First, what is Cleopas?
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 explore an extraordinary breadth of evidence, arguments and perspectives through a private conversation that moves entirely at their pace.
The problem
Questions about God, suffering, death, meaning, religion and history are important, but difficult to explore fairly. The information is fragmented across books, academic papers, ancient texts, science, philosophy, archaeology and competing religious traditions. Most people only ever see a fraction of it.
The opportunity
AI makes it possible, for the first time, to bring that depth of research into an ordinary conversation. But fluency is not enough. The system also needs to know how far each piece of evidence is allowed to take the answer.
Why we started with Christianity
We needed a subject difficult enough to expose whether the idea was real or just good language. Christianity was an obvious place to start because its central claims combine faith with claims about real people, places and events in first-century history.
The first test
Rather than begin by asking whether the resurrection happened, we started smaller: what can ordinary historical methods responsibly say happened to Jesus’ body after crucifixion?
Why that question is hard
Even this narrow question reaches Roman law, actual Roman practice, Jewish burial customs, archaeology, New Testament writings, source dependence, human motive, political discretion and competing historical explanations.
The goal was not to prove Christianity or disprove it. The goal was to see whether religious texts could be taken seriously without being automatically believed, whether supposedly neutral sources could be challenged with the same rigour, and whether all of the evidence could be combined without overstating what it proves.
What we have learned already
The first pilot was deliberately attacked by independent scholars. That did not produce a single authorised answer. It produced better rules for how the system should think.
One sentence may contain many claims
“Joseph buried Jesus in his own identifiable rock-cut tomb” is not one historical claim. Joseph’s existence, custody of the body, burial, tomb type, ownership and later identification can each deserve different confidence.
Several texts may still be one evidence line
Matthew, Mark and Luke cannot simply be counted as three independent witnesses where later accounts depend on an earlier source.
Possible is not the same as probable
Archaeology may prove that burial after crucifixion happened. It does not automatically tell us how often it happened or what happened to Jesus.
Silence only matters when evidence should exist
A missing record has weight only after asking whether the event should have produced a record, whether it would likely survive and whether we have actually searched where it should be found.
What this roadmap is trying to prove
Research and engineering do not happen one after the other. They run together. Cleopas enters the hands of real users while the Evidence Landscape is still expanding. Their questions then help decide what the research team works on next.
Technology
Build the machinery that retrieves, decomposes, compares and explains governed evidence.
Research
Build the knowledge underneath the conversation and preserve disagreement rather than hiding it.
Users
Expose what people actually ask in private, where the system fails and which topics deserve depth next.
0 to 3 months
The objective is not to build a broad religion app. It is to prove that the reviewed method can survive an unscripted conversation in software.
Technology
- Encode the burial pilot as machine-readable governed knowledge.
- Implement claim decomposition, claim dependencies and evidence relationships.
- Track source independence, source dependence, confidence, relevance and competing scenarios.
- Enforce citations, provenance and maximum defensible language.
- Use the polished scripted conversations as behavioural regression targets.
Research
- Complete the first gold-standard Evidence Landscape.
- Begin the first 10 real user questions in parallel with engineering.
- Establish repeatable source mapping, evidence-family analysis, countercase review and expert QA.
- Measure where AI can assist research safely and where human adjudication remains mandatory.
Users
- Weeks 6 to 12: private alpha with approximately 25 to 100 deliberately diverse testers.
- Include agnostics, believers with doubt, former believers, wounded users and hostile sceptics.
- Optimise for discovering failure, not sign-up growth.
Topics
- God and existence.
- Suffering and hiddenness.
- Jesus and resurrection.
- Bible reliability.
- Science and evolution.
- Comparative religion.
- What following Jesus actually asks of a person.
3 to 6 months
The product becomes useful before the research universe is complete. Every real conversation becomes a signal about what is missing.
Technology
- Launch closed beta.
- Improve automatic decomposition, evidence retrieval, adaptive depth and adversarial robustness.
- Measure overclaim rate, citation accuracy, unsupported claims, research-gap detection and human correction required.
- Instrument conversations so recurring unanswered questions become visible.
Research
- Keep the first 10 questions intensely hand-curated.
- Use questions 11 to 20 to introduce stronger scientific, historical and philosophical counterarguments.
- Let the system propose candidate subclaims, sources and counterarguments, while researchers approve the governed state.
Users
- Hundreds of users moving toward 1,000 or more meaningful conversations.
- Measure learning from conversations rather than registrations.
- Identify which user groups repeatedly encounter the same unanswered branches.
Topics
- Major existential questions.
- Evolution and cosmology.
- Bible transmission and contradictions.
- Resurrection alternatives.
- Hell, morality and religious violence.
- Other faith claims.
6 to 12 months
This is where Cleopas stops being built only from founder assumptions. Real private conversations begin showing us what people actually need help thinking about.
Technology
- Detect composite claims automatically.
- Map existing evidence and unresolved dependencies.
- Flag weak branches and recurring unanswered objections.
- Build a research-priority engine from actual user demand.
Research
- Expand the specialist network across history, theology, philosophy, science, psychology, textual criticism, archaeology and other faith traditions.
- Build tens of high-demand user questions into hundreds or thousands of governed subclaims.
- Allow AI to do more research preparation while preserving human sign-off for consequential judgments.
Users
- Broader beta through faith networks, referrals, universities, partner cohorts and direct consumer use.
- Observe where people go when no one has told them what they are supposed to ask.
Topics
- Prioritise by user demand, importance, weakness of current coverage, adjacency to governed knowledge and research cost.
- The topic plan becomes increasingly empirical rather than editorial.
12 to 24 months
The North Star is not a static library of approved answers. It is a system that increasingly knows where its own knowledge is weak, while humans retain responsibility for what becomes governed knowledge.
Technology
- Identify questions the system cannot responsibly answer.
- Detect recurring objections and weak evidence families.
- Assist the research team with decomposition, discovery and comparison.
- Continue the sovereignty pathway so product success does not create fatal platform dependence.
Research
- Shift human work from constructing every tree toward verification, adjudication, specialist review and frontier research.
- Keep deeply contested or high-impact questions under intensive human scrutiny.
Users
- The user base becomes an intelligence signal about unanswered questions, not an authority on what is true.
- Conversation behaviour increasingly shapes research priority.
Topics
- Expand across religion, science, philosophy, human existence, theology and historical alternatives according to demonstrated demand.
The flywheel
At the beginning, we decide what Cleopas needs to know. During beta, real conversations begin showing us what it needs to know. Over time, the system helps identify the gaps in its own Evidence Landscape.
1. Research
Build governed knowledge.
2. Conversations
Expose real questions, objections and failures.
3. Priority
Feed the most important gaps back into research.