Skip to content

Waiting, queues and errors

A project does not always start the second you press play, and a large download does not always arrive instantly. The platform tells you why you are waiting, so you can tell a slow moment apart from a stuck one.

While a project waits to start

The project row shows a short line under the name explaining the wait:

What you see What it means
In queue / In queue (2) The project is ready but not running yet. The number is its place in line.
…· your plan runs N at a time The wait is your own plan's limit: you already have N projects running. Finish or wait for one of them.
…· starting capacity The wait is ours, not yours. The processing fleet is scaling up; nothing about your plan is blocking it.
Queued -- starting a machine for it A machine is being booted specifically for this project. It usually clears in a couple of minutes.
starting a machine to train it Same thing, during project creation. The progress bar animates instead of sitting at 0%.
No progress for N min -- may be stuck Advisory only. Nothing is killed automatically.

The two reasons are not the same problem

your plan runs N at a time is a capacity limit on your account — running fewer projects at once, or a larger plan, changes it. starting capacity is our fleet warming up, and no plan change makes it faster. The line tells you which one you are looking at.

Projects that are running show a live ETA under the progress bar (45min, 2h 10min), refreshed about once a minute.

Note

Express census matching can legitimately sit on the same counter for a long time. The "may be stuck" warning only appears after 45 minutes without progress for that phase, precisely so a normal Express creation does not raise a false alarm.

While a download is being prepared

Big files — merged datasets, quota exports, target files, SPSS output — are built by a worker rather than in your browser. When you ask for one, a Preparing your download dialog opens:

  • Queued. It starts as soon as the current build finishes. — or Queued. N file(s) ahead of yours.
  • Building the file… once it is actually being written.
  • Elapsed: MM:SS, a live timer.
  • "The file downloads on its own when it is ready. You can keep working meanwhile."

Continue in background closes the dialog only. The build is not cancelled: you can navigate away, and the file still downloads by itself when it is ready.

One download at a time. If you request a second file while one is still being built, you get A download is already being prepared — " is still being built. Wait for it to finish before starting another download." — with a View progress button that reopens the dialog for the job already running.

If a download fails you get Download failed, with either the specific error from the worker or, when the connection dropped, "The download took too long and the connection dropped. Large SPSS exports are the slowest — try again, and if it keeps failing tell your admin the request is timing out."

When a project fails: error codes

A failed project shows a sentence, not a traceback. Click the state badge to expand it. The format is always the same:

[NI-300] Training stopped: the data file has a column or value the pipeline
could not process. Check the file (headers, empty columns, encoding) or send
the reference to support. Ref 3f9a2c1d.

Three parts: a stable code, what happened and what to do, and a reference. The reference is the one thing support needs — it points at the technical detail (exception, traceback) that the platform keeps internally and does not show you.

Code What happened Who usually acts
NI-100 The machine ran out of memory Us — a bigger machine, or fewer columns. You can retry.
NI-110 The machine ran out of disk space Us. Retry afterwards.
NI-200 Project files could not be read or written in storage Us.
NI-210 A network connection dropped Retry — your progress is kept.
NI-300 The data file has a column or value the pipeline could not process You: check headers, empty columns and encoding. If the file is fine, send us the reference.
NI-310 The project could not resume from its last saved point Us, then retry.
NI-400 The platform database was unavailable or out of date Us.
NI-500 The machine running the project was stopped before it finished Retry — it resumes from the last saved column.
NI-900 An internal error stopped the project Us. Send the reference.

Retrying is safe

Where the table says "retry", pressing play again resumes from the last saved point rather than starting over. Both error fields are cleared when a retry succeeds.

Swift and Express specification errors are not in this list: a spec that does not match the file gets its own explicit message, already written as a sentence about your file.