Published authority data.
Normalized Safety Gate, CPSC or comparable official records. The source's publication establishes the record; 4SI does not add unreported facts.
PUBLIC METHOD / VERSION 1.0
Every position should lead back to evidence.
The Presence Risk methodology keeps source fact, 4SI inference, event scoring and economic modeling visibly separate. Missing evidence widens uncertainty; it never becomes an invented fact.
Risk begins when a claim becomes authority.
Presence Risk is the exposure created when a system accepts a claim about a person, object or agent as sufficient authority for a consequential action without establishing the relevant physical or accountable presence.
THE UNIT OF ANALYSIS
A documented chain from digital claim → asserted presence or identity → granted or attempted authority → consequential action → real-world impact or material exposure.
One library.
Four different claims.
The public library does not treat normalized authority records as if each had a fully researched Claim → Authority → Action → Consequence chain.
Complete analytical chain.
A curated case where primary evidence supports all analytical stages. Validation is source validation, not independent forensic verification.
Institutional depth.
A validated incident selected for source quality, explanatory value, domain balance and institutional relevance.
Scenario range.
A transparent separation between documented monetary floor and modeled consequence. It is not a documented loss sum.
Evidence quality is not
event severity.
Evidence Assurance evaluates provenance, directness and traceability. It does not claim absolute truth, independent forensic verification or legal liability; severity remains a separate consequence signal.
Direct institutional record
Published by the responsible government authority, regulator, court or equivalent accountable institution.
Officially hosted primary notice
A traceable first-party notice on an official or directly accountable channel; company-supplied facts remain labeled.
Corroborated published evidence
Supported by reputable published evidence, but directness or source traceability remains incomplete.
Provisional signal
Retained for monitoring and not eligible for Flagship Case use without further validation.
Editorial indicators,
not probabilities.
Record scores and index positions use versioned 4SI editorial rules informed by consequence, authority depth, exposure, market action and verification difficulty.
What happened?
Documented harm, loss, intervention or material exposure.
What could proceed?
The depth and reversibility of the permission attached to the claim.
What was missing?
The difficulty of reconnecting the decision to the expected person, object or principal.
PUBLIC LIMITATION
Scores and domain positions are not statistically estimated prevalence rates, probabilities, forecasts, compliance ratings or proof that 4SI independently verified the underlying event. Domain coverage is materially unequal.
From public evidence to a defensible economic envelope.
ECR separates documented monetary impact from modeled exposure. Data coverage and evidence mode are published beside the result; missing inputs widen uncertainty.
Documented floor
Only explicit loss, transfer or physical-damage amounts qualify. Other monetary mentions do not become the floor.
Exposure vector
Published units, deaths, injuries, market action, authority depth, sector and geography inform modeling.
Translation
Welfare, remediation, disruption and financial-flow evidence is translated into a common USD-equivalent scenario frame.
Uncertainty
Observed-led inputs tighten ranges; hybrid and proxy-led cases remain deliberately wide and right-skewed.
Overlap control
Likely related records are grouped and correlated overlap is suppressed before portfolio aggregation.
INTERPRETATION BOUNDARY
Per-record ranges are uncertainty scenarios. The published portfolio envelope is not a portfolio-level P10/P50/P90 interval, prediction, valuation, reserve, damages claim or company-specific accounting loss. Calibration coefficients remain protected model IP.
Static evidence,
versioned publication.
The website serves a fixed, audited snapshot. Collection, model computation and publication remain separate from ordinary page views and filter changes.
Collect
Monitor public authority and official disclosure channels outside the website runtime.
Normalize
Preserve canonical sources and map fields without turning absence into invented facts.
Validate
Apply full chain validation only to curated Presence Incidents.
Score
Apply versioned editorial and economic methods before publication.
Publish
Serve static files with source facts, inference, limitations and correction paths.
Deterministic clustering with visible boundaries.
An Exposure is published only when the full inclusion chain is supported by at least two validated signals or a separately reviewed strong primary record. Watching signals never increase confirmed impact.
Include
Digital claim → asserted identity, object, presence or authority → attempted action → real consequence or exposure → missing trust boundary.
Validate
Evidence Assurance A–C may support public evidence. D remains outside validated Exposure status and documented economic impact.
Cluster
Reviewed incidents are grouped by trust boundary, authority type and action. Canonical URL, event fingerprint, organization, date, action and consequence prevent duplication.
Direct
New, Accelerating, Persistent, Watching and Cooling follow published time-window rules. Critical always requires documented human approval.
Model
Documented amounts remain separate from scenario ranges. Every modeled range retains incident IDs, assumptions, version, date and uncertainty.
UPDATE & CORRECTION
The website serves a static reviewed snapshot. Pipeline runs create versioned JSON before deployment; ordinary visitors never trigger scoring. Source changes and errors follow the published corrections policy.
External references for
interpretation and calibration.
The public architecture is informed by official guidance while detailed calibration priors and coefficients remain protected model IP.