MCP tool: money_board

What this tool does

money_board answers the platform’s headline question on ONE surface: what to shift money into, for how long, and how likely it is to make money. It is a standing menu of graded opportunities, composed server-side from the readers behind the plays surface, the archetype trust board, and the drawdown-risk read — never a fifth computation, so the menu can never drift from the surfaces it summarizes.

Each row carries three columns:

• what — {label, underlying, side}. The instrument class, descriptive only. side is long (a long call) or hedge (the SPY put hedge) — never anything else.
• how_long — {primary_lenses, dte_band, source}. The horizon the graded verdict holds for (a play’s own hold window, an archetype’s validated lens, or the drawdown composite’s 20-day window). dte_band reads null until the option-overlay lens ships.
• how_likely — exactly ONE labeled likelihood kind, never blended, verdict WORD first:
  • gauntlet (play rows) — the backtest verdict word (confirms / tracking / falsified / insufficient_data), per-lens verdicts (never pooled across horizons), and a sealed flag.
  • bettability (archetype rows) — the tier word (BETTABLE / PROBATION) first, then p_edge, the excess-return CI, mean_grade, n cohorts, and the n_live / n_post_lock provenance split so backfilled evidence never reads as earned-live.
  • drawdown_band (the one SPY-hedge row) — the coincident 20-day severity band word (low / elevated / high) and a descriptive trend. A gauge, not a bet count: it reads how stressed conditions ARE now, never that a specific drop is coming.
• armed — {value, abstained, reasons, last_armed_trade_date}, the trigger (not the filter). For a play it is the live SEE-gate read. For an archetype it is whether a name is firing today: value: true when at least one active screener candidate for that archetype fired on the latest screener run, with reasons naming the count and up to five firing tickers; a genuinely quiet archetype reads value: false, abstained: false; only a failed read reads abstained: true. The SPY-hedge row carries no per-name trigger (value: null). Treat armed as when to act, never whether the row is listed.

The envelope also carries session_state (fresh / stale / unknown) and ceiling_line (the platform’s evergreen reminder that no one here has a validated edge on which way the market will go).

Which rows appear

• One row per locked play.
• One row per screener archetype the trust board reads at tier BETTABLE or PROBATION — long-side only (every archetype reaches the board as a long call). NO-BET, no-evidence, and gauntlet-falsified archetypes carry no playable edge and are excluded from the menu; they stay one click away on the trust board itself.
• Exactly one SPY-hedge row, driven by the coincident drawdown band.

A row whose play is not armed today still appears, with armed.value: false — the board is the standing menu, and the armed flag is when to act, not whether the row is listed.

How an agent uses it

Read the row’s how_likely.verdict word first, its number second, and its stamps (freshness + honesty) beside it. Then apply your own governance.

Access

money_board is a thin wrapper over the internal-gated GET /api/v1/money-board. It forwards the internal token from the MCP server’s own environment; there is no compose logic in the tool. It requires INTERNAL_API_TOKEN on the MCP container.