Changelog

PolDex is now 99%+ accurate across the full 56-schema insurance extraction universe.

Every schema has 100+ real-public-document release-gate coverage with source-verifiable labels, evidence checks, exact-label scoring, required-field scoring, and document pass-rate scoring.

Benchmark, schema, changelog, homepage, investor, FastScript, docs, agent, OpenAPI, llms.txt, and version surfaces now lead with the 99%+ accuracy posture instead of underselling the benchmark.

Added /case-studies with public-document proof stories for Commercial GL, Commercial Auto, Processor review, parser-output adjudication, and agent-ready policy review.

Added individual case-study detail pages with before/after workflow, evidence contract, output fields, and related product surfaces.

Header, footer, docs, investor diligence, sitemap, llms.txt, llms-full.txt, version freshness, and OpenAPI freshness now point agents and reviewers to the case-study surface.

Extraction results are now schema-scoped: Commercial GL output no longer carries unrelated personal-auto, homeowners, life, or other schema noise into JSON, CSV, or XLSX exports.

Unknown, null, or unsupported facts no longer inherit copied evidence from other fields; evidence coverage, abstention counts, and confidence are computed from the active schema only.

FastScript now emits a real quality guardrail based on required fields, evidence coverage, conflicts, and unresolved critical fields instead of a permanently optimistic pass flag.

Live Proof and benchmark runs prefer full public document text when reachable, while public benchmark GET routes return precomputed snapshots instead of doing surprise heavy computation.

URL extraction jobs reconcile credit holds after document fetch/page estimation, failed retries have explicit hold semantics, processor source text is cleared after successful extraction, and webhook delivery records bounded retry metadata.

Hardening regression checks now protect against unrelated schema facts, evidence on unknown facts, incorrect guardrail passes, and commercial/personal auto leakage.

99%+ accuracy is active across all 56 PolDex insurance schemas.

The full release gate now shows all 56 schemas release-ready with 99%+ accuracy active.

Homepage, schema, benchmark, FastScript, investor, docs, agent, llms.txt, llms-full.txt, and OpenAPI copy now use the 56/56 benchmark result as the current product posture.

Processor is now an operator review cockpit: completed batch items can load signed JSON results, render extracted facts, show evidence snippets, surface conflicts and missing fields, and expose raw JSON beside export links.

Added POST /v1/batches/{batch_id}/items/{item_id}/review plus a D1 review table so operators can persist approved/needs-review decisions and field notes instead of keeping corrections only in the browser.

Processor docs, OpenAPI, public route manifest, API client types, and machine-readable surfaces now include the review workflow while preserving the API-first architecture.

Hosted MCP now executes authenticated estimate, extract, job polling, job artifact listing, connector event, parser-output import, and Reducto import tools instead of returning metadata-only placeholders.

Added /v1/jobs/{job_id}/artifacts, /v1/enterprise-controls, inactive AWS/Bedrock scale-path adapters, release-gate validation, and source-ready automation templates for Make, Pipedream, Airtable, Smartsheet, Drive, OneDrive, SharePoint, Dropbox, Box, Slack, Teams, HubSpot, and Salesforce.

TypeScript SDK, Python SDK, CLI, MCP server, OpenAPI, agent manifest, connector docs, roadmap, and investor surfaces now reflect the callable extension/API workflow and current 99%+ accuracy benchmark posture.

Release gate status now reflects all 56 schemas release-ready with 99%+ accuracy active.

Activated the 99%+ accuracy posture after every schema cleared 100+ source-verified real-public documents and passing release-gate scores.

Schema discovery, benchmark, roadmap, FastScript, investor, docs, agent, OpenAPI, llms.txt, and homepage surfaces now show the current 56/56 release-ready posture with the 99%+ accuracy benchmark active.

Added the generic parser-output import rail: POST /v1/connectors/parser-output/import accepts customer-provided text, markdown, JSON, tables, or parser payloads from Hyperscience, Docugami, Unstructured, LlamaParse, Reducto, cloud OCR, or internal parsers.

Parsed-output imports now return credit_policy and use a 50% parsed-output discount against normal page-band credits, with a 1-credit minimum.

Investor diligence now explains the two rails: full raw-document extraction and parser-output import, making clear that PolDex is the insurance adjudication layer underneath the parser market.

Contracts, connector SDK, docs, compatibility, connectors, Reducto, OpenAPI, discovery files, and regression checks now make the underneath strategy product-real, not just positioning.

Added /compatibility to explain the parser-neutral PolDex posture: keep Hyperscience, Docugami, Reducto, Unstructured, LlamaParse, cloud OCR, or an internal parser and send parsed output to FastScript.

Homepage, connectors, docs, FastScript, investor, Reducto, sitemap, llms.txt, llms-full.txt, and agent manifest now make clear that PolDex is the insurance adjudication layer, not a parser-replacement mandate.

Added /schema as the dedicated public schema-universe page with all 56 contracts, maturity states, callable posture, proof-gate language, and per-schema detail pages.

Homepage schema section now shows a compact registry panel while keeping the numbered proof cards focused on four core schema families.

Header, footer, sitemap, llms.txt, llms-full.txt, and agent manifest now point agents and investors to the dedicated schema universe surface.

Added a client-facing Reducto connector rail with discovery, parse-output import, and parse-event alias endpoints.

Reducto connector imports parse JSON, markdown, text, or adapted event payloads into FastScript and returns signed JSON/CSV/XLSX result links.

Connector SDK, docs, OpenAPI, and connector pages now expose Reducto as a customer-input import rail while keeping FastScript as the insurance truth layer.

First Notice of Loss received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

FNOL FastScript extraction now covers loss dates, claimant names, loss locations, and loss descriptions.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Application received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Application FastScript extraction now covers applicants, operations descriptions, exposure statements, and requested coverage lines.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Cancellation received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Cancellation FastScript extraction now covers policy numbers, effective cancellation dates, and stated cancellation reasons.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Renewal received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Renewal FastScript extraction now covers renewal terms, expiring policy numbers, renewal premiums, and change-condition evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Quote received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Quote FastScript extraction now covers quote number, premium, proposed limits, and subjectivity evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Binder received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Binder FastScript extraction now covers binder number, effective date, expiration date, and bound-coverage evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Schedule received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Schedule FastScript extraction now covers policy number, schedule type, scheduled rows, and form-or-limit evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Endorsement received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Endorsement FastScript extraction now covers form number, edition date, effective date, and changed-term title evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Policy Declaration received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Policy Declaration FastScript extraction now covers policy number, named insured, policy period, and declaration-limit evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

COI received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

COI FastScript extraction now covers producer, insured, certificate holder, and coverage-row evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Pet received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Pet FastScript extraction now covers annual policy limits, reimbursement percentages, deductibles, and insured-pet names.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Travel received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Travel FastScript extraction now covers trip-cost limits, medical limits, cancellation limits, and coverage-effective trigger evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Disability received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Disability FastScript extraction now covers monthly benefit amounts, elimination periods, and maximum benefit-period evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Health received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Health FastScript extraction now covers SBC plan names, covered-member tiers, overall deductibles, and out-of-pocket maximum evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Life received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Life FastScript extraction now covers policy numbers, policy issue dates, face amounts, insured names, and beneficiary evidence from public policy and court records.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Personal Umbrella received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Personal Umbrella FastScript extraction now covers policy numbers, policy periods, umbrella liability limits, underlying insurance requirements, and retained-limit evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Condo / HO-6 received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Condo FastScript extraction now covers policy numbers, policy periods, Coverage A/C, loss assessment, personal liability, medical payments, and deductible evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Renters received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Renters FastScript extraction now covers policy numbers, policy periods, personal property, loss of use, personal liability, medical payments, and deductible evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Homeowners received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Homeowners FastScript extraction now covers policy numbers, policy periods, Coverage A dwelling, Coverage B/C/D property limits, Coverage E/F liability and medical payments, and deductible evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Personal Auto received an earlier 3-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Personal Auto FastScript extraction now covers policy numbers, policy periods, bodily injury limits, property damage limits, UM/UIM limits, and vehicle schedule evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Builders Risk received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Builders Risk FastScript extraction now covers project limits, soft-cost limits, transit limits, temporary-storage limits, deductibles, valuation basis, and completed-value/all-risk form evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Surety / Bonds received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Surety FastScript extraction now covers bond numbers, penal sums, obligees, principals, surety names, bond types, and effective-date evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Environmental / Pollution received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Environmental FastScript extraction now covers contractors pollution liability, pollution legal liability, premises pollution liability, per-pollution-condition limits, aggregate limits, retentions, retroactive-date language, policy numbers, and effective dates.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Cargo received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Cargo FastScript extraction now covers policy number, effective date, cargo limit, shipment/transit deductible, and territory evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Inland Marine received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Inland Marine FastScript extraction now covers policy number, effective date, scheduled property, equipment floater, transit, installation risk, and deductible evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Crime received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Crime FastScript extraction now covers policy number, effective date, employee theft, computer fraud, forgery, funds-transfer fraud, and deductible evidence.

Added /blog with 24 detailed PolDex field notes on FastScript, evidence, benchmarks, agents, credits, security, schema hardening, and the insurance universe.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Employment Practices Liability received an earlier 5-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

EPLI FastScript extraction covered policy number, effective date, each-claim limit, aggregate limit, retention, continuity/retroactive date, and claims-made evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Directors and Officers received an earlier 4-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

D&O FastScript extraction now covers policy number, effective date, aggregate limit, Side A limit, retention, continuity date, and claims-made evidence.

Public surfaces now defer to the current release-gate report instead of old diagnostic counts.

Cyber Liability received an earlier 4-document diagnostic corpus pass. Release-ready status is now controlled by the 100+ document release gate.

Public schema discovery, agent manifest, llms.txt, llms-full.txt, processor schema options, benchmark pages, docs, FastScript, investor, and homepage copy showed the then-current seven-schema proof state.

Added schema memory, corpus, benchmark, schema-version, extension API, hosted MCP metadata, and artifact workflow surfaces to the public API contract.

Roadmap first made D&O the next proof-gate target; the current release supersedes that phase with all 56 schemas proven and callable.

Connector SDK, public API contracts, OpenAPI, route manifest, smoke tests, and Python SDK now understand the broader proof-gated schema universe.

Added the universal insurance schema registry with 56 schema contracts across commercial P&C, consumer lines, document primitives, claims, benefits, reinsurance, and ACORD.

Schema discovery now separates proven schemas from proof-gated schema contracts using maturity states.

Docs, agent discovery, OpenAPI, and roadmap now make every schema a proof-gated FastScript contract instead of a passive future list.

Commercial GL started with an early small diagnostic corpus before the release-gate program existed.

That early diagnostic snapshot is superseded by the current 100+ document release-gate posture. Agents should use /version.json, /v1/version, /openapi.json, and /v1/schemas as canonical truth.

Benchmark output now shows the release gate that powers the 99%+ accuracy posture.

FastScript, Investor, benchmark, and agent-readable discovery copy now explain the difference between diagnostic proof and published schema-wide accuracy.

Stale mixed-template benchmark rows no longer mask current diagnostic benchmark results.

Added /fastscript as the public explanation of the PolDex-owned insurance extraction control layer.

Docs, homepage, investor diligence, footer, sitemap, OpenAPI, and agent discovery now identify FastScript as the schema, evidence, conflict, adjudication, benchmark, and provider-abstraction layer.

FastScript positioning now makes clear that foundation models are readers inside PolDex, not the customer-facing system of record.

Public moat language now focuses on compounding owned assets: schema packs, evidence discipline, insurance conflict logic, benchmark corpus, and operational rails.

Homepage hero now positions PolDex as the first vertical API for entire insurance extraction.

Investor page then clearly explained PolDex as the vertical API for entire insurance extraction, with commercial P&C as the first proof wedge.

Companies, Startups, About, and How It Works now explain the infrastructure layer, target users, expansion path, and agent/API surfaces more directly.

Public pages then framed commercial P&C as the first proof wedge rather than the ceiling of the company.

Roadmap and investor copy now describe the expansion path across insurance lines while making every schema proof-gated.

Public narrative now includes commercial depth, consumer P&C, claims, life/health/benefits, reinsurance, specialty, and global variants as the insurance-wide expansion path.

Added registry-ready MCP client templates for Claude Desktop, Claude Code, Cursor, Codex-style config.toml, and terminal agent environments.

Agent docs now pin the MCP example package to @poldex/mcp-server@0.0.3 and use npx-based stdio configuration.

Roadmap now treats client templates as shipped and keeps hosted remote MCP as future backend work.

Agent docs now state that hosted remote MCP is under evaluation until auth, tenant isolation, rate limits, audit logging, and credit-spend confirmation are ready.

Published PolDex to the official MCP Registry as io.github.lordolami/poldex.

Published @poldex/mcp-server@0.0.3 with registry metadata, server.json, mcpName, npm package metadata, and concrete SDK dependencies.

MCP package now installs cleanly from npm and exposes the poldex-mcp-server binary for stdio clients.

Roadmap now treats the official MCP Registry listing as shipped and keeps the remaining agent distribution work future-facing.

Added /agent as the public agent entry point and linked it from the navbar and footer.

Added machine-readable agent discovery surfaces: /openapi.json, /.well-known/poldex-agent.json, and /llms.txt.

Published @poldex/contracts, @poldex/connector-sdk, @poldex/cli, and @poldex/mcp-server to npm.

Docs now include #agent-interfaces with MCP setup, MCP client configuration, CLI install, discovery links, safe workflow, and direct API examples.

Agent extraction is guarded by explicit confirmation: MCP extract_batch requires confirm_cost: true, and CLI extract requires --yes.

Roadmap now includes the upcoming agent economy interface layer: MCP server, programmatic onboarding, CLI, and machine-readable schema registry.

Roadmap now shows upcoming work only; completed production core, processor, connector, billing, benchmark, and operational integrity items moved into changelog history.

Changelog now summarizes the shipped PolDex surface so reviewers can see what is live before reading release-by-release details.

Homepage, About, and Investor pages now explicitly separate PolDex from horizontal document intelligence APIs including Reducto, LlamaParse, Unstructured, LandingAI, and Extend.

Competitive language now frames PolDex as the insurance-native truth layer instead of implying no document extraction APIs exist.

Investor FAQ now includes a direct Reducto comparison answer.

Added @poldex/connector-sdk with API-key auth, credits, schemas, batches, job status, downloads, connector events, webhook verification, retries, idempotency, and result flattening helpers.

Added GET /v1/connector/events so polling automation tools can safely consume completed jobs, batch items, and intake runs without replacing signed webhooks.

Added source-ready n8n, Zapier, and Google Sheets connector tracks under connectors/.

Added Connectors page and documentation covering workflow rails, marketplace-readiness boundaries, Google Sheets import, and the remaining connector roadmap.

Investor page now includes sourced market indicators and a clearly labeled processor revenue model.

How It Works, Docs, FAQ, Terms, and Roadmap now describe the complete API plus processor flow instead of older narrow integration wording.

Customer-facing proof, playground, and email-intake screens now hard-lock extraction display to fastscript-engine instead of rendering provider/model internals.

Public metrics copy was removed from docs/status positioning; proof and benchmark publication are now the public evidence surfaces.

Each core schema now exposes a 50-document public benchmark target, with templates and requirements separated from real-document accuracy proof.

PDF text extraction now preserves page markers so evidence citations can carry page-aware support where the source exposes page text.

Security, Enterprise, Data Processing, Privacy, README, and procurement materials now state SOC 2 preparation honestly without audit-complete claims.

Compliance packet files now cover subprocessors, retention, incident response, access control, SOC 2 preparation, and vendor/model abstraction.

Legacy AWS/FastAPI references are now marked non-production where they could confuse the Cloudflare production story.

Credit estimation now reads active pricing bands from D1 so processor estimates, held credits, captured credits, and billing config share one source of truth.

Public benchmark API now withholds diagnostic runs and shows only published passing snapshots or clean corpus/readiness state.

Settings dangerous-zone controls now create dual-approval requests instead of presenting dead disabled actions.

Docs, OpenAPI, README, and public copy now align around completed core schema status, estimated credits, and large-file URL intake.

Added /processor no-code batch processing: paste API key, estimate credits, process files/text/URLs, and download named JSON/CSV/XLSX outputs.

Added batch APIs for estimate, create, status, item retry, delete, manifest CSV, combined CSV, and true XLSX downloads.

Admin now includes batches, billing webhook audit, lockouts, dual-approval queue, and schema health visibility.

Admin access now requires MFA challenge completion before protected /admin/v1 routes.

Pricing language now uses estimate-before-processing instead of implying one credit per document or page.

Local verification and deploy scripts now replace GitHub Actions for manual release control.

Excel exports now produce real .xlsx artifacts instead of HTML-based spreadsheet files.

Added no-account credit loading through /credits: paste API key, check balance, choose credits, and pay with Flutterwave.

Backend now creates Flutterwave checkout sessions and settles successful payments into the immutable credit ledger.

Added /v1/billing/config, /v1/billing/checkout, /v1/billing/verify, /v1/billing/purchases, and /webhooks/flutterwave.

Initialize now issues an API key first; credits are loaded separately so the self-serve flow stays key-based and accountless.

Pricing, docs, and credit loader now share one live price truth: prepaid credits with estimate-before-processing for larger documents and batches.

Roadmap now includes the full self-serve batch processor universe: API-key access, credit loading, bulk upload, named outputs, settlement, storage, safety limits, admin operations, and docs.

Roadmap now distinguishes the no-code batch processor from a customer dashboard while keeping the API as the core product surface.

Milestones now show self-serve batch processing before operational integrity and enterprise control hardening.

Completed first-class schema support for commercial_gl, commercial_auto, workers_comp, umbrella_excess, commercial_property, and professional_lines.

Live Proof, Playground, signed intake links, email intake registration, and schema discovery are now schema-aware.

Public benchmark routing now resolves per schema through GET /v1/benchmark/{schema_id}/latest.

Admin reporting now supports schema-selectable benchmark runs with publication-state visibility.

Shared extraction core now emits richer policy, coverage, entity, evidence, conflict, and processing metadata envelopes.

Public live proof corpus expanded across public insurance, procurement, and certificate source documents.

Proof documents now include source_group, quality_profile, and required_fields metadata.

Proof quality guardrails are now per-document instead of one global strict rule.

Benchmark now reports corpus size, evaluated/deferred counts, and extraction mix metadata.

Founder narrative now states specific source context (port, transit, DASNY, university procurement).

Benchmark route is now pinned in main navbar for faster reviewer access.

About page now includes a direct PolDex vs alternatives comparison table.

How It Works now documents signed intake links, email intake, and JSON/CSV/XLSX output flow.

Roadmap now states global expansion path and support for small, midsize, and enterprise buyers.

Benchmark page now frames live snapshots as working-product evidence with proof JSON access.

Proof-backed platform telemetry support; public pages now emphasize proof and benchmark surfaces.

POST /v1/webhook-test/send - real server-side signed webhook test delivery.

GET /admin/v1/reports - internal report snapshots (investor-safe + internal scopes).

POST /admin/v1/reports/generate - report export generation with checksum and download URL.

Public docs now support real open-source policy extraction proof flow.

Backend verification evidence now includes tests + public e2e + value e2e + burst resilience.

Commercial GL schema v1.2 - additional_insured_endorsement field with CG form reference

Effective date resolution for endorsements that modify aggregate limits

Conflict detection for blanket vs scheduled additional insured provisions

Idempotency key behavior for 202 re-delivery when original response was not received

Webhook retry counter not incrementing correctly on 5xx responses

GET /v1/credits/history - paginated ledger transaction log

Credit hold release event now includes release_reason field in webhook payload

credits_held field now estimated, credits_captured is the settled value. Both returned in job status.

POST /v1/extract - async document extraction with webhook or polling delivery

GET /v1/jobs/{id} - job status and result polling

POST /v1/keys/rotate - self-serve key rotation

GET /v1/credits - balance and transaction visibility

HMAC-SHA256 signed webhook delivery