Frequently asked questions

A free, openly licensed way of writing down family-history evidence so that every fact keeps its source and an honest confidence score, and conflicting evidence is kept rather than deleted.

No. MAXGEN is what software uses underneath. What you notice is the behaviour: a tree that shows two possible fathers with their scores instead of silently picking one, a search that finds "Makeswell" when you typed "Maxwell", and a document that is never "corrected" away from what the clerk wrote.

Yes, and keeping the candidates is how you get there honestly. Real research often ends at "probably this man, possibly that one". A format that can hold only one answer has no room for the uncertainty a good genealogist would write in the margin. MAXGEN stores the margin note as data, so a new record can raise one candidate and lower another instead of overturning a false certainty.

They are anchored, not decorative. 0.95 and above means the fact appears verbatim in a source with no ambiguity. 0.70 to 0.95 means well attested with some ambiguity, such as a self-reported census age. 0.50 to 0.70 means plausible but needing more evidence. Below 0.50 is speculative, recorded so it is not lost. Scores must be honest: a consistent 0.4 is more useful than an inflated 0.9.

The document record (MaxRecord) keeps the spelling exactly as written. The interpretation ("that is the Maxwell family") lives separately in the person record, with a score and a source. And the name record (MaxName) can state, with citations, that Makeswell is a known variant of Maxwell in Scots records between 1600 and 1750, so searches find it.

MAXGEN is not a replacement for GEDCOM as a way to move a finished tree between programs. It is a way to exchange the evidence and reasoning behind a tree. Its event types map to GEDCOM tags, so conversion tools can go in both directions; the difference is that a MAXGEN record can say "we are 0.6 sure" and a GEDCOM record cannot.

A record about a person who may be alive is locked private by the schema itself: public systems answer "not found". DNA records are always private and can never hold raw genetic data. Research tasks about living people can never be sent to outside volunteers or third-party AI services. These are rules of the standard, not settings an application can turn off.

It can try, and the standard is built to catch it. Every claim must cite a source record. Every research task lists the evidence it must return (a link to the actual birth record, the exact search used). Results are reviewed by someone other than the contributor, and "fabricated or hallucinated" is a formal rejection reason that lowers a contributor's standing over time.

No. The standard is free under a Creative Commons Attribution licence (CC-BY 4.0): use it for anything, including commercial products, as long as you credit OpenGenealogyAI. The schemas, the specification, and the podcasts on this site cost nothing.

JSON, described by JSON Schema (draft 2020-12). Seven schema files, one per record type, all at the same version number. Records are plain JSON documents; the conventional file extensions are .maxgen for persons and .mxrecord for source documents, but any JSON file that validates is a MAXGEN record.

At their canonical addresses, which are also each file's $id: https://opengenealogyai.org/schemas/maxgen/v1/<name>.schema.json. Frozen copies of every release live under /schemas/maxgen/v1.NN/. See the schemas page and versions.

Any JSON Schema 2020-12 validator works: Ajv in JavaScript, jsonschema in Python, and equivalents in other languages. The schema repository ships validators and fixtures. Or paste a record into the browser validator on this site.

Yes, in extensions{}, which every schema has and which accepts anything. The core of each schema is closed (unknown top-level fields are rejected) so validation is exact; the extensions object is open so you never need to fork. Your records stay valid across releases.

Not within a major version. Minor releases are additive only: new optional fields, new enum values, loosened constraints. Fields are never renamed or removed in v1.x. A breaking change means v2, a new address space, a migration guide, and a 30-day public comment period first. Read the stability promise.

So there is never any doubt about which files go together. When any schema changes, all seven are stamped with the new number and released as a set. It is the model HL7 FHIR and GEDCOM use. Before v1.3 the schemas versioned independently and the skew was confusing; lockstep ended that.

Few, deliberately. A MaxPerson needs an ID, at least one name assertion, and who asserted it and when. A MaxRecord needs an ID, a record type, a licence, a source URL, an extraction confidence and the living flag. Each schema page lists its fields with the required ones marked.

Three. A record with is_living: true is tier2-private and must not be exposed publicly. MaxDNA records are always tier2-private and raw_genotype_stored is always false. MaxTask contributor_eligibility defaults to first_party_only and any task about a living person must stay that way.

The schemas store the evidence; the engine does the arithmetic. The reference approach is a noisy-OR combination weighted by source quality, which raises confidence when independent sources agree without ever reaching certainty. Because every assertion cites its source_record_id, an engine can detect two assertions that really come from one document and avoid double-counting.

Not yet. The schema repository contains JavaScript validators and fixtures. Packages for npm and PyPI are planned after launch; if you publish one, tell us and it will be listed.

Open a GitHub Issue with the problem, the proposed fields, the backward-compatibility answer and a before-and-after example. Minor changes get at least a week of open discussion, major ones a month, then a written decision. Details on the governance page.

Only to publishing records that validate against the public schemas and to honouring the three privacy rules. There is no membership, no fee, no agreement to sign, and no obligation to use the whole standard: an archive that only wants to publish transcriptions with provenance can use MaxRecognition alone.

OpenGenealogyAI holds the copyright and licenses it to everyone under CC-BY 4.0, which means anyone may use, copy, change and redistribute it with credit. The canonical version is stewarded by its author today, with a published path to an advisory board that includes an archive or library seat and can overrule the steward on major versions by a two-thirds vote. See governance.

Yes. MaxRecord has archive_ref for your reference, repository and collection for where the item sits, and extensions{} for anything else. MaxSource's access{} block records how you want to be accessed, including crawl etiquette that robots.txt cannot express.

Every record declares its own licence (CC0, public domain, CC-BY, CC-BY-SA, or never-redistribute) and carries an attribution{} object for the credit line your licence requires. The standard itself is CC-BY 4.0; your data stays under whatever terms you set.

That is what MaxRecognition is for. Each transcription gets a provenance record: who read it, their credentials and specialties, how confident they were word by word, and which readings were direct and which were inferred. The work becomes citable and its quality trackable, instead of being flattened into a text field.

It stands for the Maxwell Genealogy Standard, after its author Garlon Maxwell. The name is a mark for the canonical version: use it for anything that validates against the public schemas, but not for a fork that diverges.

OpenGenealogyAI is the organisation that stewards MAXGEN. The Probable Pedigree is a genealogy platform built on the standard that shows what it makes possible. The standard is designed to outlive any product, including that one: it has no special standing in the change process.

Because evidence, interpretation, work, names, sources, transcription and DNA are different things with different privacy needs and different owners. Separating them lets a document be re-read without touching conclusions, lets DNA be locked down harder than everything else, and lets an organisation adopt only the layer it needs.

They were produced with an AI narration tool from scripts written by the standard's author and describe MAXGEN v1.11. Where a later release changed a detail, the schema pages on this site are authoritative. Each episode has a written study guide.

They hold the records and the communities; MAXGEN is a format for describing evidence that any of them, and any tool built around them, could read and write. It is not a competing service and it does not replace anyone's database. A MaxPerson can carry scored links to the same person on FamilySearch, WikiTree or Find a Grave, so research done in different places can be reconciled instead of duplicated. The aim is a shared language, and the more sites speak it the more useful everyone's work becomes.

No. GEDCOM moves finished trees between programs and has done so for forty years. MAXGEN carries the evidence and reasoning behind a tree, which GEDCOM was never designed to hold. Event types map to GEDCOM tags so conversion runs both ways, and a tool can happily support both.

Every release is recorded in the changelog with the reason it happened. Proposals are discussed in public on GitHub. A mailing list is planned.