Insurance schema packs
Each supported line has document profiles, field contracts, normalization rules, evidence expectations, conflict states, export mapping, and benchmark requirements.
FastScript is the control system inside PolDex. It turns model reads into schema-bound, evidence-backed insurance data with conflicts, abstention, validation, exports, and stable API behavior.
The moat is not one prompt or one model. The moat is the growing insurance control layer: schema packs, evidence discipline, conflict logic, benchmark corpora, and delivery rails that compound across every document family PolDex supports.
Insurance extraction cannot stop at text. Policy data needs source authority, line-specific field contracts, missing-value handling, conflict handling, and audit-ready evidence.
Each supported line has document profiles, field contracts, normalization rules, evidence expectations, conflict states, export mapping, and benchmark requirements.
FastScript converts raw model reads into cited facts with page, form, section, source text, confidence, and provenance so downstream systems can audit every important value.
Insurance documents disagree. Declarations, COIs, endorsements, schedules, and requirements packets can all state different versions of a fact. FastScript keeps the disagreement visible instead of flattening it away.
Line-specific rules decide when a value is accepted, when it is unresolved, when evidence is too weak, and when a human or downstream system should review the output.
Foundation models are readers inside the system, not the customer-facing contract. FastScript keeps provider routing internal so PolDex can improve models without changing customer output shape.
New schemas and major behavior changes are tied to corpus targets, required-field scoring, evidence checks, and a release gate before accuracy claims are made publicly.
A competitor can call the same foundation models. They still need insurance-specific schemas, adjudication, evidence design, benchmark material, operational rails, and years of production edge cases.
| Asset | Why it compounds |
|---|---|
| Schema memory | Every supported line adds PolDex-owned field definitions, document profiles, extraction states, and export contracts. |
| Evidence discipline | The engine is trained to preserve what supported a fact, not just the fact itself. |
| Insurance conflict logic | Coverage truth is often conditional. FastScript represents conflicts, unresolved values, and authority differences explicitly. |
| Operational rails | Credits, estimates, webhooks, batch states, signed artifacts, and connector events all sit around the same controlled output. |
| Benchmark corpus | Each schema family accumulates public and private evaluation material that makes behavior harder to copy over time. |
| Interface reuse | API, processor, MCP, CLI, proof, playground, intake, email intake, and connectors all call the same contract. |
Customer-facing behavior is controlled by FastScript. Underlying model routing can change without changing the API, processor, CLI, MCP, or export contracts.
Identify schema fit, document profile, source type, page structure, and whether the input belongs to a live or future insurance family.
Use model readers to extract candidate facts, spans, tables, clauses, schedules, and endorsement signals.
Map candidates into canonical field names, money/date/entity formats, truth states, and evidence records.
Resolve supported values, preserve conflicts, abstain when evidence is weak, and mark unresolved fields without guessing.
Return stable JSON and derive CSV, XLSX, ZIP, webhooks, signed links, connector events, and agent artifacts from the same truth source.
API, processor, playground, live proof, intake links, email intake, MCP, CLI, webhooks, and connector events all consume the same FastScript-controlled output contract.
JSON is the source of truth. CSV, XLSX, ZIP files, signed links, and connector payloads are derived from that same result instead of being separately interpreted.
Estimates, held credits, completed jobs, failed jobs, retries, deletion, and exports sit around the same controlled extraction behavior.
FastScript does not expose raw completion text as the product.
FastScript assumes text extraction is only the first step; insurance truth comes from schema and evidence control.
FastScript sits under API and processor surfaces. Customers keep their own AMS, CRM, spreadsheet, agent, or workflow tool.
The behavior is versioned, benchmarked, and routed through reusable schema contracts instead of ad hoc prompts.
Every schema, benchmark, customer document family, conflict rule, and evidence pattern strengthens the control layer behind the public API.