01
Evidence over hints
Five labels — confirmed, probable, hypothesis, conflicting, disproved — on every claim. Your commercial tree host stays read-only forever: one GEDCOM seed in, nothing ever written back.
Open Source · MIT Licensed
Research your family like it matters.
LineageKit is a public template for serious family-tree research — an evidence-first methodology, a bounded research-cycle protocol, GEDCOM tooling, and a family archive website you own. Your AI assistant installs it, then works the research with you as a partner.
The Problem
Most family trees are name piles.
Commercial genealogy sites make it easy to click "accept" on thousands of hints — and easy to end up with a 5,000-person tree full of strangers, joined on nothing but a name match. LineageKit is the opposite discipline: every claim carries an evidence label, every search gets logged (including the ones that find nothing), and no two people are ever merged without a documentary bridge. A 200-person tree where every fact is sourced outlives a name pile every time.
A name variant is a search path, not proof of identity.The LineageKit rule
The Practice
Three ideas hold it together.
LineageKit is not another genealogy app or subscription. It is a research practice, packaged so your AI can install it and then work it with you.
02
Bounded research cycles
Work happens in numbered cycles, each with a charter, a report with explicit promote-or-don't verdicts per family line, and a human gate before the next cycle begins. Restraint is recorded as a result.
03
The AI carries the queue
Research questions and next searches live inside the project and are worked by your AI research partner. The hobby never becomes homework on your task list — discoveries surface in conversation.
The Vocabulary
Five words that keep a tree honest.
Confirmed
Supported by strong, consistent evidence.
Probable
The best explanation, but one important proof point is still missing.
Hypothesis
A research lead that has not been established.
Conflicting
Credible sources disagree — and the disagreement stays visible instead of being smoothed over.
Disproved
Evidence rules the claim out. A hypothesis disproved before it contaminated the tree counts as a win.
The One Instruction
You do not install LineageKit. Your AI does.
Open the repository in Claude Code, Codex, or another assistant that can read files and follow a protocol. Then give it one instruction:
Tell your AI“Read BOOTSTRAP.md in this repo and set this system up for me.”
Your assistant surveys your tools, interviews you about your family lines, proposes a system map for approval, then generates a private research project — and helps you run your first research cycle.
What You Get
The whole practice, not just a data format.
Everything runs on plain files you own — CSV tables, markdown reports, a GEDCOM seed. No subscription, no lock-in, nothing your grandchildren's tools won't be able to read.
✓
Evidence-first methodology
A phased research plan, an eight-tier evidence hierarchy, a name-variant protocol, and the dead-end worksheet for breaking brick walls.
✓
Five-table data model
People, relationships, events, sources, and assertions in plain CSV with stable IDs — diffable, greppable, future-proof.
✓
GEDCOM tooling
Dependency-free Python scripts import your tree export into staging (everything marked hypothesis) and audit your direct-line source coverage.
✓
Research-cycle protocol
Charter and report templates, plus optional schedule configs for weekly cycle kickoffs and monthly audits.
✓
Family archive website
A deployable site template (Cloudflare Workers) with branch pages, a research library, a timeline, and a moderated inbox where relatives contribute memories and photos.
✓
A worked example
A fully synthetic family with the classic traps seeded on purpose — a duplicate pair, a records conflict, a disproved claim — so you see the discipline in action before touching your own tree.
Privacy by Design
The template is public. Your family is not.
LineageKit was built by extracting the machinery from a real, active research project — and none of that family's data came with it.
The public repository
Generic forever. Every person in its examples and site template is invented. It contains the installer, the methodology, the tooling, and hard-won production lessons — never anyone's real ancestors.
Your private project
Private forever. Your records, sources, and tree stay in a project you control. The archive site publishes only what you explicitly allowlist — and living people are never published, period.
Start Yours
Your family's story deserves better than a hint button.
Start with the public template. Let your AI set up the practice, then go find the records.
Explore LineageKit on GitHub ↗