A task is enabled by an external signal that is stored until a task instance is ready to consume it. Signals accumulate in a durable queue and are never lost — each signal is guaranteed to be consumed exactly once. Contrast with 70.71 Transient Trigger, which drops signals when no consumer is waiting.
A compliance audit system monitors regulatory change feeds from multiple jurisdictions. Regulatory updates arrive irregularly — sometimes a dozen in a day during legislative sessions, sometimes none for weeks. Auditors are a scarce resource: the agent pool processes one regulation at a time and is often fully occupied. Signals cannot be dropped — every regulatory change must be reviewed, logged, and acknowledged before the next one is consumed.
The key insight: here, unlike in market data, a signal processed with a 48-hour delay is still valid and valuable. A regulation that took effect last Tuesday must still be reviewed today. The time-value of the signal does not decay to zero on the timescale of consumer availability. Persistent semantics are the correct choice: the queue absorbs burst arrivals, the agent processes at capacity, and every signal is eventually handled — in order, without loss.
| Metric | Signal |
|---|---|
| Queue depth (absolute) | Number of unprocessed signals. Primary operational health metric. Crossing a threshold triggers capacity scaling. |
| Queue age (oldest unprocessed signal) | Age of the oldest signal in the queue. Compliance posture metric — directly maps to unreviewed regulatory exposure window. |
| Consumer throughput (signals/day) | Processing rate. Must exceed signal arrival rate at steady state. Gap between arrival rate and throughput predicts queue depth trajectory. |
| Dead-letter rate | Fraction of signals requiring manual intervention. Rising rate signals data quality issues in the upstream signal source. |
| Node | What it does | What it receives | What it produces |
|---|---|---|---|
| Signal Queue | Durable, ordered queue that persists all incoming regulatory signals. XOR-split: routes to Audit Agent when one is available; loops back to wait state when the agent pool is busy. XOR-join: re-enters the wait state after each acknowledged signal if more remain. | Regulatory change signals from external feeds | Next signal in queue, delivered to the next available Audit Agent |
| Audit Agent | Consumes a single signal from the queue. Reads the regulatory change, cross-references existing policy, and determines what actions are required. Does not consume the next signal until the current one is acknowledged. | Single regulatory change signal from the queue | Audit findings: policy gaps, required actions, affected systems |
| Process Regulation | Generates the compliance response: updates policy documents, flags affected workflows, creates remediation tasks. Produces a structured compliance record for the audit trail. | Audit findings from the Audit Agent | Compliance record: updated policies, remediation tasks, audit trail entry |
| Acknowledge | Confirms successful processing of the signal to the queue. Removes the signal from the durable store. XOR-split: routes back to Signal Queue if more signals remain, or routes to End if the queue is empty. | Compliance record | Acknowledgment token to the queue; final audit record to the durable log |
| Origin of Value | Where it appears | How it is captured |
|---|---|---|
| Future Cashflow | Process Regulation node | Value is created when compliance records are generated. The queue ensures no signal is missed — 100% coverage is the value guarantee. The cost of a missed regulatory signal (regulatory penalty, audit finding) typically far exceeds the cost of processing delay. |
| Governance | Signal Queue + Acknowledge nodes | The queue is the governance artifact. At any moment, the queue depth shows the unprocessed compliance backlog. The Acknowledge step creates an immutable audit trail: every signal has a processing timestamp, a responsible agent instance, and a compliance record. This is the evidentiary chain for regulatory audits. |
| Conditional Action | Audit Agent | The agent is activated only when a signal is available. Idle agents incur standby cost but no processing cost. Cost scales with signal frequency, not pool size. Persistent queuing decouples arrival spikes from agent spend — the organization pays for processing capacity, not burst handling. |
| Risk Exposure | Signal Queue (queue depth) | A growing queue is a lagging compliance posture. Regulations that arrived 10 days ago are unreviewed — the organization is exposed for every day of processing lag. Queue depth is therefore a direct proxy for unhedged regulatory risk. |
VCM analog: Work Token with durable escrow. Each signal is a token held in escrow by the queue. Tokens cannot be destroyed — they can only be transferred to the consumer on consumption and to the audit log on acknowledgment. The escrow guarantees that no token is lost between issuance and settlement.
Regulatory signals arrive during a legislative session at 50 per day for two weeks. The audit agent processes 5 per day. Queue depth reaches 630 signals. Processing the backlog takes 126 days. During this period, the organization is non-compliant with all queued regulations. Fix: establish a maximum acceptable queue depth (e.g., 14 days of backlog). When depth exceeds the threshold, scale the agent pool or escalate to human reviewers. Queue depth should be a first-class operational alert.
The Audit Agent consumes a signal, processes it, generates the compliance record, but crashes before the Acknowledge step. The queue, not receiving an acknowledgment, marks the signal as undelivered and re-delivers it to the next available agent. The signal is processed twice — two compliance records are generated for the same regulation. Fix: use exactly-once delivery semantics with idempotency keys. The compliance record for a given signal ID should be a no-op if it already exists.
A malformed signal enters the queue — invalid schema, missing required fields. The Audit Agent fails to process it and returns it to the queue. The queue re-delivers it. The agent fails again. This signal is at the head of the queue, blocking all subsequent signals indefinitely. Fix: implement a dead-letter queue. After N failed delivery attempts, move the signal to the dead-letter queue for manual investigation. Never block the main queue on a single failing message.
| Variant | Modification | When to use |
|---|---|---|
| Priority Queue | Signals are assigned priority scores. High-priority signals are consumed before lower-priority ones regardless of arrival order. | Not all regulatory changes are equal — a major jurisdiction mandate takes precedence over a minor clarification. FIFO order is suboptimal. |
| Competing Consumers | Multiple agent instances consume from the same queue concurrently. Each signal is delivered to exactly one consumer. | Signal volume exceeds single-agent throughput. Scale out by adding consumers — the queue handles coordination, agents run independently. |
| Deduplicating Queue | The queue detects and drops duplicate signals (same regulation ID received multiple times from different feeds) before delivery. | Multiple regulatory feed sources may emit the same change. Processing duplicates is wasteful and creates conflicting compliance records. |
| Partitioned Queue | Signals are partitioned by jurisdiction. Each partition has a dedicated consumer, maintaining strict per-jurisdiction ordering. | Order within a jurisdiction matters (amendment must be processed after the original regulation) but cross-jurisdiction ordering is irrelevant. Parallel processing across partitions is safe. |
| Pattern | Relationship |
|---|---|
| 80.83 Transient Trigger | The lossy alternative — signals are dropped when no consumer is available. Use when signal time-value decays rapidly and stale processing is harmful. |
| 50.51 Structured Loop | The consumer side of this pattern is typically a structured loop — an agent that drains the queue until empty, then exits. |
| 10.14 Retry-Fallback | Combine with persistent trigger to handle processing failures — failed signals are retried up to N times before routing to the dead-letter queue. |
| 20.22 Human-in-the-Loop | Dead-letter queue signals escalate to human review — persistent trigger is the mechanism, human-in-the-loop is the handler for exceptions the agent cannot process. |
Persistent trigger queues are compliance infrastructure. Any organization operating in a regulated environment needs exactly-once, ordered, auditable signal processing. The pattern is not optional — it is the minimum viable architecture for regulatory signal handling. The competitive differentiation is in queue depth management: organizations that maintain near-zero queue depth have faster compliance postures than peers with growing backlogs.
The moat is in throughput density. Compliance teams that have built AI agents that process regulatory changes faster and with higher accuracy per signal can absorb larger signal volumes without capacity scaling. A 2x improvement in per-signal processing speed is equivalent to doubling the agent pool at zero marginal cost.
Red flag: a compliance system that cannot report current queue depth and oldest unprocessed signal age in real time is flying blind. These two metrics are the minimum observability requirements for any persistent trigger system in a regulated context. Due diligence should treat absence of these metrics as a material operational risk.