AI Automation Agency Deliverable & Resilience Workbench

Inbound B2B Lead AI Qualifier & CRM Sync IDLE / READY

⚡ Failure-Mode Chaos Injection (Stress Test Agency Resilience)

Status

READY

Total Latency

0 ms

Token Cost

$0.00000

Resilience Score

100%

Nodes Executed

0 / 5

EXECUTION TRACE & AUDIT TRAIL

[READY] Simulator initialized. Select any node to inspect or click 'Execute Pipeline'.

Node Inspector & Client Deliverable Packager Select a node

Configuration: Inbound Webhook Receiver

Timeout Limit (ms)

Max Retries

LLM Model Target

Temperature

Confidence Threshold 0.85

DETERMINISTIC ROUTING OUTCOME

Enterprise Lead -> High Priority HubSpot Pipeline + Dedicated Account Exec Slack Ping

Contractor Client Deliverable & Blueprint Packager

Package this workflow simulation, resilience benchmarks, and failure mitigation SOP for agency review.

Agency pipeline fixtures, recovered failures and export limits

Your browser does not support video.

Read the explanation

The default lead pipeline contains five nodes with latency labels forty five, two hundred twenty, two hundred ninety, sixty five and sixty milliseconds. Their sum is six hundred eighty. The classifier alone has assigned cost point zero zero zero eight four dollars. Bars compare the six hundred eighty fixture total with eight hundred milliseconds of scheduled animation waits at point two five pixels per millisecond. Each node actually waits one hundred sixty milliseconds regardless of its latency label. Neither total is an observed network benchmark. The simulation does not call a model, webhook, enrichment service, CRM or Slack. Reset records success-shaped export telemetry while the visible status reads ready and nodes zero of five, before any execution. The rate limit toggle assigns a failure to the classifier, increments failed node count and then marks its fallback successful without sending a request. It adds three hundred fifty latency units, so the caption rises from six hundred eighty to one thousand thirty milliseconds. Bars use point two pixels per unit. All five nodes succeed, but the failure count remains one. Final success requires zero failed nodes, so this recovered run is labeled degraded rather than success, with assigned resilience eighty eight point five. Its log says exponential backoff and retry, but the branch contains no exponential delays, real secondary queue or actual provider request. A single local branch writes the result and continuing nodes. Schema drift targets enrichment or guardrail nodes. In the lead fixture it halts at the second node after one successful webhook, recording one success, one failure, forty five latency units and zero cost. A webhook timeout halts at the first node with zero successes and zero accumulated latency. Bars use three pixels per latency unit. Both write a dead letter caption and assigned resilience fifty four without sending anything to a queue. Timeout, retry and temperature inputs are displayed configuration but not read by the engine. The guardrail slider changes its displayed value only, not routing. Export preview uses fixed three thousand timeout metadata and reset telemetry; JSON and handover downloads export authored fixtures. The handover three retries and observed latency wording does not establish execution evidence. Paired native simulations, actual downloads, resize and local playback preserve these limitations without public publication or exhaustive verification.