Skip to main content

Detailed Changelog

Every release. Every change. Every fix.

Full version-history release notes for procurement, technical reviewers, and customers running compliance evaluations. The high-level changelog lives at /changelog — this is the auditor's view.

Releases

3 versions, sorted by recency

v2026.5.2Bid Window

May 2, 2026

Confidence-gated review queue, vector PDF parser, bidirectional spec links, active-learning foundation, dashboard rewire, marketing surface expansion, branded proposal editor v1, GDPR/CCPA Phase 1, daily digest composer, change order pricing engine, proposal narrative assembler.

Added
  • Review queue with confidence-floor bulk-confirm and per-line confirm/edit/reject actions.
  • Vector PDF parser — direct extraction for born-vector sheets, vision-model fallback for raster scans.
  • Bidirectional spec link resolver with CSI 40% / keyword 35% / equipment-tag 25% scoring.
  • Symbol library + active learning v1 (per-trade ground-truth corpus, recognizer retraining job).
  • Dashboard KPIs wired to real backend telemetry (active projects, takeoff value, docs processed, avg processing time, accuracy rate, active pipelines).
  • Marketing surface: 12 trade pages, 8 platform pillars, 5 industry pages, expanded resources hub.
  • Branded proposal editor v1 with 8 reorderable section blocks, live preview, PDF export, org-level branding.
  • Daily digest composer with morning + end-of-day flavors, per-user quiet hours, opt-out controls.
  • Change order pricing engine — full markup waterfall (direct → contingency → overhead → profit → bond → tax) with two profit modes, two tax modes, org min-CO waiver, min-charge floor, and >10% project warning.
  • Proposal narrative assembler — 22 CSI titles, healthcare ICRA milestone detection, data-center critical-systems energization milestone, structured ProposalNarrative output.
  • Bid lifecycle finite-state machine with 13-state enum, dual-approval threshold $250K, BFS shortest-path routing, transition history validation.
  • Vendor scoring engine — 5-component composite (CSI 30% / Credentials 20% / History 20% / Delivery 20% / Proximity 10%) + preferred bonus, RFQ response ranker.
  • Demo slot allocator — TZ-aware host filtering, 15-min back-to-back buffer, 3-alternative suggestion fallback.
Changed
  • Sitemap auto-pulls dynamic per-slug routes from content arrays (TRADES, PLATFORMS, INDUSTRIES, DOCS, etc.).
  • Marketing shell footer restructured into 10 columns (Platform / Trades / Industries / Solutions / Use Cases / Resources / Services / Company / By Audience / Legal).
  • Marketing component prop names normalized: `headline` → `title`, `body` → `subtitle` across Hero / SectionHeading / FinalCta.
  • FaqItem type now uses `q`/`a` (was `question`/`answer`).
  • Help articles renamed marketing route from `/help` → `/learn` to avoid collision with in-app `/help`.
  • All proposal-editor blocks now keyed by stable UUIDs (was index) to prevent reorder churn.
Fixed
  • Recognizer accuracy curve no longer divides by zero when a trade has no annotations.
  • Review queue bulk-confirm now respects confidence-floor edge case where `floor === threshold`.
  • Sitemap excludes drafts (where `published === false`) instead of emitting empty entries.
  • PDF export of branded proposals now embeds the org logo at full DPI (was downsampled).
  • Dashboard accuracy rate now defaults to 100% on empty corpus instead of NaN.
  • Vendor scoring no longer returns negative composite when proximity coordinates are missing.
  • Demo-slot conflict detection now correctly handles UTC midnight rollover.
Security
  • Continued SOC 2 program work with an independent audit firm; attestation status is shared under NDA via the Trust Center.
  • Rotated all internal service-account credentials per quarterly rotation policy.
  • Patched dependency advisory in `python-multipart` (CVE-2024-XXXX, low severity, no exposure).
  • GDPR Phase 1: right-to-access data export, right-to-erasure soft-delete with PII scrub.
  • CCPA Phase 1: opt-out preference endpoint + data-processing addendum endpoint.
  • Audit log immutability now enforced at the row-level via append-only constraint.
Breaking
  • None. All API changes additive. Existing webhook payloads unchanged. Existing OAuth scopes unchanged.
Known issues
  • 3D BIM viewer (Layer 2 B7) still in beta — recommended only for IFC files under 200 MB.
  • Plan-overlay v1 (Layer 2 B4) accuracy degrades on rotated underlay PDFs; cite the page rotation explicitly until 2026.6 release.
  • Daily digest delivery latency in the 90th percentile averages 11 seconds (target: 5s) — being optimized in 2026.5.3.
v2026.4.0Audit Trail

Apr 4, 2026

Federal-audit evidence-trail hardening, vendor RFQ ranker, healthcare ICRA scope detection, branded proposal editor preview, marketing batch 10 (industries + solutions).

Added
  • Evidence-trail audit-log schema v2 with retention-class tagging.
  • Vendor RFQ response ranker (`rank_responses()`) with ship-date weighting.
  • Healthcare ICRA milestone auto-insertion in proposal narratives.
  • Branded proposal editor live-preview iframe.
  • Marketing batch 10: 5 industry pages + 4 solution pages.
Changed
  • Proposal narrative scope paragraphs now grouped by CSI division (was flat list).
  • Vendor scoring `proximity_signal` clamped to [0, 1] when zip-distance is missing.
Fixed
  • Proposal-editor reorder no longer drops the cover-letter block when moved to position 0.
  • Audit log no longer drops events when DB connection times out (now retries with exponential backoff).
Security
  • Federal-audit posture maintained; specific audit outcomes shared under NDA on reference calls per /security-deep-dive policy.
  • Patched two low-severity dependency CVEs.
Known issues
  • Active-learning recognizer retraining job latency exceeds SLA on corpora over 50k annotations.
v2026.3.0Active Learning

Mar 7, 2026

Active-learning v0 production rollout, takeoff line review queue MVP, customer-pilot conversions during the cycle, marketing batch 6 (compare pages, ROI calculator).

Added
  • Active-learning training-job orchestrator (Celery + Redis).
  • Takeoff line review-queue MVP with confirm / edit / reject actions.
  • ROI calculator marketing page with shareable URL state.
  • Marketing comparison content (/pricing-comparison + /best-ai-estimating-2026).
  • Trade-context filter for symbol detector (electrical / mechanical / plumbing / fire).
Changed
  • Symbol library schema migrated to per-org per-trade partitioning (was global).
  • Compare-page CTAs unified to `/book-demo`.
Fixed
  • Recognizer no longer ignores symbols at the page edge (was clipped by 8px buffer).
Security
  • Annual independent penetration test completed; results shared under NDA via the Trust Center.

How we version

CalVer + Codename

Calendar versioning

Releases are numbered YYYY.M.PATCH. The year and month tell you exactly when the build was cut. Patches roll out continuously between minor releases.

Codenames

Each minor release gets a codename drawn from construction-estimating vocabulary. The codename ages well in customer support tickets where v2026.5.2 is hard to remember but “Bid Window” is not.

Breaking-change policy

Breaking changes are flagged at least 30 days before landing. Migration guides published at /docs/migrations. Customer-facing API endpoints maintain a v1 contract; breaking changes ship behind v2.

Subscribe

Never miss a release.

Subscribe via RSS for changelog updates, or by email for monthly digests. Procurement teams: ask us about the SOC 2 + audit-log artifacts that change between releases.

Detailed Changelog — OmniTakeoff release notes