The pulse check

A pulse check is the fast platform-state read — under a minute for humans on the dashboard, one HTTP call or MCP tool invocation for agents. It exists because regimes flip, alignment shifts, alerts arrive, and calibration drifts; you need a habit for sweeping the whole platform without getting lost in a single signal card.

The eight-step sequence

The dashboard’s section ordering implements this sequence top to bottom — you can pulse-check by scrolling. The same ordering drives /api/v1/agents/pulse’s composition so the agent narrative cannot drift from the human read.

1. Signal Alignment — does SPY’s 3-day direction agree with the eleven (or twelve, when news_sentiment is enabled) signal categories? Strong divergence is the loudest “something is wrong here” tell on the page. Start here.
2. Market Status — regime, score, bridge text, and the leading/lagging paragraph. The five-second read once alignment has set the frame.
3. Energy & Commodities — energy_regime is a top-of-funnel macro signal that shifts before equities; transitions here cascade into alignment and breadth before they show up in price.
4. Active Alerts — anything CRITICAL or WARNING needs immediate attention; INFO alerts (bullish confirmations, position reclaims) are context.
5. What Changed — every meaningful delta vs the previous daily snapshot — regime flips, ≥5pt health-score moves, threshold crossings, label flips. Quiet here = quiet day.
6. Key Levels — gamma call/put walls + ZGL. Where are the support/resistance levels for today’s session?
7. Scan raw data sections — colored dots make this fast. Glance for any signal cards whose dots stand out vs your expectation given the regime + alignment read.
8. Economic calendar — upcoming catalysts can override every technical signal. A CPI print at 8:30 next morning means today’s read has a ~14-hour shelf life.

Suggested order in one line

For agents and humans alike, the canonical sweep order is regime → score → alignment → severity → top movers → alerts → calibration drift → AI brief. The dashboard order above optimizes for visual flow; the sweep order optimizes for diagnostic reasoning (“what’s the world look like, does the platform agree, where’s the disagreement, what’s news”). Both end up in the same place.

Human vs agent pulse

A human pulse is a visit to the dashboard at bigclawd.com. The page is intentionally read-once-top-to-bottom — sticky nav links to specific sections; the alignment card always lives at the top. Read time is 30-60 seconds when the day is quiet, 2-3 minutes when alignment is divergent or alerts are firing.

An agent pulse is one HTTP call or one MCP tool invocation:

• HTTP — GET /api/v1/agents/pulse. Returns the stitched payload that previously required /signals/latest + /alerts/active + /signals/alignment composed by the agent. Reuses the same helpers /signals/latest uses so the narrative cannot drift. schema_version mirrors /signals/latest.
• MCP — the market_pulse tool in mcp/mcp_server/tools/public.py wraps /agents/pulse with the standard cache. Use this from any MCP-enabled AI agent (Claude Code, Cursor, ChatGPT desktop) for the same single-call platform state read.

Either way the agent reads the same regime, score, alignment, alerts, and what-changed surfaces the dashboard renders. There is no separate “agent-only” data path.

When to run a pulse check

• Start of trading session — before placing a single trade. The overnight cycle may have flipped the regime; the morning DIX print may have crossed a threshold.
• After major news — Fed announcement, CPI print, major geopolitical headline. Alignment can flip in minutes when the news hits a load-bearing macro input.
• After a regime flip — when the dashboard’s regime label changes, every downstream interpretation of every signal recalibrates. Pulse-check immediately to re-anchor the read.
• Before closing a position — five-minute pulse check before any meaningful position change. Don’t let yesterday’s regime drive today’s exit.
• Mid-session if alignment looks suspicious — a single dot flipping isn’t enough to act on, but if you sense the page disagreeing with your thesis, sweep top-to-bottom and let the platform’s read overwrite your intuition.

The calibration-drift overlay

A pulse check is incomplete without checking calibration drift. /api/v1/ai-brief/calibration-drift?horizon=1d (and 3d, 5d) surfaces whether the AI brief’s directional calls are tracking realized returns — a drift_hit_rate flag means the recent 30d rolling accuracy fell below 45%, a drift_quintile_inversion means the top-confidence quintile is underperforming the bottom-confidence quintile by ≥15 points. Either flag means the brief’s directional read is structurally unreliable right now; humans should weight it down accordingly, agents should pass the flag through to downstream consumers.

The discipline frame

The pulse check exists so you have a habit that’s stronger than your bias. Run it the same way every time — same order, same surfaces, same disposition. When you feel yourself wanting to skip step 1 because “alignment looks fine, I already know,” that’s exactly when the platform has the most to tell you that you don’t already know.

See also

• Five-level signal scale — the per-signal dot levels you scan during step 7.
• Leading vs lagging — which categories deserve the most weight during step 1.
• Hybrid analogue matcher — “compare-to-history” for when the pulse read is ambiguous.
• Implementation: app/signals/routes.py (/agents/pulse), mcp/mcp_server/tools/public.py (market_pulse).