🚀 Highlights

A fund-manager voice, not a data dump. firma now talks back like a private fund manager — decisive, evidence-based, micro-macro bridged in dollars. Four explicit Voice and judgment principles in SERVER_INSTRUCTIONS make vague hedge phrases ("could pressure tech", "조건부로 보라") explicit violations.

Short-horizon quant view. simulate_scenario is the deterministic sibling of project_net_worth's stochastic Monte Carlo. High-impact macro events in the brief (FOMC / CPI / PCE / NFP within 7 days) auto-trigger it — Claude doesn't wait for you to ask "what if".

Two new behavioral tools. show_thesis_track joins your recorded reason ↔ outcome for past trades. show_anti_portfolio follows the post-exit prices of positions you fully sold — the data brokerages deliberately don't show because it highlights bad sells.

Brief = diagnostic only. The brief now strictly returns currently true facts; forward-looking probability routes to project_net_worth. The two epistemic families never mix.

🆕 New MCP tools

• simulate_scenario (#48) — beta-driven shock impact per holding + per-ticker overrides + one-sigma daily range when you pass daily_vol_pct
• show_thesis_track (#49) — past trade reasons ↔ current outcome labels (validated_holding / underwater_holding / cut_loss / neutral_exited)
• show_anti_portfolio (#49) — post-exit performance of fully-sold positions (missed_rebound / good_call / neutral_exit)

🔌 New MCP resources

Attach as context in Claude Code with @firma::

• firma://portfolio — current holdings table
• firma://thesis-log — recent trades with recorded reasons
• firma://profile — your goals

📊 Brief surface

• risk_summary.daily_vol_pct now exposed (1σ daily range — input to simulate_scenario)
• holdings[].fundamentals.beta_5y now exposed
• economic_calendar capped at 25 events, high-impact first then medium (chronologically sorted)
• goal_tracking no longer carries projected_net_worth or on_track (diagnostic only)

🎙 Voice and judgment principles

`SERVER_INSTRUCTIONS` now codifies five new rules that govern how Claude talks:

• Principle 13. Decisiveness over neutrality — name a lean + the tripwire
• Principle 14. Micro–macro bridge — every macro statement MUST end with a dollar number on the user's book
• Principle 15. Behavioral context, not platitudes — chain into `show_thesis_track` on buys, `show_anti_portfolio` on sells
• Principle 16. Evidence-based conviction — every BUY/HOLD/SELL stance cites the firma field justifying it
• Principle 17. Always capture the thesis on new trades — Claude asks for `reason` if missing
• Principle 8b. High-impact macro events auto-trigger `simulate_scenario` even when the user didn't ask "what if"

⚠️ Breaking changes

• `simulate_goal_projection` renamed to `project_net_worth` (#44). API generalized: `target_value` optional, `horizon_months` 1–720 (one tool for both short-term outlook and long-term FIRE planning). Update any external invocation.
• Brief no longer returns `projected_net_worth` or `on_track`. Forward-looking probability routes to `project_net_worth`.

📚 Docs

• New `scenario.{en,ko}.mdx` reference page (parallel to `projection.{en,ko}.mdx`)
• `why-firma` adds fund-manager voice + decision-history features as front-page identity
• Installer vs npm flow clarified on the install page
• Audited and cleared stale references (`target_net_worth_usd`, `/morning`, `--json`, `tax_outlook`, `simulate_goal_projection`, `world-intel`, `show_insider`, `home_currency`)
• README front page reflects the identity shift to fund-manager voice

🧹 Maintenance

• Cleanup pass — dead code, brief abstractions, `buildPortfolioView` extraction (#45)
• Projection response size: yearly resolution by default + rounding (#43)
• Docs h4 subheader styling fix (#47)

Install

```bash

npm

npm install -g @evan-moon/[email protected]

macOS

Download firma-mac-arm64.pkg below

Windows

Download firma-win-x64.exe below

```

After install, restart Claude Desktop so the new MCP tools and resources show up.