Methodology

Every claim cited, or absent.

A dossier shows you exactly where a claim came from, graded for how much weight it can bear.

FIG 01 Citation or drop
The citation-or-drop rule A claim must attach a citation marker that resolves to a real evidence record before it can enter a dossier. A claim with no marker, or a marker pointing at nothing, is dropped rather than shipped. Claim a sentence that asserts a fact Citation marker must resolve to a real record, checked in code Evidence record source, URL, license, retrieval time, and the content an agent may cite No marker, no record: the claim is dropped, not shipped

A claim with no citation does not ship.

How the guarantee works

That is the guarantee behind every Atlas dossier. A dossier does not tell you something is true and ask you to trust it: it tells you something is true and shows you exactly where that came from, graded for how much weight it can bear. If Atlas cannot show its work for a claim, the claim does not appear, or it is flagged as a gap instead of quietly smoothed over.

Why an instrument

Same data, different answers.

The evidence under a liquefaction judgment is largely shared and public; the judgments built on it diverge widely, and rarely show their working.

FIG 02 Three documented divergences
Shared evidence, diverging judgments In each of three studies, one shared target drew widely diverging judgments. Thirteen physical-risk vendors scoring one common portfolio of 100 real properties produced flood damage-ratio estimates whose pairwise correlations ranged from 0.2 to 0.9. Italy's three official seismic hazard models diverge by more than 35% in predicted ground motion at specific locations, and 50 years of records cannot arbitrate between them. Forty-nine experts asked to distribute 100 points of influence across 17 liquefaction susceptibility parameters gave the single top parameter anywhere from 0 to 80 points. One shared target in each study Thirteen vendors, one portfolio flood damage-ratio estimates for 100 identical properties correlations 0.2 to 0.9 Italy's three official models predicted ground motion at specific locations; 50 years of records cannot arbitrate more than 35% apart Forty-nine leading experts asked to weight 17 parameters with 100 points of influence, the single top parameter drew anywhere from 0 to 80 Each divergence is published, measured, and unresolved

The evidence is shared. The judgments are not.

Thirteen vendors, one common portfolio

In an October 2025 study by the GARP Risk Institute, commissioned by the Climate Financial Risk Forum (convened by the Bank of England's Prudential Regulation Authority and the Financial Conduct Authority), thirteen physical-risk vendors scored one common portfolio of 100 real properties; pairwise correlations between vendors' flood damage-ratio estimates for identical properties ranged from 0.2 to 0.9.

Three official models, one country

Italy's three official seismic hazard models (MPS04, MPS19, ESHM20) diverge by more than 35% in predicted ground motion at specific locations, and formal testing against 50 years of records from 143 stations could not statistically distinguish which model is right (Iervolino et al., PLOS One).

Forty-nine experts, one parameter list

PEER Report 2023/02 documents a 2022 workshop of 49 experts averaging 19 years of experience: the field's leaders do not share a common understanding of liquefaction susceptibility, and when asked to distribute 100 points of influence across 17 parameters, the single top parameter received anywhere from 0 to 80 points.

Atlas does not promise the agreement the field itself does not have. Its offer is a judgment you can re-run, ablate, and date: take one apart in the gallery of worked examples.

Liquefaction is a secondary effect of the earthquake primary peril, not a hazard that arrives on its own: it only happens where shaking meets ground already prone to it. That is the relationship every Atlas dossier is built to trace, evidence by graded evidence.

What the guarantee covers

The guarantee, one glance.

Eight guarantees, each pinned to one of these five stages.

FIG 03 Five stages, checked in code
The Atlas pipeline, one glance Connectors feed a content-hashed evidence ledger. Specialist agents turn evidence records into cited claims, quoting them into records that carry source, license, and timestamp. Every source carries a fixed reliability grade. An adversarial reviewer challenges the draft. A fixed battery of deterministic checks gates the prose before it can ship; a violation renders flagged, never silently. The signed judgment comes out the end. No model predicts a hazard number: the LLM only narrates a decision the code made. 01 Gather Connectors geology, subsurface, hazard, and case history sources 02 Quote Evidence ledger content-hashed, with source, license, and timestamp attached 03 Grade A to F, per connector a fixed reliability grade travels with the evidence 04 Challenge Adversarial reviewer reads the draft against its own ledger, then a retry, or a flag Gate Deterministic checks 05 Sign Signed judgment the call and the decisive record, HMAC-signed No model predicts a hazard number The LLM only narrates a decision the code made A fixed battery of checks gates the prose A violation renders flagged, never silently

Every stage is checked in code before a claim can ship.

Thin evidence

Where the evidence runs out.

An outcome is only asserted with confidence when the evidence actually supports it.

FIG 04 Stated, never rounded up
Sufficient evidence versus thin evidence Where the evidence supports an outcome, the dossier states the outcome. Where it does not, the dossier states plainly that the data is insufficient rather than rounding up to a conclusion. The public evidence coverage map shows which case a location is in. Sufficient evidence the outcome is stated, with the graded records that carry it Evidence thin INSUFFICIENT DATA is stated plainly, never rounded up to a conclusion The coverage map states which case a location is in

Thin evidence is stated, never rounded up.

Coverage is uneven by design

An outcome is only ever asserted with confidence when the evidence actually supports it. When it does not, the dossier states plainly that the data is insufficient rather than rounding up to a conclusion. The same honesty drives the public evidence coverage map. Coverage is uneven by design, and we would rather show exactly where we are strong and say plainly where we are not.

Boundaries

What it does not cover.

Three limits that belong on the page, not in the fine print.

FIG 05 Limits stated, not hidden
  • L1

    Atlas is a desk-study screening instrument, not a site investigation.

    What it reasons over instead

    It reasons over evidence that connectors can retrieve remotely: published mapping, catalogued subsurface data, hazard models, case histories. It does not replace a boring, a CPT, or a site visit, and it is not a substitute for site-specific geotechnical engineering judgment.

  • L2

    A citation guarantees traceability, not correctness beyond its stated grade.

    What the grade is actually for

    Citing a source means the claim can be walked back to that source and checked; it does not mean the source itself is error-free. That is exactly what the reliability grade is for: it tells you how much independent weight to put on a source, not just that one exists.

  • L3

    Coverage varies by region, sometimes sharply.

    What the coverage map states

    Atlas can only cite what its connectors can actually retrieve for a location, and that is not uniform across the world today. The coverage map states plainly where that evidence is strong, where it is partial, and where it is not yet there, rather than implying a global reach the connectors do not have.

Three limits Atlas states, not hides.

The division of labour

What Atlas refuses to do.

The model is never asked to choose the outcome, or to produce a number.

FIG 06 Model narrates, code decides
What the model does and what the code does The model does not predict a hazard number and does not choose the outcome. A frozen, founder-signed rule table selects the outcome deterministically, and every numeric value comes from published models and measured data through cited connectors. The model does not predict a hazard number does not choose the outcome It narrates, in plain language, a decision already made The code a frozen rule table selects the outcome every numeric value comes from published models and measured data The same table, every run

The model narrates. The rule table decides.