SimpleFunctions Thesis Monitor Setup

When to use

• After creating a thesis with sf-thesis-build, ready to put it into autonomous monitoring
• Adjusting cadence or kill conditions on an existing thesis
• Pausing or resuming thesis monitoring without removing the thesis

Prerequisites

npm install -g @spfunctions/cli
sf login

You need an existing thesis id. Create one first with sf-thesis-build.

How the heartbeat works

The SimpleFunctions heartbeat is a per-thesis monitoring loop that:

1. Polls news and social feeds for material to the thesis
2. Runs LLM evaluation at the configured cadence
3. Updates node confidences when signals arrive
4. Trips kill conditions if confidence drops below threshold or price levels are hit
5. Optionally triggers closed-loop intents (when allowed)

Each thesis has its own heartbeat config — different thesis, different cadence.

Instructions

Step 1: Get current heartbeat state

sf heartbeat <thesis-id> --json

Returns:

• cadence: how often the heartbeat fires (e.g., "30m", "1h", "24h")
• evaluationModelTier: which LLM tier (cheap, standard, heavy)
• monthlyBudgetUsd: max spend per month
• killConditions: list of conditions that pause the thesis
• closedLoop: closed-loop intent behavior
• status: active, paused, killed
• lastFiredAt, nextFireAt
• monthlySpend so far

Step 2: Configure cadence

sf heartbeat <thesis-id> --cadence 1h

Cadence guidelines:

• 5min - 30min: high-stakes, near-resolution markets (election day, FOMC day)
• 30min - 2h: active monitoring (most theses)
• 6h - 24h: long-tail theses with slow news cycle

Faster cadence = more LLM cost. Match cadence to actual news cadence.

Step 3: Set evaluation model tier

sf heartbeat <thesis-id> --evaluation-tier standard

Tiers:

• cheap: light model (e.g., Haiku) — for routine ticks, costs ~$0.001/tick
• standard: balanced (e.g., Sonnet) — for normal evaluation, ~$0.01/tick
• heavy: top-tier (e.g., Opus) — for major signals or scheduled deep eval, ~$0.10/tick

Most theses run standard for routine, heavy for triggered evaluation.

Step 4: Set budget

sf heartbeat <thesis-id> --monthly-budget 5.00

Hard cap on LLM spend per month. When hit, the heartbeat auto-pauses and emails the operator.

Step 5: Define kill conditions

sf heartbeat <thesis-id> --kill-conditions '[
  { "type": "node_confidence_below", "node": "n1.2", "threshold": 0.3 },
  { "type": "market_price_above", "ticker": "KXFEDDOVE-26JUL", "price": 80 },
  { "type": "drawdown_above", "percent": 15 }
]'

Kill condition types:

• node_confidence_below: pause if a specific node drops below confidence
• market_price_above / market_price_below: pause if a price crosses threshold
• drawdown_above: pause if any linked position drawdown exceeds %
• signal_keyword: pause if a signal contains a keyword (e.g., "ceasefire", "resignation")

When tripped, the thesis status becomes killed, monitoring stops, and the operator is notified.

Step 6: Configure closed-loop intent behavior

sf heartbeat <thesis-id> --closed-loop-entry true --closed-loop-exit true --json

Closed loop = the agent can place orders without human approval when conditions are met.

Safety: even with closed-loop on, generated strategy intents default to autoExecute: false. The local runtime must still evaluate and execute through the explicit intent path.

Step 7: Pause / resume

sf heartbeat <thesis-id> --pause --json   # pause the heartbeat (thesis stays active)
sf heartbeat <thesis-id> --resume --json  # resume

Pause keeps the thesis lifecycle active. To fully deactivate, use sf unpublish + adjust thesis status via dashboard.

Step 8: Verify state

sf heartbeat <thesis-id> --json | jq '{cadence, status, nextFireAt, monthlySpend, killConditions}'

Confirm the config landed.

Recommended starting profile

For a new thesis with moderate news cadence:

sf heartbeat <thesis-id> \
  --cadence 1h \
  --evaluation-tier standard \
  --monthly-budget 3.00 \
  --closed-loop-entry false \
  --closed-loop-exit false \
  --json

Then layer in 2-3 kill conditions specific to the thesis.

Errors and recovery

• `Permission denied`: heartbeat is user-scoped, ensure sf login matches the thesis owner.
• Budget exhausted, auto-paused: review monthly spend, raise budget or reduce cadence, then --resume.
• Kill condition tripped, status=killed: review the trigger, decide if it's a real signal or noise. If noise, edit conditions, set thesis status back to active.

Related skills

• sf-thesis-build — create the thesis being monitored
• sf-portfolio-review — see thesis-linked position state
• sf-discipline-audit — adversarial review when kill conditions trip

Reference docs

• <https://docs.simplefunctions.dev/concepts/heartbeat>
• <https://docs.simplefunctions.dev/build/thesis-lifecycle>
• <https://docs.simplefunctions.dev/api-reference/thesis>
• <https://docs.simplefunctions.dev/concepts/risk-gates>