NewAgeSolution
All articles

Data preparation

What multi step products need from evaluation data

· 10 min read

In short

Multi step products need sequence records: the task, each action, the observed result, the recovery when a step fails and an outcome judgement on the full sequence rather than on individual replies.

What multi step products need from evaluation data

Workflows that plan and act over several steps cannot be assessed the way a single reply is. What matters is whether the whole sequence reached a correct end state.

What a useful sequence record contains

  • The original task and any context supplied
  • Every system action with its inputs and returned result
  • Points where the workflow recovered from a bad step
  • The final state, judged against what the user asked for

Failures worth collecting deliberately

Timeouts, malformed inputs, wrong action selection and confident stops before the task is finished. These improve recovery behaviour, and they are rare in data collected only from successful runs.

Judge the outcome

Reviewers should mark whether the task was completed, partially completed or failed, with the step where it went wrong. Step level scores alone will call a sequence good that never delivered the result.

Practical checklist

  • Define the acceptance rule before any volume starts
  • Review a small pilot before committing the full budget
  • Track errors by category, language and reviewer
  • Keep consent, source notes and version history with the files

Before you ask for a quote

A clear brief saves days. Share a sample file, target language or region, expected volume, deadline, quality threshold and any privacy restrictions. A supplier can then price the work on real effort rather than assumptions.

  • Which languages, markets or user groups must be represented?
  • What format does the final file need to arrive in?
  • Who will approve ambiguous cases during the pilot?
  • workflow
  • evaluation
  • action traces

Need this done rather than read about it?

We run collection, annotation, transcription and localization projects for teams who would rather spend their time on the product.

Start a conversation