Verifex
Create an API key

A product history your team can review.

Shipped product, screening, evidence, data, and reliability updates—kept in chronological order with implementation detail available when it matters.

What changed recently.

The newest updates stay easy to scan. Expanded notes preserve the operational and technical context behind each release.

  1. Screening & reliabilityProduction Engine vNext

    vNext Screening Engine in Production, with a Rebuilt Benchmark & Site

    • The vNext decision engine is now the authoritative engine in production. Measured on the live corpus of 1,325,118 listed records across 53 sources, it produced 0 severe false positives across 5,000 adversarial negatives (95% CI ≤ 0.077%).
    • Sensitivity to real listed parties held: exact-name recall was 97.33% (95% CI 94.83–98.64%, n=300), and all 300 sampled sanctions positives were actionable.
    • Weak resemblance now resolves to a low-confidence watch signal (MONITOR → low) rather than a match, so near-miss negatives concentrate in the review band instead of surfacing as sanctions-grade results.
    3 additional release notes
    • Political exposure stays distinct from sanctions: across 150 PEP-only cases, 0 were served at sanctions-grade severity.
    • Semantic (FAISS) vector retrieval was built, measured on an identical corpus, and left disabled — on this corpus it cost precision for a recall gain that was statistically indistinguishable from zero.
    • The public landing page and /benchmark were rebuilt around the deployed engine. Every published figure is driven by a single typed source of truth, and superseded synthetic-benchmark numbers were moved into a clearly labelled historical archive. Availability is reported separately and is never counted as a false positive.
  2. Screening & reliability

    Coverage Transparency, Fail-Closed Reliability & Refreshed Product Experience

    • Screening coverage status now follows results through the API, screening history, and CSV/JSON exports. When required coverage is stale, unavailable, partial, missing, or malformed, affected workflows fail closed instead of presenting an unsupported clear result.
    • Single, batch, bulk, public OFAC, entity-composite, UBO, cached, and export paths now use the same authoritative finalized-decision semantics, reducing discrepancies between customer-facing surfaces.
    • Screening cache keys are partitioned by the requested source coverage, cached payloads are validated before use, and relevant source or plan transitions invalidate affected cache state.
    5 additional release notes
    • Atomic idempotency claims prevent concurrent duplicate screening execution. Conflicting in-flight reuse now returns HTTP 409 rather than running the same operation twice.
    • Input and export safeguards now handle zero-width obfuscation, ambiguous mixed-script input, and spreadsheet-formula injection conservatively.
    • Source operations gained pre-write validation, scheduler-coverage checks, deterministic effective-health and staleness classification, and recovery probes for temporarily unavailable feeds.
    • The public site and dashboard now use the refreshed, light-first Verifex Evidence Engine interface, with dashboard and admin feature parity restored after the redesign.
    • Privacy-safe browser error monitoring is now live to improve detection and diagnosis of customer-facing failures without sending screening payloads.
  3. Screening & reliability

    Stronger Identity Safeguards Across Screening Workflows

    • Strengthened how screening records are attributed to canonical identities. Similar names alone are no longer treated as sufficient evidence that two records represent the same identity.
    • Conflicting authority identifiers from supported OFAC, UN, EU, Wikidata, and GLEIF records now prevent unsafe canonical identity associations.
    • Ordinal and generational distinctions such as Fund II versus Fund III and Jr versus Sr are now preserved. A one-sided ordinal remains ambiguous unless supported by stronger corroborating evidence.
    3 additional release notes
    • Mixed-script confusables are handled conservatively, and name-only cross-script similarity does not establish canonical identity.
    • The same safeguards now apply consistently across single screening, batch screening, cached results, screening history, and Evidence Capsule output.
    • This update improves canonical identity attribution and the evidence attached to screening results. Existing block and allow verdict policies remain unchanged.
  4. Screening & reliability

    Batch Screening Reliability & Auditability

    • Improved auditability for batch screening workflows. Successful batch screening items now follow the same evidence capture path as single-name screenings, making high-volume screening results easier to review and explain later.
    • Strengthened consistency checks between single-name and batch screening so teams can rely on the same screening behavior across both workflows.
    • Added stronger release checks for customer-facing screening APIs, including batch screening, evidence export, quota handling, API-key behavior, and response schema validation.
    2 additional release notes
    • Clarified matching mode guidance: broad mode is recommended for sanctions screening workflows where fuzzy and phonetic recall matters, while exact mode is stricter and may miss similar-name matches.
    • Added additional release and post-deploy smoke checks to improve operational reliability for future API updates.

Earlier product history.

21 earlier releases remain available in full without making the current update feed difficult to read.

Productv3.1.0

Adverse Media Screening API

7 notes
  • Adverse Media Screening API live — POST /v1/adverse-media/screen screens individuals and companies against real-time news via GDELT DOC 2.0, with Claude Haiku 4.5 classification into 7 risk categories (CRITICAL, HIGH, MEDIUM, LOW, CLEAR, DISMISSED, UNKNOWN).
  • LLM-powered relevance scoring with entity disambiguation — distinguishes between homonyms, evaluates article tone and context, and returns per-article confidence scores with human-readable reasoning.
5 additional release notes
  • Plan-gated access: Growth (50/mo), Pro (200/mo), Enterprise (unlimited). Overage pricing: $0.15 (Growth), $0.10 (Pro) per additional check.
  • Dual-layer caching: Redis hot cache (24h for GDELT articles, 7d for LLM classifications) + PostgreSQL warm fallback. Zero quota consumption on cache hits.
  • LLM failure resilience: if Claude is unavailable, returns CLEAR with llmFailed: true and consumes zero quota.
  • GDELT query optimized with exact-phrase matching, tone<-3 filter, and domain exclusions (social media, press releases). 3 retries with exponential backoff.
  • Cost per screening: ~$0.009 for 20 articles (~3,200 tokens).
Productv3.0.0

Engine v3 — Probabilistic scoring pipeline with calibrated confidence

9 notes
  • New probabilistic scoring engine (v3) replaces the legacy rule-based pipeline. Log-odds aggregation with Fellegi-Sunter contextual weighting produces calibrated confidence scores.
  • Temperature scaling (T = 0.20) integrated into aggregateComparisons. Makes the sigmoid ~5× steeper, fixing underconfidence. ECE: 1.32% (191-case), 1.36% (500-case).
7 additional release notes
  • Context-gating penalty: −1.2 log-odds applied when a moderate name match lacks DOB, country, or identifier support. Prevents high-confidence matches on ambiguous common names.
  • 500-case synthetic benchmark published: 99.68% F1, 99.36% recall, 100% precision, 3 false negatives, 0 false positives.
  • 191-case smoke benchmark: 100% F1, 100% recall, 100% precision, 0 false positives, 0 false negatives.
  • Adversarial safety suite (247 cases, 10 corruption categories): 37.79% recall on the tested set, high precision, 0 danger-scored cases in the 247-case adversarial suite. Conservative by design (benchmark result, not a guarantee).
  • Explainability API: add explain: true to any screen request and every match returns v3_explainability with per-field contributions, threshold reasoning, and evidence summary.
  • Benchmark page redesigned with interactive suite selector, animated stats, Engine v3 banner, adversarial safety visualization, and calibration metrics.
  • Homepage updated with Engine v3 value proposition, animated stat counters, and updated benchmark preview.
Product

Trust & Conversion Sprint — SDK verification, demo pages, guides, and security packet

6 notes
  • SDK runtime verification — Node.js, Python, Go, and Rust SDKs tested against live API. SDK health tests updated to handle unauthenticated responses correctly.
  • Evidence Capsule page enhanced with live demo section — fictional sample data showing query, result status, matched source, score breakdown, reviewer action, audit metadata, and JSON export preview.
4 additional release notes
  • KYB Sample Dossier page live at /kyb-sample-dossier — standalone page showcasing the fictional Caspian Trade Holdings dossier with hero, deliverables, and CTAs.
  • Use Verifex With Existing Provider page live at /use-verifex-with-existing-provider — six use-case cards (secondary screening, evidence capsule, benchmark, audit export, regional KYB, fallback API) with FAQ and pilot CTA.
  • Security Packet page live at /security-packet — architecture overview, data flow, subprocessors, API key security, webhook signing, rate limiting, backup/retention, incident response, and honest certification status.
  • Six SEO guide pages created: /guides/sanctions-screening-api, /guides/ofac-screening-api, /guides/sanctions-screening-evidence, /guides/build-vs-buy-sanctions-screening, /guides/kyb-turkey-checklist, /guides/kyb-azerbaijan-checklist.
Product

Verifex repositioned as sanctions screening evidence infrastructure

5 notes
  • About page rewritten with institutional product positioning — explains Verifex's evidence-first approach, core product areas, operating principles, and decision-support disclaimer.
  • Homepage repositioned — headline, trust strip, and benchmark section reframed around 'Every screening decision needs evidence.' The site now reads as evidence infrastructure, not just a screening API.
3 additional release notes
  • Evidence Capsule public page copy refined — metadata and body copy now emphasize decision evidence, point-in-time list versions, and reviewable match rationale.
  • Sanctions Screening API public page copy refined — hero and metadata now describe preserving evidence behind every decision rather than 'reviewable decisions' alone.
  • Site metadata updated — title and description across the homepage, Evidence Capsule, and Sanctions Screening API pages now reflect the evidence infrastructure positioning.
Product

Evidence Capsule public page, KYB Sample Dossier, and design audit fixes

9 notes
  • Evidence Capsule public page live at verifex.dev/evidence-capsule — explains the structured audit record format, tamper-evident hash, bitemporal timestamps, and how each screening decision maps to a retrievable capsule with query input, list versions, match rationale, and reviewer action.
  • KYB public page live at verifex.dev/kyb — Know Your Business screening for Turkey and Azerbaijan markets. Includes company identity verification, registry evidence, ownership chain mapping, sanctions screening, risk summary, and structured deliverables.
7 additional release notes
  • KYB Sample Dossier Showcase added to the KYB page — interactive 7-module preview of a fictional 'Caspian Trade Holdings Ltd.' dossier with company identity, registry evidence, ownership chain, sanctions screening results, risk summary, evidence timeline, and deliverables grid.
  • About page now includes the shared site header and footer for consistent navigation.
  • Compare index and all comparison subpages now include the shared site header and footer.
  • Footer Product column updated with links to Evidence Capsule, KYB, and About pages.
  • Sitemap expanded to include /evidence-capsule and /kyb.
  • Solutions page gradients updated from blue-500 to cyan #3BBFDF for visual consistency with the rest of the site.
  • Homepage Evidence Capsule section CTA now links to /evidence-capsule instead of /audit-trail.
Screening & reliability

Pricing, docs, accuracy fixes, and UI improvements

13 notes
  • Enterprise plan no longer shows a fixed price. The pricing card now displays 'Custom pricing' — reflecting that Enterprise terms are negotiated per contract.
  • Enterprise plan 'Get Started' / 'Subscribe' button replaced with a single 'Talk to sales' action that opens a pre-filled email to [email protected]. Annual billing toggle no longer shows a strikethrough price for Enterprise.
11 additional release notes
  • Pricing page now includes an 11-row plan comparison table covering screens per month, rate limits, audit retention, continuous monitoring watches, bulk screening, Evidence Capsule export, GLEIF/LEI lookup, UBO 50% Rule, webhooks, custom list integration, and dedicated support.
  • Public docs page (/docs) redesigned to match the site's full design system — now uses the shared header, footer, and theme tokens instead of appearing as a separate product. Sidebar navigation, code blocks, and section layout all updated.
  • Sign-up form now shows a 4-segment visual password strength bar (Weak / Fair / Good / Strong) as you type, replacing the static hint text.
  • Hero stat corrected: '10yr Audit Trail' updated to '365d Max Audit Retention' — accurately reflects the Enterprise plan retention limit.
  • Hero demo and stat corrected: source count updated from 49 to 54, consistent with the current configured Enterprise source total.
  • Homepage CTA now shows 'No credit card required · 50 free screens/month' microcopy directly under the Start screening free button.
  • Security page now links to the live status page (status.verifex.dev). Verifex makes no public uptime target or SLA — a service-level commitment exists only where expressly agreed with a customer.
  • Dashboard playground: clear screening results now show a disclaimer — 'A clear result is screening evidence — it is not a legal compliance determination.'
  • Continuous monitoring page header now states that watched entities are re-screened daily and that you receive a webhook event if risk level changes.
  • UBO page now includes a beta notice explaining that the OFAC 50% Rule calculator requires a corporate entity UUID from the Verifex corporate entity database.
  • Email verification success page now auto-redirects to sign-in after 3 seconds with a visible countdown.
Screening & reliability

Phase 2 FP Reduction, API Clarity Fields & PEP Near-Match Policy

5 notes
  • Benchmark updated to May 2, 2026 verified run: 122-case published benchmark: 96.6% F1, 100% recall, 93.4% precision, verified May 2, 2026.. Phase 2 entity-name FP reduction brought false positives from 8 down to 6.
  • Phase 2 FP reduction: entity-name false positives fully eliminated. Generic business names (e.g. Global Trading LLC, Eastern Star Consulting) now correctly clear via Strategy 3 name-token gate and Penalty 14 (company evidence gate).
3 additional release notes
  • New API response fields on every match: source_severity (primary / secondary / pep / watchlist) and source_type (sanctions / pep / adverse_media). These explicitly separate the regulatory weight of a source from the match confidence score.
  • PEP near-match policy: low-confidence PEP-only matches (confidence 40–49, all matches from PEP source) now return human_review_required: false and recommended_action: review_if_policy_requires instead of triggering mandatory manual review. PEP hits require enhanced due diligence, not a sanctions block.
  • Remaining 6 borderline FPs are intentionally tolerated: all are low-confidence PEP near-matches at exactly confidence 42. The adjudication engine marks all as AUTO_CLEAR. See the benchmark page for the full FP cluster breakdown.
Product

Source Transparency Page, Trust Layer API Fields & New Data Sources

7 notes
  • Source Transparency: every configured source was published with its publisher, classification (Risk / Jurisdiction / Enrichment), and last successful ingestion timestamp. (The standalone data-sources page was later retired; source attribution and freshness now ship in every screening response.)
  • Trust Layer fields added to every screening response: verdict (clear / possible_match / confirmed_match), risk_score (0–100), confidence_score (0.0–1.0), human_review_required, recommended_action, and reasons array explaining the decision.
5 additional release notes
  • audit_log_available field returned on every screen — confirms an append-only decision record is stored for audit retrieval.
  • missing_fields returned when a screening call omits date_of_birth or country, indicating reduced match confidence due to insufficient context.
  • 4 data sources restored and re-activated: Singapore MAS (40 designations), Netherlands Terrorism List (119 designations), Ukraine NSDC Sanctions (21,000+ designations), EBRD Ineligible Entities (1,082 debarred firms). All previously unavailable due to upstream URL changes.
  • Turkey MASAK TF designation list (2,287 entries) re-activated with improved URL fallback logic.
  • Screening API throughput improved — cluster mode now active on the API tier, distributing requests across all available CPU cores.
Screening & reliability

Benchmark page refreshed — Phase 1 FP reduction results

4 notes
  • April 25 rerun: 96.6% F1, 100% recall, ~91.4% precision, 8 false positives (vendor-authored benchmark with disclosed methodology). Superseded by the May 2 rerun above after Phase 2 FP reduction.
  • Phase 1 FP reduction: surname-mismatch penalty, common-name evidence gate, and stable sort eliminated all common-name false positives.
2 additional release notes
  • Latency snapshot from this rerun showed sub-second response times; we're not republishing the specific historical percentile figures until they're reconciled against a fresh, independently-verified load test.
  • Benchmark page rewritten to show the current live breakdown instead of a frozen perfect-score snapshot.
Product

Entity Resolution, GLEIF / UBO Chain & Compliance Dashboard

6 notes
  • Entity Resolution Graph — search and de-duplicate across configured source records. A single canonical record aggregates known aliases, transliterations, and identifiers (OFAC ID, EU reference, GLEIF LEI) into one searchable profile.
  • GLEIF LEI Lookup — look up any Legal Entity Identifier (LEI) to retrieve entity details, direct parent, and ultimate parent from the Global LEI Foundation database (3.28M+ entities). Fully integrated into the dashboard.
4 additional release notes
  • UBO Chain Screening — enter an LEI or corporate UUID to traverse the full ownership chain and calculate aggregate sanctioned ownership under the OFAC 50% Rule (FAQ No. 401). Identifies constructively-blocked entities across multi-hop chains up to depth 10.
  • Compliance Dashboard — dedicated UI for compliance teams with a live Alert Queue (pending/reviewed/escalated), Monitoring dashboard for all watched entities, and generated PDF/CSV Reports with audit export.
  • PEP vs Sanctions distinction — PEP-only screening hits now return risk level MATCH (displayed in purple) instead of CRITICAL. Entities that appear only in PEP lists are politically exposed but not sanctioned — this prevents PEP hits from being conflated with actual sanctions violations.
  • Bitemporal audit history — every canonical entity record tracks both transaction time (when we learned it) and valid time (when the change was effective), enabling point-in-time screening queries.
Screening & reliability

Archived April 11 benchmark snapshot

7 notes
  • Archived benchmark snapshot from April 11. Superseded by the April 25 and May 2 reruns.
  • New: Zero-width Unicode evasion detection (FATF 2025 typology). Mixed-script query rejection. Patronymic-derivative detection (Ivan≠Ivanov). 9 penalty layers for false-positive elimination.
5 additional release notes
  • Benchmark includes exact match, spelling variations, Arabic/Cyrillic transliteration, phonetic matching, word-order variations, entity/organization names, PEP screening, multi-source matching, partial names, substring traps, common-name disambiguation, and adversarial inputs.
  • Benchmark page published at verifex.dev/benchmark with full methodology, real test cases, and comparison to published research (OpenSanctions Pairs, Federal Reserve FEDS 2025-092).
  • 11 new data sources added: BIS Unverified List, BIS Military End User List, OFAC Sectoral Sanctions (SSI), OFAC Palestinian Legislative Council, OFAC Chinese Military-Industrial Complex (CMIC), Nonproliferation Sanctions (ISN), ASIC Banned Persons (Australia), Thailand AMLO Designated Persons, Brazil PEP (95K+ officials), and IDB Invest Sanctions.
  • Coverage messaging now distinguishes raw source records from normalized, active screening entities.
  • New pricing: Starter $49/mo, Growth $99/mo, Pro $249/mo, Enterprise $499/mo. Free tier unchanged at 50 screens/month.
Product

Go & Rust SDKs

2 notes
  • REST API examples updated with Go and Rust request snippets in the docs page — zero-dependency patterns for both languages.
  • Security hardening — improved webhook SSRF protection, XXE prevention on all XML parsers, stricter input validation across the API.
Screening & reliability

Smarter Matching & 7 New Data Sources

4 notes
  • Contextual scoring — date of birth, nationality, and entity type now adjust match confidence. Each match includes an evidence breakdown explaining why it matched.
  • Faster screening — new in-memory fuzzy index reduces broad search latency from 2s to under 750ms.
2 additional release notes
  • 7 new data sources: BIS Entity List (3,329), BIS DPL (469), Europol Most Wanted (87), FATF High-Risk Countries (23), France Trésor (6,062), Netherlands Terrorism (119), ADB Debarment (1,331).
  • Country name resolution — full country names (e.g. "Azerbaijan") now work alongside ISO codes.
Data coverage

4 New Sanctions Sources — 14 Total

6 notes
  • Interpol Red Notices — 6,400+ wanted persons from 195 countries, synced daily
  • Japan MOF — 6,700+ entities under Foreign Exchange and Foreign Trade Act sanctions
4 additional release notes
  • Israel NBCTF — 2,000+ terror designated individuals and organizations
  • World Bank Debarred — 1,200+ debarred firms and individuals (re-enabled with new data source)
  • Total entity coverage expanded across all sanctions sources
  • All new sources included in Starter, Growth, Pro, and Enterprise plans at no extra cost
Product

Enterprise Features & Configured Source Coverage

7 notes
  • API Key Rotation — rotate keys with 24-hour grace period (Enterprise)
  • IP Allowlisting — restrict API key usage to specific IP addresses (Enterprise)
5 additional release notes
  • Webhook Retries — 5x automatic retry with exponential backoff on delivery failure (Enterprise)
  • Priority Screening — force a fresh screening path and skip cache where available (Enterprise)
  • Audit Retention — 1 year for Enterprise, 90 days Pro, 60 days Growth, 30 days Free/Starter
  • PEP source coverage expanded through configured PEP datasets
  • Coverage reporting now separates raw source records from the active searchable screening index
Product

Dashboard Features & Monitoring

4 notes
  • SDK Quickstart Wizard — step-by-step onboarding for Node.js, Python, Go, Rust, and cURL
  • Continuous Monitoring — watch names and get alerts when risk levels change
2 additional release notes
  • Webhook Management — configure and manage endpoints from the dashboard
  • Public Status Page — live API health and uptime at /status
Product

Subscription Billing & Payments

3 notes
  • Subscription billing — Starter ($49/mo), Growth ($99/mo), Pro ($249/mo), Enterprise ($499/mo)
  • One-click plan upgrades and downgrades from the dashboard
1 additional release note
  • Customer billing portal for invoice and payment management
Screening & reliability

Screening Accuracy Improvements

4 notes
  • Structured adjudication layer — LLM-assisted entity resolution for ambiguous matches with documented rationale
  • Entity-type mismatch detection — companies no longer match person entries
2 additional release notes
  • Name rarity scoring — common names like 'John Smith' get reduced confidence
  • Context noise penalty — 'Sberbank Cafe' no longer matches 'Sberbank'
Product

Switzerland Sanctions & API Tools

3 notes
  • Switzerland (SECO) sanctions list fully integrated — 7,641 entries
  • Postman collection with 24 ready-to-use API requests (download from docs page)
1 additional release note
  • Batch screening — screen up to 100 entities in a single API call
Product

Python SDK & Free OFAC Tool

2 notes
  • REST API quickstart examples published for Python (httpx/requests) and Node.js (fetch/axios) in the docs page.
  • Free OFAC search tool — check names against OFAC SDN, no signup required
Data coverage

Sanctions Data Sources

8 notes
  • OFAC SDN source added with automated publisher synchronization
  • EU Consolidated: coverage planned; not currently under active automated sync (official feed requires authenticated access)
6 additional release notes
  • UN Security Council: 1,000+ entries (synced every 12 hours)
  • UK Sanctions List (UKSL): 5,100+ entries (synced daily)
  • Canada SEMA: 3,400+ entries (synced daily)
  • Australia DFAT: 10,900+ entries (synced daily)
  • Switzerland SECO: 7,600+ entries (synced daily)
  • PEP sources: configured politically exposed person datasets by plan and environment

Follow the latest updates

Have questions about a recent change? Reach out directly.

[email protected]Back to home