← Back to App

ARBITER / Docs

Changelog & Decisions Log

Commit records and engineering reviews extracted from development audits.

Below is the sequence of core improvements, model inventory migrations, and orchestration tweaks completed during the initial design audits.

Engineering Log Entries

2026-07-13@mehul

Model ID Refinement & Auto-Title Correctness

Corrected Mistral API IDs to match callable slugs (e.g. migrated mistral-medium-3.5 to mistral-medium-3-5).
Restored standard legacy models (gpt-4o, claude-3-5-sonnet-20241022, etc.) to prevent frontend model picker desynchronization.
Gated auto-title executions to run only on successful pipeline response turns.

2026-07-10@mehul

Pricing configurator rewrite

Removed hardcoded rates from lib/agents/utils.ts, migrating all rates to a single configuration record at config/modelPricing.ts.
Rewrote calculateRunCost to support Advanced Caching: OpenAI/DeepSeek cached token rates, Anthropic cache writes/reads, and Gemini tiered prompt constraints.
Added batch mode 50% discounts across all supported models.

2026-07-08@mehul

Database Telemetry & Schema Migration

Migrated message_model_runs table columns to support recording cached_tokens counts and ttft_ms times.
Updated SSE generators to capture latency benchmarks on the initial chunk payload delta.