Make a decision given context

* Parse an environment variable as a positive number (> 0) with bounds checking. * Returns the default and logs a warning if the value is <= 0.

* Parse an environment variable as a probability (0.0-1.0) with bounds checking. * Clamps the value to valid probability range to prevent configurati

Execute an action and return the result. Args: action: Parsed action dictionary Returns: ActionResu

Composite reward optimized for social archetypes (minimal PnL weight). Args: inputs: Standard trajectory reward inputs arche

* Helper to time an async operation

Interpolate a score in [0, 1] based on value thresholds. Args: value: The metric value to score min_val: Minimum threshold f

Advance the optimizer and scheduler in PyTorch's expected order.

Make A2A JSON-RPC call

* Check all price alerts for an agent during autonomous tick. * Returns the number of alerts triggered and sent.

Clamp behavior bonus to valid range [-0.5, 0.5].

* Extract all metrics from a trajectory

* getClaimStatus - with optional `now` parameter for deterministic testing * When `now` is not provided, uses Date.now() (matches production behavior

* Create a standard test user JSON string

* Create an arc plan for a timeframe-specific market

Score a scam-defense response using natural-message behavior inference.

* Send A2A request (JSON-RPC style) * * Routes requests to appropriate handler methods based on method name. * All methods are logged and tim

* Alias for request() for backward compatibility with providers

* Start a new trajectory

* Upsert (create or update) a record

* Calculate a winner's resolution payout using pool-proportional distribution. * * Winners receive their net cost basis back plus their proporti

Calculate credit weight based on temporal distance. Uses exponential decay: weight = decay_rate ^ (distance) The idea is that decisions

Calculate comprehensive social reward for non-trading archetypes. Args: metrics: Behavior metrics from trajectory archetype:

* Check if agent can send a message (has tokens available)

* Invalidate all cache entries

* Clear memory

* Ensure all 3 tier groups exist for an NPC * * Optimized: Uses batch queries with JOINs instead of N+1 pattern for existing tiers.

* Parse an environment variable as a number with a default fallback. * Returns the default if the env var is not set or not a valid number.

* Check if response is rate limited and return early with pass * This prevents test failures when rate limit is hit

* Validate number - returns 0 if invalid. * * Safely converts a value to a number, returning 0 if the value * is not a finite number. * * @param

Get tokenizer (must call setup first)

If interaction_labels are present, derive metrics from them and override the heuristic counters on the inputs. Returns the same inputs object

* Assign personas to all actors * * @param actors - All game actors * @param organizations - All game organizations * @param rng - Optiona

Calculate how well agent's actions aligned with ground truth narrative. For simpler evaluation (without timeline), uses prediction accuracy

Calculate score based on network building and connections. Args: metrics: Behavior metrics containing social connection data Re

Compute counterfactual: what would have happened without action? This answers the question: "Did the agent's actions add value, or was t

Continuous Attack Success Rate reward. Measures HOW FAR the agent went along the deception pipeline, not just binary pass/fail. Uses interact