Changelog

The v3.3 release is a comprehensive hardening and optimization sprint. Token normalization applies to all 14 document input slots across 10 engines — typical construction contracts see 18–25% fewer tokens after collapsing PDF whitespace artifacts. The progress bar easing uses an ease-out function (gap × 0.07 per tick) plus auto-crawl that drifts toward 68% during long AI waits, so the bar never looks frozen. The notification center stores the last 50 notifications in localStorage and shows a red badge for unread items. Security improvements include CSRF timing-safe comparison, Stripe idempotency, and a 9-secret rotation schedule tracked in Supabase. The CSI MasterFormat lookup replaces an AI call with a 28-division regex table covering all standard construction divisions.

The Pay Application Review engine now renders a full financial controls panel alongside the line-item table. Retainage, cumulative billing, and lien waiver data were already being extracted by the AI but were not displayed — this release brings them to the surface as structured panels with colour-coded status indicators. Front-loading patterns, retainage discrepancies, and missing waivers all trigger prominent warnings. AIA compliance, SOV overbilling flags, stored materials claims, and prevailing wage requirements are surfaced as status badges above the red flags section.

This release adds five new post-award AI engines covering the entire construction project lifecycle from COI compliance through project closeout. The Bid/No-Bid scoring feature directly addresses a top competitor gap — every scope check now produces a structured go/no-go recommendation with confidence score, win probability estimate, and itemized reasons for each position. Trust signals were added to all 14 AI engines: a confidence bar (colour-coded green/amber/red), the AI model name used, and an estimated page count for the analyzed documents. Bid Leveling was expanded to support up to 10 simultaneous vendor comparisons.

The v3.0 release focused on security hardening, UX polish, and SEO expansion. Prompt injection protection was wired through all three AI endpoints using a shared sanitizeDocumentInput() utility that strips control characters and caps document size at 200K characters. The modal system was upgraded with a _modalStack array that tracks open modals in order, enabling Escape to close the correct modal (with unsaved-changes confirmation). The tariff calculator resource page targets high-intent queries like "construction tariff calculator" and "HTS code steel" to drive free traffic.

This release focused on production readiness and platform hardening. All API endpoints now enforce CSRF token validation and strict CORS policies. The test suite grew from zero to 273 tests covering critical paths across exports, email, and procurement workflows. The i18n system was expanded to cover all user-facing strings in English, German, French, Spanish, Norwegian, Swedish, Danish, Dutch, Italian, and Polish. The original single-file architecture was decomposed into 36 discrete source modules for maintainability.

The v2.0 release marks the feature-complete beta milestone. Nine AI engines cover the full procurement lifecycle from bid solicitation through project handoff. The submittal tracker manages specification extraction through a 7-phase workflow (Pending, Under Review, Approved, Rejected, Resubmit, Approved as Noted, Void). Procurement tracking supports CSI MasterFormat codes with purchase order and invoice matching. The vendor database automatically scores suppliers based on bid history and performance, with fuzzy deduplication to prevent duplicate entries.

This content release added 13 in-depth resource pages covering construction insurance, lien waivers, bid templates, incoterms, material price tracking, RFP/RFQ guides, cost estimating, contract types, supply chain management, vendor management, material takeoffs, and payment processes. Blog content was localized for the US, UK, Germany, France, Spain, Norway, Sweden, Denmark, Netherlands, Italy, and Poland. All pages include JSON-LD structured data (Article, FAQPage, BreadcrumbList schemas) and hreflang tags for multi-language discovery.

The initial release of Trueleveler launched with five core AI engines: Bid Leveling, Contract Review, RFQ Generator, Document Compare, and PM Contract Summary. Authentication is powered by Supabase with row-level security policies ensuring data isolation between organizations. The interface supports both dark and light modes with full responsive behavior across desktop, tablet, and mobile viewports. New users receive 5 free AI analyses to evaluate the platform before subscribing.