You need to already be signed into Audema, and you need the internal tools' shared access
password (the same one used by BuiltWith Research).
🔒 Every one of this run's simulated jobs is fully synthetic — the job-processing engine that generates
durations, success/failure outcomes, and cost figures below never calls a real AI provider and never
spends real money on its own. The ONLY exception, ever, is an explicit, opt-in "Calibrate against a real
build" toggle below, which — only if you turn it on — makes exactly ONE real, disclosed, credit-spending
call before a run starts, purely to measure real latency/cost and parameterize the simulation with it.
Test Scenario
Total: 100%
Spike traffic
Periodic bursts up to a peak virtual-user count
Artificial AI delay
Simulated generation time drawn from a realistic latency distribution (disable for pure queue/concurrency testing)
API failure injection
Simulated Claude / Image API / Deployment / Database error rate
⚠️ Calibrate against a real build
Off (default): this run stays 100% simulated, exactly as before — no real calls, no real spend.
⚠️ Turning this on makes ONE real website-mockup generation call (a real Gemini image
request) before the simulated run starts, and that ONE call spends real image-generation
credits. It is used only to calibrate this run's simulated duration and cost — it does not
repeat, and no other call in this run is ever real.
Leave blank to check the generated mockup image's own URL. Provide a real website URL
instead to verify that specific site serves successfully — this is a single real HTTP
request, made once, never repeated for the rest of the run.
Dashboard
Live Activity
Sites requested
—
Sites successfully built
—
Success: —
Median build
—
P95
—
P99
—
Peak concurrent users
—
Peak concurrent builds
—
Maximum queue
—
Claude errors
—
Image API errors
—
Deployment errors
—
Database errors
—
Average generation cost
—
Synthetic — configured constant, not a real charge