Evidence Intelligence
Method
What the figures mean, what they do not mean, and where the method stops.
What a score is
Every figure is an evidence confidence: an assessment of how strongly the available evidence supports one specific claim. It is not a probability that the claim is true, and it is not a measure of anyone’s faith.
A reasoning model never returns a score. It returns structured factor assessments — the relevance, directness and independence of specific cited items, and bounded penalty estimates — and application code computes the number from those. Two analyses run months apart by different providers are comparable because the arithmetic is fixed even though the judgement is not.
Scores above 95 require multiple independent primary sources with negligible counter-evidence. Interpretive and theological readings are capped below the top band, because the text underdetermines them by nature however well attested the manuscripts are. A claim is never scored low merely for being supernatural.
Interpretive and theological claims: words, not figures
For interpretive, theological, canonical, prophetic and metaphysical claims, no figure and no probability-style label is shown. A number attached to a reading of a text would be read as the probability that the reading is true, which no evidence can supply. The assessment is stated in the epistemic vocabulary used across AbrahamMoses instead, always relative to the evidence and assumptions on record and never a verdict on whether a tradition is true. The interpretive dimension of every profile is shown the same way.
Given the assumptions in force and the evidence on record, the claim is supported. Change the assumptions and the result may change.
Some of what the claim needs is supported and some is not: for example, the text says it, but the argument that makes it a supported reading does not hold.
The claim is still supported, but on fewer independent arguments or weaker evidence than before.
Under the assumptions in force, no recorded argument establishes the claim. This is a statement about this argument as recorded, not about whether a tradition is right.
The evidence on record is too thin to support or weaken the claim. Missing evidence is not evidence against it.
Qualified scholars read the evidence in materially different ways, and the record does not settle between them.
How the internal certainty ladder is worded for these claims: established or highly probable → supported under selected assumptions; probable or plausible → partially supported; possible or speculative → insufficient evidence; unsupported or contradicted → not established under selected assumptions; any claim on which scholarship is significantly divided → disputed interpretation. The internal figures still drive calibration and are kept with each analysis.
Three dimensions, never combined
Textual confidence, historical corroboration and interpretive confidence are assessed and displayed separately. A passage can be textually secure and historically uncorroborated at once; a claim can have high archaeological correspondence and low interpretive certainty. Collapsing them into one figure would hide the only thing worth knowing.
Dimensions are also bounded by claim type. Archaeology cannot establish the Trinity. Manuscript counts cannot establish whether God exists. Attaching a dimension that cannot bear on a claim is treated as a category error and the evidence item is excluded, not merely down-weighted.
Certainty taxonomy (textual and historical claims)
Strongly supported across the available evidence.
The best explanation, with strong supporting evidence.
The evidence favours the conclusion.
Reasonable, but not sufficiently established.
Cannot be ruled out on the available evidence.
Requires assumptions that are not themselves strongly evidenced.
Adequate supporting evidence is presently absent.
Available evidence materially conflicts with the claim.
The claim is not the kind of proposition historical or textual method can adjudicate. It may still be assessed for canonical grounding, coherence and reception.
Note that unsupported and contradicted are different findings. “Contradicted” requires evidence that actually conflicts with the claim; weak support alone never produces it.
Source classes
Manuscripts, inscriptions, excavation reports, ancient documentary evidence, critical editions.
Peer-reviewed scholarship, academic monographs, major critical commentaries, university publications.
Denominational scholarship, theological institutes, established apologetic and established critical scholarship.
Encyclopedic sources, educational publications, popular academic material.
Advocacy material, popular claims, and material that has not been verified.
Class affects weighting; confessional standpoint does not. Denominational scholarship is weighted by its class like any other source, and its perspective is labelled so a reader can see the shape of the evidence base.
Absence of evidence
Nothing bearing on the claim has been found. Given survival rates, excavation coverage or the nature of the claim, this carries little evidential weight either way.
Evidence would reasonably be expected to survive and to have been found by now, and it has not been. This does carry evidential weight against the claim.
Positive evidence has been recovered and it conflicts with the claim. This is the strongest of the three and is not an argument from silence at all.
The claim is not of a kind that predicts recoverable material evidence.
These three are routinely conflated and are not equivalent. An absence counts against a claim only when the evidence would plausibly survive, the region has been excavated, the predicted footprint is large, the dates are precise and the material culture is distinctive — all five, not on average.
Citations
Every citation is resolved against a curated corpus before it can enter an analysis. A reference that does not resolve is dropped from the evidence ledger and recorded as a flag — never quietly reshaped into a plausible-looking footnote. Where a source cannot be independently verified, the interface says so and does not display it as authoritative.
Sources that descend from a common original are discounted through a dependency graph: two chronicles copying one earlier account are one witness, not two.
Where the method stops
Claims that historical and textual method cannot adjudicate — whether a text is of divine origin, whether an execution accomplished redemption — receive not directly scorable under historical method rather than a low number. Assigning zero to a metaphysical claim because archaeology is silent about it would be a category error dressed as rigour.
AbrahamMoses will say that evidence is strong, that evidence is weak, and that we do not know. What it will not do is assign a truth percentage to a religion.