Free concept playground · no account

Workflow Studio

Watch a pipeline halt exactly where the model's output lied.

Workflow Studio

AvailableFree

A five-step AI pipeline. One step emits malformed output — the pipeline halts there and everything downstream is skipped. Toggle the broken step to “fix” it and watch the flow reach the end.

  1. 1Extract fields✓ ok
  2. 2Classify intent (tool call)✓ ok
  3. 3Score priority✗ priority expected integer 0–5, got "🔥 super urgent"; sla_hours expected number, got "asap"
  4. 4Human checkpointskipped (upstream halt)
  5. 5Route & notify (mutating)skipped (upstream halt)

✗ Halted at "Score priority" — only 3 of 5 steps ran.

The pipeline halts where the output liedone bad step, everything downstream skipped

The pipeline reached step 3 and stopped: that step's output didn't match what the next step expected, so nothing after it could run. 2 downstream steps were skipped.

An AI pipeline is only as reliable as its weakest hand-off. When a step emits malformed output — wrong shape, missing field, a hallucinated value — the failure surfaces downstream, often far from the real cause. Validate every hand-off at its boundary so the pipeline halts loudly at the broken step, not silently three steps later.

Was this playground useful?

Workflow studio

Watch a five-step AI pipeline run — and halt exactly where one step's output lied. When a step emits malformed output, the pipeline stops there and every downstream step is skipped. Toggle the broken step to "fix" it and watch the flow reach the end. Runs entirely in your browser (0 uploads, works offline).

How to use it

  1. Read the pipeline top to bottom — each step hands its output to the next.
  2. Find the step that failed (its output didn't match what the next step expected) and hit fix this step.
  3. Watch the halt point move and the skipped steps light back up as the flow completes.

What this clears up (the fundamentals)

  • A pipeline is only as reliable as its weakest hand-off — one malformed output breaks every step downstream of it.
  • Failures surface far from their cause — a bad value in step 2 often throws in step 4, so the error message points at the wrong place.
  • Validate at every boundary — check each step's output against what the next step needs, so the pipeline halts loudly at the real break, not silently later.
  • Skipped ≠ passed — steps after a halt didn't succeed, they never ran. Counting "green" steps hides the break.

Where it's used

An AI-engineering literacy tool — the fastest way to feel why orchestrated LLM pipelines need validation at every hand-off. It's a miatz build-lab concept playable; learn to build it yourself.

FAQ

Why does the whole pipeline stop at one step?

Because each step consumes the previous step's output. If that output is malformed, the next step has nothing valid to work with, so it can't run — and neither can anything after it.

Why do failures show up in the wrong place?

Because a bad value flows downstream until some later step chokes on it. The step that throws is rarely the step that introduced the problem; you have to trace back to the first bad hand-off.

How do real pipelines prevent this?

By validating each step's output against a schema at the boundary, so a malformed hand-off is caught and reported at the step that produced it, not three steps later.

Is anything uploaded?

No. The pipeline is simulated entirely in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.

Limits

A teaching model with a fixed five-step pipeline and a single seeded failure — real pipelines have branching, retries, and many failure modes. The lesson (validate every hand-off) transfers directly.

Related

Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.

Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.

Ninety playgrounds. Zero setup.

Every concept here is playable free, no account — and inside the program you learn to rebuild the machinery yourself.