How the decision gets made.

Drift-free weighted rotation that spreads load proportionally to weight.

Weighted random pick — simple probabilistic distribution by weight.

Sample two candidates, pick the less loaded. Near-optimal and O(1).

Routes to the recipient with the most free capacity right now.

Sticky routing — the same key always lands on the same recipient.

Active-passive HA — promote the next tier when the active recipient is unhealthy.

Boosts under-assigned recipients to repay fairness debt over a window.

Highest score wins — blends weight, load, and availability.

One composable score blending value, capacity, fairness, deadline, distance, and margin.

Deadline-aware routing that prefers recipients likely to beat the SLA.

Maximizes expected value — conversion × value × quality × margin.

Caps each recipient or group's share of recent assignments.

Routes to the recipient with the highest predicted outcome value.

Bayesian bandit balancing exploration and exploitation.

Contextual bandit that learns per-arm reward models from features.

Routes to the geographically closest recipient.

Blends distance with weight for proximity-aware load distribution.

Broadcast to N candidates — first to claim wins.

Single winner at the highest bid amount.

Winner pays the second-highest price — incentive for truthful bidding.

Select the N highest bids, all above a floor, or above a reserve.

Durable session that solicits bids, collects them, and clears a winner.

Typed winner slate with prices, reserve reasoning, and pricing-ledger refs.

Throttle or deny market operations under load or budget pressure.

A durable ordered list of outcomes — selected, alternates, eliminated, pending.

Wrap any explainable single-winner strategy into a ranked slate.

Durable batch-level assignment contract optimized across all items at once.

Greedy batch optimizer that deducts capacity as it assigns.

Enforce per-recipient and per-group caps across an entire batch.

Hard requirements, soft preferences, and trait-based filtering.

Reusable coverage area — region, postal set, polygon, radius, or account list.

Exact skill filter, then weighted distribution among the qualified.

Normalizes circuit state, latency, error budget, and saturation into one readiness signal.

Base score plus readiness penalties; blocked targets excluded.

Primary→secondary failover with half-open canary probing.

Route work to humans, teams, queues, AI agents, or service accounts.

Simple work to AI agents; complex or edge cases to human experts.

Compose filter → select → fallback stages into one reusable primitive.

A versioned declaration of a strategy's required facts, outputs, and determinism.

A named, versioned bundle binding a strategy to concrete config.

The immutable, typed bundle of feature values a decision actually consumed.

A durable, tenant-scoped home for stateful-strategy state.

Govern real live traffic with sticky A/B variant assignment.

Run candidate strategies on live traffic without affecting decisions.

Auto-tune strategies and emit reviewable change proposals with expected lift.

Conformance suites that gate strategy activation and promotion.

Bring your own strategy over gRPC, Connect, subprocess, WASM, or in-process.