Preserved analysis · Document 01

The strategic assessment

The full feasibility analysis that preceded this research programme, retained unchanged: the thesis, the paths, the scenarios, the risks and the decisions requested. It remains the record of how the direction was framed.

Document 01 was written for internal deliberation before the ATLAS research chapters existed. Its figures are the same internal assumptions used elsewhere on this platform; they are not quotations, valuations or approvals.

The strategic thesis

A question of extension, not reinvention

CoreBon is not proposing to become a frontier AI laboratory. It is proposing to find out whether existing, proven systems can be brought inside its own infrastructure, and what that would be worth.

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 document 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.

This document is

An initial strategic feasibility study
A map of credible paths to capability
A set of measurable success criteria
A request to authorise structured research

It is not

An investment proposal or approved transaction
A vendor selection or evaluation result
A costed, negotiated or finalised arrangement
A commitment of capital of any size

Established foundations

A decade of operating blockchain infrastructure has produced an indexed view of on-chain history, an institutional counterparty network and the discipline required to run systems that cannot fail quietly.

Capability is now contractible

Advanced AI systems can be licensed, integrated and specialised. That converts a frontier research problem into an engineering, evaluation and commercial one — a domain this organisation already inhabits.

Relevance is earned annually

Institutional relevance is not a legacy asset. It is maintained by continuing to build infrastructure that others depend on, and by being early to the next layer rather than late to it.

Evidence standard

A standard of evidence applied to every claim

This document separates what is established, what exists and can be inspected, what is assumed for the sake of reasoning, and what remains genuinely unknown. The distinction is the point.

Established fact

Supported by a reliable source or by internal record. Stated plainly, without extrapolation.

Documented technology

A technology that exists and can be inspected — not a claim about our own capability or readiness.

Assumption

An illustrative scenario or estimate used to reason about magnitude. Not validated, not a forecast.

Open question

A genuine gap, listed openly because the document's credibility depends on naming it.

Decision required

Requires board authorisation. Nothing beyond research proceeds without it.

Verified evidence

Vendor material

Illustrative scenario

Working hypothesis

Evidence required

Research integrity — the four 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.

The 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.

Fig. 02 — Illustrative shape only. No axis values are measured, and no advantage is claimed as proven.

We already hold

Indexed on-chain history
Institutional distribution
Live settlement operations
A decade of operational record

→

We would access

Financial world models
Chain reasoning systems
Agentic execution tooling
Evaluation and assurance

→

We would ship

Institutional research products
Risk and settlement intelligence
Metered API services
Agent-mediated operations

Fig. 03 — Proposed composition. Access to intelligence augments existing foundations rather than replacing them. No partnership or licence described here exists.

Capability landscape

Four domains where licensed intelligence meets our foundations

These are the domains worth evaluating first, selected because they combine external capability with something CoreBon already holds. Select a domain to see what it would enable and what remains unresolved.

Documented technology

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.

What it would enable

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?

Paths to capability

Three routes, assessed 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.

Assessment
Speed to first production capability
Capital efficiency
Control and differentiation
Access to scarce talent
Reversibility of the decision

Indicative analytical judgement — not a measured result. Scores to be re-derived from diligence.

Selected path

Licence an existing system

Contract for access to an already-trained system and integrate it.

Months to a first production integration

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.

To be validated by vendor diligence

On the cost claim

Commentary frequently cites savings of 80–90% against frontier development. This document does not adopt that figure. The magnitude is plausible, but it is only meaningful once a like-for-like total-cost comparison exists — including evaluation, data preparation, integration, governance and periodic re-baselining. Producing that comparison is the first deliverable of the research we are recommending.

Partnership structures identified for evaluation

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

Structures only. No provider has been named, approached or evaluated; no terms have been discussed.

Illustrative 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.

Scenario A — illustrative envelope

Buy information, not capability. Fund diligence, a like-for-like total-cost model and two bounded proofs of concept. No production commitment.

Decision gate

Proceed only if the evaluation protocol produces measurable, reproducible results.

Indicative allocation of envelope

Technical and commercial diligence40%
Two bounded proofs of concept35%
Governance, legal and model risk15%
Independent advisory10%

All envelopes and allocations are illustrative assumptions prepared to test feasibility of reasoning about magnitude. No amount has been approved, requested, quoted or negotiated. They are not forecasts and must not be used for planning until validated.

Proof of concept

What a proof of concept must prove, and how it is allowed to fail

A credible programme defines success before it begins, and defines its stop conditions with equal care.

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

0–90 days

Research authorisation

Vendor shortlist, legal review of terms, published evaluation protocol, and a like-for-like total-cost model.

90–180 days

Two bounded proofs of concept

Chain intelligence and financial simulation tested against the same protocol, with failure cases reported.

180–365 days

Decision and first service

Board decision on partnership or deferral, then a single production service with an approved model risk regime.

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.

Financial sustainability (secondary)

Supporting mechanisms, recorded but not relied upon

These mechanisms matter to how a programme could be funded over time. They are not the reason to proceed, and they are deliberately kept out of the case for investment.

S1Supporting

The Company Token as an access and settlement asset

Within new intelligence services, the existing token can serve as an access right and settlement unit. This is an operational role, not an investment case, and it is not presented here as a reason to proceed.

S2Supporting

Agentic payments and x402-style settlement

Machine-to-machine payment flows allow services to be metered and paid programmatically, opening usage-based revenue that does not depend on human onboarding.

S3Supporting

API and data revenue

Analysis endpoints and indexed data access create recurring infrastructure revenue, and a feedback loop: usage improves our evaluation data, which improves the service.

S4Supporting

Capital recycling

Sustainable surplus could fund continued research. A buyback-and-burn mechanism is one possible instrument among several; it is not a thesis, is not modelled here, and involves no price expectation.

These mechanisms are recorded because they affect how a future programme could be funded over time. They are not the reason to proceed, they are not modelled here, and nothing in this section should be read as a view on token price, market behaviour or speculative return.

Risks and open questions

What could make this direction wrong

A direction is only credible if its failure modes are named as plainly as its advantages.

Vendor concentration and lock-in
Commercial
Model access, pricing and roadmap sit with a third party. Exit provisions and inference portability must be contractual, not aspirational.
High
Unvalidated cost advantage
Financial
The widely repeated 80–90% saving is not established. It depends on scope, retention of evaluation effort and integration cost.
High
Model risk in financially consequential outputs
Regulatory
Confident but unsupported analysis is the primary failure mode. Human review and stated uncertainty are non-negotiable.
High
Data rights over proprietary indexed history
Legal
Grounding a licensed model on our own data may interact with provider terms and with the rights of data subjects.
Medium
Capability and talent gap
Operational
Evaluation, data engineering and model governance skills are scarcer than general engineering skills.
Medium
Reputational overstatement
Strategic
Publicly presenting research as achievement would damage the credibility this direction depends on.
Medium
Provider capability withdrawal
Commercial
Products are discontinued and terms change. Continuity planning is part of diligence.
Medium
Regulatory divergence across jurisdictions
Regulatory
AI and digital-asset rules are converging unevenly; the operating footprint determines exposure.
Medium

Questions this research has not yet answered

Q01Which capability is genuinely defensible for us, and which will commoditise within two years?
Q02What is the true total cost of ownership once evaluation, data preparation and periodic re-baselining are included?
Q03Do provider terms permit fine-tuning on proprietary on-chain data — and who owns the resulting weights?
Q04What governance must exist before any model output informs a customer-fund decision?
Q05Can a licensed system be resold inside our own products without a redistribution licence?
Q06What happens to the service if the provider raises prices, changes models or exits the segment?

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.

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.

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.

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.

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.

Immediate recommendation

Authorise structured research, vendor due diligence and a measurable proof of concept — and defer any capital commitment until the evidence from that work exists.

No approval has been given. This document records the decisions being requested, not decisions taken. The opportunity is plausible and worth testing; it has not been validated, priced or negotiated.

Claim register · preserved

Every substantive claim in Document 01, with the original classification retained.

2
Established fact: 3
Documented technology: 5
Assumption: 4
Open question: 2
Decision required

Established fact
CoreBon operates established blockchain infrastructure with accumulated technical expertise and ecosystem relationships.
Organisational fact, drawn from internal records rather than external measurement.
Documented technology
Advanced 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.
Established fact
Frontier-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.
Documented technology
Financial world-model approaches can produce calibrated, interrogable market simulations.
Demonstrated in research settings. Production readiness and calibration on our instruments remain unverified.
Assumption
Licensed 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.
Assumption
Licensing 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.
Documented technology
x402-style machine payment flows can support metered, usage-based service revenue.
Specification and early implementations exist. Institutional adoption at scale is unproven.
Assumption
Bounded agentic execution can reduce operational latency without reducing control.
Plausible and testable. Depends entirely on authorisation boundaries and audit design.
Assumption
A 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.
Assumption
A 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.
Open question
Which capability is genuinely defensible for CoreBon over a five-year horizon?
Unresolved. Depends on proprietary data advantage, distribution and evaluation depth.
Open question
What 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.
Open question
Do provider terms permit fine-tuning on proprietary indexed chain data?
Unresolved and contract-specific. Requires legal review before technical evaluation.
Open question
Which vendor, price, structure or performance level is selected?
None. No vendor has been evaluated, no price has been quoted and no arrangement exists.
Decision required
Authorise research, due diligence and two proofs of concept.
The immediate recommendation. Requires board approval.
Decision required
Commit capital to a partnership or a production programme.
Not recommended at this stage. Deferred until evidence from D1 and D2 exists.