[JanGradkowski]

What changes

The "Ask StockWatch" assistant (#46) now asks gemini-3.6-flash first and falls back to gemini-3.8-flash. Until now the order was the other way round.

Why

In the first live evaluation:

	gemini-3.6-flash	gemini-3.8-flash
Questions answered	18	6
Which	All the hard ones: "then" sequences (including a three-step Polish one), groups, candlestick and harmonic patterns, Elliott waves, insider clusters, mixed timeframes, impossible requests, refusing investment advice	Only simple ones: RSI below 30, a golden cross, a price level, a breakout, a 10% drop, a senator buying
Wrong answers caused by the model	0	0
Answer time	2–6 s	4.5–31 s, often 503 "high demand" or a 60 s timeout

The answers are grounded in the prompt, and every rule is checked by the same validator as the Save button, so the older model is enough. Swapping the two keeps both free daily quotas (20 requests each), and gemini-3.8-flash still answers when gemini-3.6-flash is busy.

Changes

• application.properties and .env.example: the defaults for GEMINI_MODEL and GEMINI_FALLBACK_MODEL are swapped.
• AssistantLiveEvalTest: the same defaults. GeminiAssistantModelTest: the fixture names are swapped to match.
• docs/ai-assistant-plan.md: the model notes explain the choice.

Testing

• The assistant, Gemini client, knowledge, template and production-config tests pass.
• Not run yet: the four Polish live questions. They need the free quota, which resets at midnight Pacific time.

🤖 Generated with Claude Code