GROKIE: connect one free-tier provider

The application runs in free_only mode. No paid fallback is implemented.

Recommended first setup: Groq
1. Sign in at https://console.groq.com/keys and create one API key.
2. Confirm the organization uses the Free Plan. Keep billing disabled.
3. For a new installation only, copy .env.example to .env. Preserve an existing .env.
4. Set GROQ_API_KEY to your key. Set GROQ_FREE_TIER_CONFIRMED=true.
5. GROQ_MODEL defaults to openai/gpt-oss-20b. Check its availability in your account.
6. Run npm run build, then npm start. For the active preview, stop only its terminal/process before restarting.
7. Run a token analysis, then inspect Sources, methodology & actual usage.

Do not send keys in chat or put them in browser code.

Fallbacks, optional:
- Gemini: GEMINI_API_KEY, GEMINI_MODEL, GEMINI_FREE_TIER_CONFIRMED=true.
  Use a project without paid billing. Free-tier availability depends on region/account.
  Get a key: https://aistudio.google.com/apikey
- OpenRouter: OPENROUTER_API_KEY and a pinned OPENROUTER_MODEL ending in :free.
  Requests enforce zero max input/output pricing and disable model provider fallback.
  Model catalog: https://openrouter.ai/models?max_price=0

Order: enabled Groq -> enabled Gemini -> enabled OpenRouter.
On HTTP error or timeout, short-circuit that provider and try the next enabled free option.
One format repair is allowed. Retry-After creates a cooldown; no endless retry loop.

Local defaults: 40 LLM attempts/day, 80,000 reserved or measured tokens/day,
one queued run at a time, up to five pending/running jobs, 1,800 output tokens/call,
22-second provider timeout, five-minute report refresh cache.
Provider limits can be lower. The ledger is persistent in SQLite.
Groq/Gemini free-account status must be confirmed in the account; the API does not
prove billing configuration. Costs are marked unverified when inference is used.

Without a key:
Live market snapshots and seven deterministic role coverage notes are saved.
They are facts-only, not AI analyses. Follow-up performs labeled evidence lookup.
AI follow-up is an explicit separate button and sends that question to the provider.

Docs verified during implementation:
https://console.groq.com/docs/rate-limits
https://console.groq.com/docs/models
https://ai.google.dev/gemini-api/docs/pricing
https://openrouter.ai/docs/guides/routing/model-variants/free
https://docs.dexscreener.com/api/reference