Understand checks
Checks help you answer: did this run deliver what I asked for? Save clear requirements with a task, then ConsoleX compares the delivery and available evidence with those requirements. Open the report to see what passed, what failed and what still needs your judgment.
For example, a client update might need three labeled sections, a word limit and a clear request for the client's next decision. These are more useful checks than “write a good email.”
Three things to keep separate
| What | What it means | Where you use it |
|---|---|---|
| A metric | One requirement to check, such as a length limit or whether sources are named | Keep reusable metrics in Library → Check metrics |
| A task's acceptance criteria | The requirements you chose and saved for this task | Set them on the task's Checks tab |
| A check report | What happened when one delivery was checked against a saved version of those requirements | Open a run's Check result |
Adding a library metric to a task creates an independent copy. You can adapt it to that task without changing the library or other tasks. See Save and reuse check metrics.
Start with one task
- Decide what a usable result must contain. Open the task's Checks tab. Draft criteria from the task, add your own, or choose library metrics. Review the requirements and save them.
- Run the task. With automatic checks enabled, ConsoleX checks the delivery after the run. You can also check a saved run manually.
- Read the result and its evidence. Open Run history, select a run, then open Check result. A verdict is most useful when you can see why it was reached.
- Decide what to do next. Use the result, review an uncertain judgment, revise the work, or improve the task for next time.
The same saved task standard can be used for deliveries submitted through an external agent. The client's guide explains what evidence its integration can submit.
Understand the result
| Result | Meaning | A useful next step |
|---|---|---|
| Passed | The checks found the requirements met within the available evidence | Read the delivery; you can spot-check it yourself |
| Failed | A check found a requirement that was not met | Read the reason and evidence, then decide whether the work or the criterion needs changing |
| Needs review | The evidence was not enough to decide one or more items | Inspect the result yourself or supply the missing evidence |
| Check error | The check could not finish | Retry the check or review the saved result yourself |
A completed task run and a passed check answer different questions: the first says the work finished; the second says how its delivery matched the standard. A missing source or unreadable file may leave a check undecided rather than prove the result wrong.
How ConsoleX checks a requirement
Rule checks test a specific condition, such as text length or a recorded tool call. AI checks interpret the requirement and available evidence, such as whether a client update clearly explains the next decision. AI checks use the configured checking model and have their own usage cost.
Write criteria that name the result you need and the evidence that could establish it. A citation link alone does not prove the source supports a claim, and a file path alone does not prove its contents were checked. See Set acceptance criteria for evidence types and checking settings.
Your judgment and the next run
You can record a human review after reading the delivery. Your saved conclusion is shown alongside the automatic check and is used in the result summaries; the automatic verdict and evidence stay available.
If the check missed a problem or judged too harshly, review the result and adjust the criteria if needed. If the task's method needs work, request an improvement and review the proposed changes before applying them. Reviewing one delivery does not automatically rewrite the task or its standard.
Choose a guide
| I want to… | Guide |
|---|---|
| Decide how a task should be checked | Set acceptance criteria |
| Understand a verdict, see evidence or check again | Read run results and checks |
| Record my own conclusion or handle a wrong check | Review a result yourself |
| Save requirements for use in other tasks | Save and reuse check metrics |
| Change how the task does the work next time | Improve a task |