Vartovii logo

Release notes

Vartovii Changelog

Release archive
3.2.35

ADK-03 (Session Continuity) & ADK-04 (Configurable ADK Instructions)

Release v3.2.35 focuses on ADK-03 (Session Continuity), ADK-04 (Configurable ADK Instructions), and PROMPT-01 (VC Assumption Bias).

FixedInternal
ADK-03 (Session Continuity) & ADK-04 (Configurable ADK Instructions)

Release digest

Fixed

  • ADK-03 (Session Continuity) — ADK chat now reuses session IDs per conversation_id instead of creating a fresh session on each message.
  • ADK-04 (Configurable ADK Instructions) — ADK agent prompts were externalized into backend/ai/prompts/adk.py and are now overridable via AIConfig env keys (ADK_*_INSTRUCTION).
  • PROMPT-01 (VC Assumption Bias) — Chat system prompt now explicitly forbids assuming traditional VC backing or founder pedigree when data is not provided. Tokenomics runtime logic also removes legacy missing-lead penalty in RiskEngine to avoid unfair score loss for fair-launch/community-first projects.
  • PROMPT-02 (Sarcasm Classification Gap) — Sentiment prompts now include explicit sarcasm/irony handling instructions for intent-based classification.
  • LOG-01 (Execution Persistence) — Chat runtime execution metadata is now persisted and returned to clients:
    • new DB column: chat_logs.execution_context (JSONB)
    • POST /api/ai/chat response includes execution payload
    • frontend chat message history now persists executed tool traces per assistant message (not only transient loading tracker).
  • CLEAN-01 (Duplicate Formula Drift) — backend/ai/tools/db_utils.py trust-score helper now delegates to canonical backend/api/agent_tools.py::calculate_trust_score.
  • CLEAN-02 (Report Schema Drift) — Reports API now normalizes section aliases to canonical v2 keys in orchestration core and emits schema metadata (report_schema, X-Report-Schema). PDF renderer now prioritizes canonical v2 keys, with legacy aliases as fallback only.
  • CLEAN-03 (Env Template Drift) — env-example contract remains standardized to a single committed template: .env.docker.example.

Internal

  • POST /api/ai/chat request supports optional conversation_id.
  • POST /api/ai/chat response includes:
    • tools_used
    • execution metadata object
    • message_id.
  • GET /api/reports/{company_name} includes report_schema: "v2".
  • GET /api/reports/download/{company_name} includes X-Report-Schema: v2.
  • Added tests/test_block15_adk_prompt_cleanup.py (11 tests).
  • Full regression: 223 passed, 2 skipped, 22 deselected (pytest -q).
  • Targeted quality gates passed for all changed files (pre-commit run --files ...).
  • Added docs/reports/2026-03-06_BLOCK15_VERIFICATION_REPORT.md.
  • Synced Block 15 updates across main/internal/public documentation surfaces.

Technical notes