How FaithGraph knows what it claims

Posture

FaithGraph is Christ-centered, Scripture-first, historically careful, and hospitable across historic orthodox Christian traditions. We confess what the Apostles' and Nicene Creeds confess. Where faithful Christians differ on secondary matters, we describe the responsible views fairly and label any editorial choice we must make. We never write in the first person as God, Jesus, or the Holy Spirit.

Six kinds of content, never blurred

• Scripture: verbatim text from an authorized provider, always with its translation label and attribution.
• Scripture reference: a translation-neutral canonical reference such as John 3:16.
• FaithGraph explanation: our original editorial summary or teaching.
• Historical context: a sourced claim about geography, chronology, culture, or archaeology.
• Traditional interpretation: an identified Christian reading that is not explicit in the passage.
• AI-assisted material: not used on public pages in this phase. See the AI policy below.

Evidence classes

Every relationship in the graph uses exactly one of these user-visible labels:

• Explicit in Scripture: The text directly states this relationship.
• Direct quotation: A New Testament passage quotes an earlier passage.
• Explicit fulfillment: The New Testament explicitly frames this as fulfillment.
• Strong allusion or echo: A substantial textual relationship supported by cited evidence.
• Historic Christian reading: A recognized typological or theological interpretation, not explicit in the passage itself.
• Historical inference: A reasoned conclusion from historical or geographic sources.
• Disputed interpretation: Multiple responsible views exist and are described.

Confidence (high, medium, low, disputed) is stored separately from the evidence class. A direct quotation can still carry a note about how the New Testament author uses it.

Review workflow

Records move through draft, in review, approved, and published. Only published records appear on the public site. A material change returns a record to review. Scripture licensing review and theological review are separate approvals. Until launch, unreviewed records carry a visible Draft badge.

Chronology and geography

Ancient dates are stored as earliest and latest plausible years with a named dating scheme and a confidence level. We do not force uncertain dates into a single year. Places carry ancient and modern names, a precision level, an uncertainty radius, and a source. Disputed identifications are labeled disputed. Reconstructed journey routes are interpretations and will be labeled as estimates.

Genealogies

Matthew 1 and Luke 3 are recorded as two separate lineages. We never merge them into one list, and every genealogical relationship states which list it comes from.

Sources

Open datasets produce candidates, not conclusions. Every candidate relationship needs a human-assigned evidence class and rationale before it can be published. See the source registry.

AI policy

In this phase AI is used only internally: proposing entity matches, drafting candidates for human review, and checking consistency. AI cannot approve or publish anything. No licensed Bible text is placed in prompts or embeddings. A public, graph-grounded question tool is a separate future phase that begins only with publisher and owner approval, and it will cite sources for every claim or decline to answer.

What we will not do

• Invent or silently correct Scripture wording.
• Present an interpretation as an explicit statement of the text.
• Claim archaeological consensus where identifications are disputed.
• Manufacture certainty about dates, authorship, or lineage.
• Use engagement tricks, guilt, or pressure.