Watch the overview
About 2 minutes. A narrated explainer; the same words are below.
Read the video text
- 0:00You upload a survey file and the checks come back with a list.
- 0:04Some say error, some say warning, and the totals also count skipped checks. What should you do with each one?
- 0:13Think of it as a proofreader's marks on a report.
- 0:16Each mark tells you how much attention it needs, and a page nobody read is not a clean page.
- 0:23Before reading the totals, confirm you uploaded the right file and chose the right check template.
- 0:29A result only describes the checks that ran on that file.
- 0:34An error blocks the file. It cannot go to Send for review and co-sign until it is fixed or covered by a recorded exception.
- 0:48That block also holds at co-sign and at certification.
- 0:53A warning does not block anything.
- 0:55But it is a request to look. Decide whether it matters for how the data will be used, and keep a note of your decision.
- 1:03A skipped check did not run.
- 1:06The result counts these as Skipped. The template decides which checks run, so ask why. A skipped check is not a pass.
- 1:16First, read the rule, the message, and the example rows it shows. Look at the neighbouring records too.
- 1:22Then decide: is this an export fault, a wrong template, or a genuine measurement?
- 1:29If the export is wrong, fix the source and reupload. Never edit a value just to make an error disappear.
- 1:37Checking again puts the file back under review, and the earlier result is no evidence for the new file.
- 1:44If the value is genuine, someone with the right permission can choose Record exception.
- 1:48The reason must be written down. Two processing notes cannot be dismissed at all.
- 1:55A clean result means the checks that ran found nothing. It does not prove the measurements are right.
- 2:03An engineer still has to judge whether the data suits its purpose.
- 2:09Errors need action. Warnings need a look. Skipped checks need an explanation.
- 2:15Keep a short review note for every decision. The full guide is at TMH18 dot com, slash articles.
Start with the file and the selected checks
Before interpreting the totals, confirm that the file and check template are the ones you intended to use. Check whether analysis has finished. A result describes the checks performed on that submission; it is not a guarantee that every possible survey issue has been tested.
For each finding, read the rule reference, message and any row or field information provided. Look at the source record and its neighbouring records. A surprising value may be a genuine measurement, an export problem or the result of choosing an unsuitable template. The next action depends on which explanation the evidence supports.
Errors need action before approval
An error identifies an issue that must be resolved or covered by an allowed, recorded exception before the file can proceed through approval. Start by checking the source data and the rule that generated the finding. Do not change a measurement merely to remove an error from the screen.
If the export is wrong, correct it at the source and submit a replacement through the available workflow. If the value is supported by survey evidence, check whether the finding allows an exception and whether you have the necessary permission. Some findings cannot be overridden. A written reason should explain the evidence and the decision, not simply say that the value was checked.
Review warnings and understand skipped checks
A warning flags something that needs attention, even where it does not block approval. Confirm whether it affects the intended use of the dataset. Record a review decision where the workflow allows it, and retain any supporting survey information.
A skipped check did not run. Missing inputs or an inapplicable condition can prevent a check from being evaluated. Read the reason shown and decide whether more information or a different template is needed. Do not report a skipped check as evidence that the data passed.
- Read the rule and the affected records.
- Separate an export fault from a supported measurement.
- Investigate warnings before handing the file over.
- Account for checks that did not run.
Close the loop after a correction
After replacing a file, wait for its new analysis to complete and read the new findings. A correction can affect more than one rule. The previous result is not evidence for the revised bytes.
Keep a concise review note: what was found, the source evidence consulted, what changed and why the revised file is suitable for the next review step. This gives the co-signer and final certifier a clear basis for their own decisions.