CAT Synapse Pipeline Simulation · One Autonomous Process, Parameterized by Peril
Illustrative demo · uses Event 252793 policyholder pool (PH-MO-001, Cape Girardeau MO)
One pipeline, parameterized by peril
CAT Synapse is one autonomous process. The peril slot below is a parameter — the same DSN → DSTCE → CCM → HPE → CIL pipeline runs regardless of which peril is selected. Switching perils only swaps the DSTCE model instance (M-01 to M-13); every downstream stage is peril-invariant.

PERIL — INSTANCE SLOT FOR DSTCE (STAGE 2)

DSTCE INPUT PARAMETERS · CURRENT PERIL INSTANCE

POLICYHOLDER CONTEXT

Alert polygon 3 counties
Number of counties inside the NWS alert polygon
Distance to event 4.2 mi
Nearest event boundary to PH-MO-001 property
Delivery channel preference

PIPELINE STAGES

STAGE 1
DSN
Ingest & normalise
CLM-1 CLM-3
STAGE 2
DSTCE
Semantic inference
CLM-1
STAGE 3
CCM
Composite scoring
CLM-1 CLM-2
STAGE 4
HPE
Publish & deliver
CLM-1 CLM-5
STAGE 5
CIL
Retrain
CLM-1 CLM-4
READY
Select a peril, adjust parameters, then step through the pipeline one stage at a time.

DSTCE OUTPUT Stage 2

Select a peril and Run Pipeline to see DSTCE inference.

CCM SCORING Stage 3

Composite Score will be computed at Stage 3.

HPE ALERT Stage 4

Personalised alert content will be generated by M-21 (NLG) and routed by M-22.
Single Patent Filing · Read this first
CLM-1 through CLM-5 are five claims within one patent filing — not five separate inventions, and not one invention per peril domain. They describe a single autonomous pipeline with a Composite Score at its center. The peril slot (wildfire, flood, hurricane, tornado, hail, winter storm) is a parameter of the invention, not a dimension along which the invention multiplies. M-01 through M-13 are pluggable instances of the same DSTCE stage — same input contract, same output contract, only the learned weights differ.
Baseline nomenclature: This document uses CLM-1 through CLM-4 as the current claim labels. Any prior C-1 / C-2 / C-1(iv) / C-4 references in earlier drafts should be treated as retired. The mapping is CLM-1 ← C-1, CLM-2 ← C-2, CLM-3 ← C-1(iv), CLM-4 ← C-4.
Proposed Patent Claims CLM-1 through CLM-4

The five claimed innovations of CAT Synapse

CAT Synapse is one autonomous catastrophe risk intelligence pipeline — DSN → DSTCE → CCM → HPE → CIL — that ingests raw natural and man-made hazard feeds, performs semantic inference through a peril-parametric DSTCE stage, computes a Composite Score fusing event risk with location proximity, and publishes personalised outputs to policyholders, insurer operations, and underwriting audiences without human intervention between stages. The five claims below assert distinct innovations within that one pipeline: CLM-1 the pipeline method, CLM-2 the Composite Score fusion, CLM-3 the synaptic-activation dynamic architecture, CLM-4 the versioned event evolution + retraining mechanism, and CLM-5 the autonomous hyper-personalized engagement. Click any claim to see it in the simulator.

Claim architecture — at a glance

Five claims, one autonomous pipeline. CLM-1 is the foundational method claim covering the whole pipeline. CLM-2 is a system claim covering the Composite Score fusion at Stage 3. CLM-3 is the Synaptic Activation State / dynamic architecture at the DSN layer. CLM-4 is the versioned evolution + retraining mechanism across the event lifecycle. CLM-5 is the autonomous hyper-personalized engagement at Stage 4. Click a tile to jump to its stage in the simulator.
CLM-1 · METHOD
Autonomous Risk Cognition & Engagement Core
The pipeline itself — DSN → DSTCE → CCM → HPE → CIL — with peril-parametric semantic inference, two-constituent scoring, and fully autonomous execution across all stages.
CLM-2 · SYSTEM
Composite Risk Scoring Engine
Sits inside CLM-1 at Stage 3. Fuses CAT Event Risk Score (asset-independent) with Proximity Risk Score (location-dependent) into a Composite Score — the terminal CAT Synapse output.
CLM-3 · METHOD
Synaptic Activation State — Dynamic Architecture
At the DSN layer. Compute topology reconfigures per event; pathways instantiate on threat and auto-prune on subsidence — self-liquidating, just-in-time infrastructure with zero idle overhead.
CLM-5 · METHOD
Autonomous Hyper-Personalized Engagement
The Stage 4 engagement claim. Three parallel audience-specific artifacts — policyholder alert, insurer operations record, risk narratives dashboard — produced without human staging or approval.
CLM-4 · SYSTEM & METHOD
Continuous Improvement Loop
Runs alongside CLM-1 throughout the event lifecycle. Each new alert version has its own REFID, URN, start time, and polygon composition; M-25 feeds observed outcomes back into M-01–M-13, M-17, M-18 for closed-loop refinement.

The five claims in detail

CLM-1 · METHOD CLAIM

Layered Autonomous CAT Risk Processing Pipeline

A computer-implemented method for autonomous catastrophe risk intelligence comprising: (i) multi-source ingestion and normalisation of natural-hazard, man-made-hazard, and social-signal feeds; (ii) peril-specific Transformer inference producing semantic classification of Likely versus Observed event class, severity tier, immediacy tier, and structured narrative; (iii) two-constituent scoring combining event risk and proximity risk into a Composite Score; and (iv) fully autonomous execution end-to-end, with no human intervention between stages.
Pipeline stages
Stage 1 · DSN Stage 2 · DSTCE Stage 3 · CCM Stage 4 · HPE Stage 5 · CIL
Every stage is autonomous. The method covers the entire chain from raw feed to published Composite Score.
Module basis · AI Model Registry v2
M-23M-24 — DSN ingestion & quality (Stage 1)
M-14M-15M-16 — DSTCE source sub-models (Stage 2)
M-01 → M-13 — DSTCE peril-specific Transformers (Stage 2)
M-17M-18M-19 — CCM scoring & optimisation (Stage 3)
M-21M-22 — HPE personalisation & delivery (Stage 4)
Novelty basis
Prior-art CAT modelling systems separate peril inference (deterministic parameters or single-peril ML), scoring (rule-based composites), and delivery (human-staged messaging). CAT Synapse binds these into one autonomous pipeline in which peril-specific Transformers produce semantic output that scores compose downstream and NLG delivers without human handoff. The invention is the compositional method, not any individual model or stage.
CLM-2 · SYSTEM CLAIM

Composite Risk Scoring Engine

A system, embedded within Stage 3 of the pipeline of Claim CLM-1, comprising a Composite Score Synthesis Engine that autonomously computes a Composite Score as a function of two factors: a CAT Event Risk Score (asset-independent, derived from peril-specific DSTCE inference) and a Proximity Risk Score (location-dependent, calibrated per alert polygon). The Composite Score is the terminal output of the CAT Synapse system and is published to the client insurer through Stage 4.
Composite Score — formula
Composite Score = f(CAT Event Risk × Proximity Risk) CAT Event Risk = M-17 output · Proximity Risk = M-18 output · fused directly (no separate synthesis engine)
The fusion — not either factor alone — is what the claim protects. Neither M-17 nor M-18 by itself constitutes CLM-2.
Module basis
M-17 — CAT Risk Score (CRS) Engine — Constituent 1 (asset-independent event risk)
M-18 — Proximity Delta Scorer — Constituent 2 (location-dependent proximity risk)
The Composite Score is the two-constituent fusion: normalize[ CAT Event Risk × Proximity Risk ] × 10. There is no third factor — PHRS is out of scope, computed downstream by the client insurer.
Novelty basis
Prior-art insurance risk scoring either uses a single global risk score (peril hazard only) or an insurer-side composite that combines external hazard with proprietary policyholder attributes. CLM-2 claims a system-level Composite Score computed inside the CAT intelligence platform, fusing an asset-independent event score with a per-polygon proximity score — a pre-insurer intermediate output that stops precisely at the boundary of proprietary policyholder data. The Policyholder Risk Score (PHRS) is expressly out of scope and computed downstream by the client insurer.
CLM-3 · METHOD CLAIM

Synaptic Activation State — Dynamic Architecture

A computer-implemented method for autonomous dynamic-architecture management at the Dynamic Synapse Network (DSN) layer of the pipeline of Claim CLM-1, comprising: (i) instantiating peril-specific data pathways on threat detection, with compute topology reconfiguring per event; (ii) weighting active pathways by real-time urgency; and (iii) autonomously dissolving pathways when urgency falls below a learned threshold — self-liquidating, just-in-time infrastructure that eliminates idle overhead without human infrastructure management.
Maturity model — L1 to L5
  • L1 · Static monitoring — fixed limits, legacy polling (prior art)
  • L2 · Triggered pipelines — rule-based gates, sequential (prior art)
  • L3 · Sentinel state — cognitive Likely-vs-Observed interpretation (novel)
  • L4 · Synaptic Activation State — topology reconfigures per event; just-in-time compute follows the threat (novel)
  • L5 · Self-organizing optimization — closed-loop auto-pruning; self-liquidating micro-services (novel)
CLM-3 covers L4 (Synaptic Activation State) and L5 (Self-Organizing Optimization). For the worked example, 10 DSN pathways instantiated on the wildfire trigger; 7 self-liquidated on 01/19 as urgency subsided.
Module basis
M-23 — Synaptic Activation Controller (SAC) — pathway activation + dissolution
M-24 — Anomaly Detection & Data Quality Scorer — validates activated pathways
Activation rules are expert-defined (peril type → data source); the dissolution threshold (lambda) is ML-calibrated by the CIL. The self-liquidating infrastructure is the novelty.
Novelty basis
Prior-art systems either poll continuously (always-on, high idle cost) or use static triggered pipelines with reactive allocation. CLM-3 claims a self-organizing compute topology that instantiates pathways on threat and autonomously dissolves them on subsidence — the system builds and destroys its own data pathways per event, with no human infrastructure management. L4/L5 are the inventive step.
CLM-4 · SYSTEM & METHOD CLAIM

Real-Time Event Impact Evolution + Autonomous AI Retraining

A system and method for versioned event evolution and closed-loop model refinement comprising: (a) for each material update to an underlying event, generating a new version instance having its own reference identifier (REFID), Uniform Resource Name (URN), start time, and polygon composition, while preserving prior versions in a persistent history; (b) for each version, computing an updated Composite Score and re-publishing to all three audience artifacts; and (c) autonomously feeding observed outcome data from concluded versions back into the pipeline's peril-specific Transformers (M-01–M-13) and scoring modules (M-17, M-18) via a delta-learning refinement engine (M-25).
The versioning mechanism — one alert, evolving over time
Example: single Flood Watch evolving through four versions (matches REFID-987805 reference pattern)
v1.0 REFID-987685 · started 08:32 PM · polygon: Butler; Clarion; Jefferson
v2.0 REFID-987707 · started 08:43 PM · polygon: Clarion; Jefferson; Armstrong
v3.0 REFID-987805 · started 09:38 PM · polygon: Jefferson; Armstrong; Westmoreland
v4.0 REFID-987xxx · started 09:55 PM · polygon: Armstrong; Westmoreland; Indiana
Every version has its own REFID, its own URN, its own start time, and its own polygon composition. Counties drop off (Butler was in v1, gone by v3) and are added (Indiana is in v4, wasn't in v1) as NWS refines the boundary. All versions are retained in history for audit, claims traceability, and delta-learning. Advance to Stage 5 in the simulator to see this mechanism live.
Module basis
M-25 — Post-Event Refinement (CIL-MOD) — delta-learning weights back into upstream models
M-27 — Event Version Tracker — polygon-evolution regression, stakeholder mutation
Feedback edges from M-25 flow back into M-01→M-13 (peril DSTCE weights), M-17 (CRS calibration), and M-18 (proximity decay tuning). Closed-loop autonomous retraining — no human labelling step.
Novelty basis
Prior-art CAT alerting either overwrites a prior alert with each NWS update (losing history) or maintains a versioned log without downstream re-scoring (history captured, but no re-computation of composite risk per version). CLM-4 claims a system in which each version is a first-class instance with its own identifiers, spatial boundary, and re-computed score, while a closed-loop feedback pathway autonomously refines upstream models from observed outcomes across the version family. The combination — per-version instance identity + boundary evolution + closed-loop retraining — is the invention.
CLM-5 · METHOD CLAIM

Autonomous Hyper-Personalized Engagement

A computer-implemented method for autonomous multi-audience publication of catastrophe risk intelligence, executing at Stage 4 of the pipeline of Claim CLM-1, comprising: producing three audience-specific artifacts in parallel — a per-policyholder alert with NLG-generated risk narrative and channel routing (policyholder audience); a client insurer operations record with URN, provenance, and versioned history (claims / operations audience); and a Risk Narratives Dashboard with CAT Event Risk Score gauge and structured narrative (underwriter / risk-analyst audience). All three artifacts are produced and published without human staging or approval between production and delivery.
Three parallel audience-specific outputs
  • Policyholder alert — SMS / app / email / IVR to the insured, with peril-specific recommended actions.
  • Client insurer operations record — dashboard entry with URN, agency provenance, status pills, versioned history.
  • Risk Narratives Dashboard — CAT Event Risk Score gauge, Risk Category, structured narrative (Summary / Threats / Factors), peril-specific data panel, weather outlook.
One pipeline run, three distinct artifacts, three distinct audiences. No human handoff between Composite Score and delivery.
Module basis
M-21 — Alert Personalisation & NLG — content generation across all three artifacts
M-22 — Channel & Timing Selection — autonomous routing
Neither M-21 nor M-22 computes a Policyholder Risk Score. They personalise and route the Composite Score plus DSTCE narrative — the PHRS is calculated downstream by the client insurer.
Novelty basis
Prior-art alerting systems produce a single audience-generic message (typically NWS CAP forwarded verbatim) with human staging for insurer distribution. CLM-5 claims simultaneous, audience-differentiated publication — the same pipeline run yields distinct content for policyholder, claims, and underwriting audiences — with autonomous channel selection. The three-artifact-per-run structure is the innovation.

The client insurer boundary — what is deliberately out of scope

The Policyholder Risk Score (PHRS) is NOT a CAT Synapse output

The claims above stop at the Composite Score. The Composite Score is what CAT Synapse publishes to the client insurer through Stage 4. Downstream of that boundary, the client insurer computes their own Policyholder Risk Score (PHRS) by combining the Composite Score with their proprietary data:

  • Individual policyholder claims history
  • Total Insured Value (TIV) per property
  • Proprietary underwriting attributes
  • Portfolio concentration and accumulation modelling

This boundary is deliberate architectural design, not a capability gap. Placing the PHRS outside the CAT Synapse boundary keeps the claims focused on the pre-insurer risk-intelligence layer, avoids overlap with existing insurer-side rating and underwriting IP, and cleanly delimits what the five claims cover.

Prior-art distinction summary

Capability Prior-art CAT modelling & alerting CAT Synapse Claim(s)
Peril inference Deterministic parameters or single-peril ML models trained on structured inputs. Peril-specific Transformer models (M-01 to M-13) producing semantic classification: event class, severity, immediacy, narrative. CLM-1
Risk scoring Single global risk score, or insurer-side composite mixing hazard with proprietary policyholder data. Two-factor Composite Score fusing asset-independent CAT Event Risk with location-dependent Proximity Risk; PHRS deliberately downstream. CLM-2
Pipeline autonomy Human staging between inference, scoring, and delivery. Approval gates between stages. Fully autonomous execution across all five stages. No human handoff between Composite Score computation and multi-audience publication. CLM-1, CLM-5
Compute architecture Always-on polling (high idle cost) or static triggered pipelines with reactive allocation. Synaptic Activation State — pathways instantiate on threat and auto-prune on subsidence; self-liquidating, just-in-time infrastructure with zero idle overhead. CLM-3
Alert publication Single audience-generic message (typically NWS CAP forwarded verbatim) with manual re-formatting for insurer channels. Three audience-differentiated artifacts (policyholder / claims / underwriting) produced in parallel from one pipeline run. CLM-5
Event evolution New NWS update overwrites the prior alert, or is logged as version without downstream re-scoring. Each version is a first-class instance with own REFID, URN, start time, and polygon composition; each triggers re-scoring across the pipeline. CLM-4
Model refinement Offline batch retraining from human-labelled outcome data. Closed-loop autonomous retraining: observed outcomes from concluded versions feed back into M-01 to M-13, M-17, M-18 via M-25 delta-learning. CLM-4
CAT SYNAPSE · PIPELINE SIMULATION · ILLUSTRATIVE ONLY · CONFIDENTIAL — ATTORNEY-CLIENT PRIVILEGED