Vartovii logo

Release notes

Vartovii Changelog

Release archive
3.2.40

SPRINT-03-AI-RUNTIME (Structured Output & Chat Decomposition) & SPRINT-04-EVALS-FRONTEND (Eval Visibility & Frontend Cleanup)

Release v3.2.40 focuses on SPRINT-03-AI-RUNTIME (Structured Output & Chat Decomposition), SPRINT-04-EVALS-FRONTEND (Eval Visibility & Frontend Cleanup), and SPRINT-05-TELEMETRY (Event Sink & Deep Research PDF v2).

AddedFixedInternal
SPRINT-03-AI-RUNTIME (Structured Output & Chat Decomposition) & SPRINT-04-EVALS-FRONTEND (Eval Visibility & Frontend Cleanup)

Release digest

Added

  • SPRINT-03-AI-RUNTIME (Structured Output & Chat Decomposition) — Completed the AI runtime modernization track:
    • Sentiment, forensic, and report generation now use structured Gemini output contracts instead of brittle manual parsing.
    • Shared Gemini executor, model runtime adapter, and structured runtime telemetry are now in place.
    • chat_service.py was decomposed into dedicated routing, prompt, conversation, company, crypto, forensic, and analysis adapters.
  • SPRINT-04-EVALS-FRONTEND (Eval Visibility & Frontend Cleanup) — Added routing eval fixtures, gold output fixtures, CI-visible AI runtime eval and backend coverage gates, and completed the dashboard cleanup sweep:
    • Legacy App.jsx shell removed.
    • Runtime HTTP access unified.
    • Polling backoff, lazy loading, and production logging cleanup delivered.
  • SPRINT-05-TELEMETRY (Event Sink & Deep Research PDF v2) — Added repository-backed telemetry collection for Farcaster, dashboard, report, and marketing flows, sourced runtime metadata from deploy artifacts, and extended Deep Research PDF v2 with structured payload plus multi-agent workflow fusion.
  • SPRINT-06-PUBLIC-API-ALERTS (Public API v2 & Alert Beta) — Productized the public API and shipped the first alert beta surfaces:
    • Cursor-based search pagination, explicit freshness metadata, and canonical error schema.
    • Public API Python and JavaScript examples.
    • Watchlist CRUD, trust-score delta delivery, and dashboard alert subscription UI.

Fixed

  • PROD-HOTFIX (Admin Yield & Report Trust Score Alignment) — Closed the production inconsistency where completed non-vacancy admin jobs showed zero yield after reviews were already persisted, and aligned AI report trust-score rendering with the canonical overview score path.
  • CRYPTO-METADATA (Project Surface Polish) — Crypto entity metadata links now render as compact icon actions, while FDV remains intentionally N/A for open-ended supply assets where it is not a meaningful metric.

Internal

  • Regression baseline advanced from Sprint 2 closure to 579 passed, 2 skipped, 22 deselected (pytest -q).
  • Added dedicated regression coverage for AI runtime adapters, public API contracts, telemetry sink, alert delivery/watchlist flows, frontend polling and lazy-loading cleanup, Deep Research PDF v2, and changelog/public docs sync contracts.
  • AI runtime evals, backend coverage, dashboard build, knowledge-base build, and changelog app build all pass locally.
  • Synced Sprint 3, Sprint 4, Sprint 5, and Sprint 6 execution and closure across main, internal, and public-safe documentation surfaces.
  • Regenerated changelog app content from the canonical root CHANGELOG.md and preserved local SVG fallback cover generation for release art.

Technical notes