | Legal | Medium | Grounding a licensed model on our own data may interact with provider terms and with the rights of data subjects. |
| Capability and talent gap | Operational | Medium | Evaluation, data engineering and model governance skills are scarcer than general engineering skills. |
| Reputational overstatement | Strategic | Medium | Publicly presenting research as achievement would damage the credibility this direction depends on. |
| Provider capability withdrawal | Commercial | Medium | Products are discontinued and terms change. Continuity planning is part of diligence. |
| Regulatory divergence across jurisdictions | Regulatory | Medium | AI and digital-asset rules are converging unevenly; the operating footprint determines exposure. |
The two that decide the case
Vendor concentration and unvalidated cost advantage are not peripheral. If either cannot be resolved contractually or empirically, the licensing path loses the characteristic that makes it attractive — the ability to stop.
Evidence register
Established, documented, or assumed
Every substantive claim in this pack is recorded here with its classification. The register is the basis on which the reader should judge the rest of the document.
01
Established factCoreBon operates established blockchain infrastructure with accumulated technical expertise and ecosystem relationships.
Organisational fact, drawn from internal records rather than external measurement.
02
Documented technologyAdvanced AI capability can be acquired through licence, integration or partnership rather than in-house frontier training.
Documented commercial practice: several providers offer licensed access, fine-tuning and joint development.
03
Established factFrontier-scale model development requires very large compute and specialised research teams.
Established industry pattern. Specific figures vary by generation and are not relied upon here as forecasts.
04
Documented technologyFinancial world-model approaches can produce calibrated, interrogable market simulations.
Demonstrated in research settings. Production readiness and calibration on our instruments remain unverified.
05
AssumptionLicensed access delivers an 80–90% saving against frontier development.
Not established. This figure appears in industry commentary and requires a like-for-like total-cost comparison before use.
06
AssumptionLicensing plus integration costs roughly an order of magnitude less than a frontier programme of comparable scope.
Illustrative magnitude derived from our own envelope assumptions. Not a quote, not a validated result.
07
Documented technologyx402-style machine payment flows can support metered, usage-based service revenue.
Specification and early implementations exist. Institutional adoption at scale is unproven.
08
AssumptionBounded agentic execution can reduce operational latency without reducing control.
Plausible and testable. Depends entirely on authorisation boundaries and audit design.
09
AssumptionA Company Token can serve as an access and settlement asset inside new services.
Mechanism-level assumption. No price, valuation or market behaviour is assumed anywhere in this document.
10
AssumptionA buyback-and-burn mechanism could recycle surplus into continued innovation.
One option among several. Not modelled, not recommended, and deliberately excluded from the investment case.
Evidence register
Unresolved, and requiring authorisation
What remains genuinely unknown is listed with the same prominence as what is established. Nothing in this section has been resolved by the document that contains it.
01
Open questionWhich capability is genuinely defensible for CoreBon over a five-year horizon?
Unresolved. Depends on proprietary data advantage, distribution and evaluation depth.
02
Open questionWhat is the true total cost of ownership including evaluation and data preparation?
Unresolved. This is the central diligence question and the largest gap in the current evidence.
03
Open questionDo provider terms permit fine-tuning on proprietary indexed chain data?
Unresolved and contract-specific. Requires legal review before technical evaluation.
04
Open questionWhich vendor, price, structure or performance level is selected?
None. No vendor has been evaluated, no price has been quoted and no arrangement exists.
05
Decision requiredAuthorise research, due diligence and two proofs of concept.
The immediate recommendation. Requires board approval.
06
Decision requiredCommit capital to a partnership or a production programme.
Not recommended at this stage. Deferred until evidence from D1 and D2 exists.
Register composition
| Classification | Entries |
|---|---|
| Established factEstablished fact | 2 |
| Documented technologyDocumented technology | 3 |
| AssumptionAssumption | 5 |
| Open questionOpen question | 4 |
| Decision requiredDecision required | 2 |
| Total recorded entries | 16 |
Five of sixteen entries are assumptions. Two are decisions. No entry asserts a finalised price, vendor or performance level.
Sources and references
Basis of evidence, and the limits of this document
This pack quotes no external study by name and relies on no third-party dataset. What follows records the basis on which each class of claim rests, and what must be gathered before any claim is strengthened.
Basis of each class of claim
Organisational facts
Drawn from CoreBon's own operating record: infrastructure history, indexed data, counterparty relationships.
Documented technology
Referenced as commercial or technical practice that exists and can be inspected. It is not a claim about our readiness.
Industry commentary
Referred to only to be rejected as evidence. No commentary figure is relied upon anywhere in this pack.
To be gathered during the research phase
- Provider licence, fine-tuning and redistribution terms, reviewed by counsel.
- Published model documentation and independent evaluation results.
- Machine-payment specifications and their current institutional adoption.
- Like-for-like cost benchmarks from counterparties who have already integrated.
Disclaimers
01This pack is preliminary strategic research and a feasibility assessment. It is not an investment proposal, an offer, a solicitation or a recommendation to acquire any asset, security or service.
02No vendor, price, performance level, partnership or commercial arrangement described here has been finalised, quoted or approved. Nothing in this document constitutes a commitment of capital by CoreBon or by any affiliated entity.
03All financial envelopes and allocations are illustrative assumptions constructed to test the reasoning at different scales. They are not budgets, forecasts, projections or promises, and must not be relied upon as such.
04Quantitative statements carry an evidence label and should be read with it. Figures drawn from industry commentary are classified as assumptions and remain unvalidated until a like-for-like comparison supports them.
05The organisation's legal identity and existing token symbol are withheld from this document by design. CoreBon is the working name used throughout; its use implies no registered entity, listing or instrument.
06Prepared for board deliberation and circulation among institutional stakeholders under confidentiality. Not for external distribution, publication or reproduction in whole or in part without written authorisation.
Document
CoreBon · Board Research Pack · Document 01 · November 2026 · Prepared for board deliberation · Confidential. Companion to the CoreBon publication at the same version. Where the two differ, the publication governs.