Build the evidence. Get to users early.

Life’s biggest questions are too important to be reduced to one person, one tradition or one model answer.

Cleopas is building a new way to explore life’s biggest questions about faith and existence. It combines deeply governed research with AI so one person can privately explore an extraordinary breadth of evidence, arguments and perspectives through a conversation that moves at their pace.

Questions such as Does God exist?, Why does suffering exist?, Can I trust the Bible?, Did Jesus rise from the dead?, Has science made religion unnecessary?, and Why Christianity rather than another religion? are not hard because information is absent. They are hard because the information is scattered across different disciplines, worldviews and source traditions.

A pastor can know theology but not Roman law. A Roman historian can know imperial practice but not textual criticism. A philosopher can understand an argument about consciousness but not first-century archaeology. General AI can discuss all of them, but fluency is not the same as evidence discipline. Cleopas exists to bring the relevant angles into one place and then control how far each piece of evidence is allowed to carry the answer.

A private, deeply researched conversation that does not need to choose the conclusion for you.

The user should experience something simple. Ask a difficult question. Receive a clear answer. Push deeper if you want. Ask for the strongest opposing explanation. See the evidence. See where it stops. Make up your own mind.

Underneath that conversation is an Evidence Landscape. Claims are broken apart, evidence is linked to the specific proposition it supports, source dependence is tracked, competing explanations are preserved and the system records the strongest wording the evidence permits. The conversation adapts to the person. The governed evidence state does not.

We chose a subject difficult enough to expose weakness in the method.

Christianity was not selected because the system is allowed to assume Christianity is true. It was selected because Christianity combines faith with claims about real first-century people, places and events, and because those claims have been defended, attacked and reinterpreted for nearly two thousand years.

The first deep pilot asked a narrow question: what can ordinary historical methods responsibly say happened to Jesus’ body after crucifixion? That apparently small question forced Roman law, Roman practice, Jewish burial norms, archaeology, New Testament texts, source dependence, human motive and competing historical scenarios into the same analysis.

The aim was not to prove the resurrection. The aim was to test whether religious texts could be examined seriously without simply believing them, whether non-religious sources could be examined without automatically privileging them, and whether a system could combine all of that evidence without claiming more than the total record allows.

Independent scholars were asked to attack the work, not endorse it.

The external review has already changed the product architecture. It reinforced the need to separate several sources from several independent witnesses, to split composite claims into smaller propositions, to distinguish a textual fact from event historicity, to treat contextual possibility differently from evidence of a specific event, and to use arguments from silence only when we can first show that evidence should reasonably have existed.

This is important commercially because the research is not decorative content. Each methodological correction becomes a product control, a data field, a research rule or an evaluation test. The stronger the research process becomes, the harder it is for the conversational layer to overstate a case.

People sit on overlapping scales of belief, doubt, curiosity, hurt and hostility.

Cleopas is not designed around rigid personas. A believer can be wounded. An atheist can be uncertain. An agnostic can be deeply intelligent and hostile to apologetics. A former believer may want evidence without any attempt to turn their experience into a faith lesson.

The early product will deliberately test conversations with agnostics, wounded or former believers, believers with hard questions and atheists or sceptics. The evidence underneath each conversation remains the same. What changes is pace, depth, language and the amount revealed at once.

For the first time, depth of research and depth of conversation can exist together at scale.

The enabling shift is not simply that AI can answer questions. It is that digitised scholarship, instantaneous retrieval, long-form dialogue, source tracing and tone adaptation can now exist inside one conversation.

That creates a new category. AI makes a depth of research that was previously impractical for an ordinary conversation available in an ordinary conversation. The product opportunity is to make that capability trustworthy enough for the questions where confidence and bias matter most.

Research and technology are parallel production systems.

Engineering builds the machinery that retrieves, reasons, cites, evaluates and converses. Research builds the governed Evidence Landscape that gives the machinery something reliable to reason over.

The first questions are deliberately expensive and hand-curated. Over time the system should increasingly assist researchers by decomposing claims, proposing source families, identifying gaps, comparing scenarios and highlighting repeated unanswered user questions. Human researchers remain responsible for what becomes governed knowledge.

The critical scaling metric is not simply how many topics exist. It is whether human research hours per governed topic fall while evidence quality, traceability and overclaim performance remain stable or improve.

Get into users’ hands early enough that real conversations can change the roadmap.

Speed matters. The plan is not to research hundreds of topics before anyone can use the product. Engineering and research begin in parallel, a narrow governed product reaches private alpha within months, and paid users are recruited specifically to spend meaningful time trying to use and break it.

The first 1,000 intensive beta participants are not vanity sign-ups. The working assumption is roughly eight to ten hours of use across one to two weeks, plus structured feedback, interviews and question-by-question evaluation. At a planning honorarium of roughly US$1,000 per participant, the programme is intentionally expensive because it buys depth of learning rather than shallow traffic.

The first cohorts can come from Alpha-style programmes, church networks, believers referring friends, former believers, deliberately recruited sceptics, universities and other communities already engaging with questions of faith and meaning. The product should learn what people ask in private before the founders pretend to know the long-term content roadmap.

The first 50 questions become both product content and the training ground for the research system.

The first 10 questions are broad human questions likely to be asked by agnostic, curious or wounded users. Questions 11 to 20 introduce sharper counterarguments and technical objections. The next 30 broaden the Evidence Landscape across Jesus, scripture, suffering, science, philosophy, comparative religion and Christian practice.

A planning assumption of US$1,000 per expert review, with three to five independent reviewers on each of the first 50 questions, implies US$150,000 to US$250,000 before premium specialist work. A separate reserve is needed for highly recognised scholars, complex specialist questions and repeat review when the methodology changes.

Internal researchers do not replace specialists. They make specialist time efficient. Their job is to assemble sources, decompose claims, maintain provenance, tag evidence, identify conflicts and prepare work for external attack. Expensive scholars should spend their time on the judgments that actually require them.

One cheque should remove fundraising risk from the period when Cleopas needs to learn fastest.

The previous operating plan modelled A$2.3 million in year one and an A$9 million five-year philanthropic ask. The updated strategy is intentionally more aggressive. It front-loads engineering, research, scholar review and paid user testing, with the objective of reaching market faster and reducing the chance that a second fundraising round becomes a dependency.

Recommended capitalisation: A$15 million in a single commitment. Planning range: A$12 million to A$15 million. The recommendation is the upper end because the expected lead funder is a one-cheque funder and because unused capital can remain protected as runway rather than being spent simply because it was raised.

The core principle is simple: capital should buy speed, independence and evidence. It should not buy premature scale.

Front-load the expensive learning.

The current working model deploys approximately A$9.05 million across the first 24 months. That is intentionally higher than the old plan because the revised strategy pays for speed: more engineers, a real internal research function, external review of the first 50 questions, and approximately 1,000 intensive paid beta participants.

An A$15 million capitalisation therefore does not imply A$15 million of immediate spending. The remaining capital stays as protected runway for years three to five, sovereignty work, unplanned specialist research, additional user cohorts and downside protection if supporter revenue develops more slowly than expected.

The first 24 months should not depend on revenue to survive.

The historical plan assumed a supporter model with an average contribution of A$10 per month and eventual break-even from a large recurring supporter base. That remains a useful benchmark, but the updated plan should treat it as a hypothesis to test rather than a promise.

The first two years should test three revenue surfaces in parallel: recurring supporters, aligned institutional partnerships or licensing, and donor-funded free access for users who should not face a paywall. The exact mix should be determined by real retention and willingness-to-pay behaviour, not by an early spreadsheet.

The illustrative five-year scenario below is not a forecast. Its purpose is to show that an A$15 million single capitalisation gives Cleopas enough room to learn even if revenue arrives later than hoped.

Illustrative planning scenario only. It is not a revenue forecast and should be replaced as beta produces real retention, contribution and institutional demand data.

The capital does not disappear into ephemeral model spend.

A substantial share of the funding creates durable intellectual property: a structured Evidence Landscape across major questions, atomic claim maps, source and provenance relationships, scholar disagreement records, tagging ontology, reviewed evidence summaries, evaluation sets, behavioural training conversations and a repeatable method for turning contested knowledge into governed AI output.

That asset is useful beyond one consumer interface. It can support education, seminaries, publishers, research organisations, other faith projects and potentially other high-disagreement knowledge domains. The research method itself is designed to be portable.

This should not be presented as a guaranteed A$15 million liquidation value. The stronger and more defensible point for a mission-led funder is that even an unsuccessful consumer outcome can leave behind a substantial, reusable research and governance asset rather than a failed app and a pile of model invoices.

A fast, governed learning loop.

Within months: a real narrow product in private alpha. Within six months: closed beta with real evidence about user behaviour and failure modes. Within twelve months: a meaningful governed question set, a functioning internal research operation, system-assisted research tooling and enough conversation data to show what users actually need Cleopas to know next.

By twenty-four months: the company should be able to demonstrate that real conversations drive research priorities, that external experts can attack new topics without collapsing the methodology, that research cost per topic is falling, and that the conversational system can remain warm and natural without outrunning the governed evidence underneath it.

Take the plan with you.