Troubleshooting Case ContextCopy Anchor Link

If something is not working as expected with Case Context, check the common issues below. Most problems have straightforward solutions.

Empty context warningCopy Anchor Link

What you see: A warning that Case Context is empty or incomplete when you try to use the prompt generator.

Why it happens: The prompt generator draws on Case Context to create tailored prompts. When there is no Case Context (or it has not been approved yet), the generator warns you that prompts will be generic rather than case-specific.

What to do:

• If no one has created Case Context yet, follow the Setting Up guide to create one.
• If a context exists but is still a draft, review it and approve it, or ask a teammate to. Anyone on the case can approve a context.
• If you want to proceed without Case Context, you can dismiss the warning. Claira will still work, but prompts will not be tailored to your matter.

Approval workflow questionsCopy Anchor Link

"What do Draft and Approved mean?"Copy Anchor Link

A context is always in one of two states:

• Draft: anyone on the case can edit it, and the prompt generator does not offer it.
• Approved: editing is locked, and the prompt generator offers it.

There is no pending state and no separate approval workflow to set up.

"Who can approve a context?"Copy Anchor Link

Anyone on the case can approve a draft with Approve. The panel and the PDF and Word downloads show who approved the context and when.

"I need to change an approved context"Copy Anchor Link

Click Revert to draft, make your changes (they save as you type), then approve the context again. Until it is approved again, the prompt generator offers only the other approved contexts. Prompts already generated from it do not change.

Multiple users editingCopy Anchor Link

"The context is read-only for me"Copy Anchor Link

One person edits a context at a time. While someone has a context open and is working in it, Claira holds an edit lock for them, and everyone else sees the context read-only with a note that another user is editing it. The lock is released when that person leaves the context, and it lapses a few minutes after they stop working in it. Open the context again after that to edit it.

While Claira generates a context from documents, the Agent cannot edit it either: it says the context is being generated, changes nothing, and asks you to try again once the generation finishes. See While Claira generates the context.

"This context changed somewhere else"Copy Anchor Link

You see this when the context was updated while you were typing, for example by the Agent in a Revise with Agent chat, in another window, or by a generation. Claira stops saving so that it does not overwrite that change, and your latest edits stay on screen, not yet saved. Choose one:

• Load the latest version shows the context as it is now. The edits on your screen that were not saved are discarded.
• Keep my version saves the text on your screen over the other change. Claira asks you to confirm first, because the changes made elsewhere are replaced.

"Someone else is editing this context now"Copy Anchor Link

You see this when another person took the edit lock while you still had the context open, for example after you stepped away for a while. Your latest changes could not be saved, and the editor becomes read-only. Click Copy my changes to put your text on the clipboard, then Load the latest version to see the context as it is now, and add your text once the other person has finished.

The same questions keep coming backCopy Anchor Link

What you see: The Missing information list shows the same questions after every generation.

Why it happens: A question stays on the list until the documents or the context answer it. Each time Claira generates, it checks every question again and removes only the ones that are answered. Some questions ask for something only your team knows, such as which party you represent. Claira does not work that out from the documents (for example from whose documents are being reviewed), so the question stays until a section of the context says it.

What to do:

• Add the answer to the section the question names. The next generation finds it and drops the question.
• Or click Revise with Agent and tell the Agent the answer. It can update the section and take the question off the list.
• Hide for now only hides the list. Questions that are still unanswered come back after the next generation.

Generated prompt covers more than you asked forCopy Anchor Link

What you see: You type a short request such as "relevance review", and the generated prompt comes back covering several things at once: a relevance pass, a tag for each issue in your Case Context, sometimes a privilege pass as well. Trimming it down takes longer than starting over.

Why it happens: The generator reads your Case Context and your description together. If your description names one kind of review, the generator now builds the prompt around that one dimension only. The issues in your Case Context still do all their work (they define what relevant means for your matter), but they are written into that definition rather than turned into a second set of codes.

What to do:

• Name the review you want. "Relevance review", "privilege review", "PII review". One or two words is enough for the generator to hold the line.
• Ask for issue coding when you want it. Add something like "and tag which issue each document goes to" and the generated prompt includes issue tags. Leave it out and it will not.
• Ask for more than one dimension explicitly. "Relevance and privilege" produces a prompt covering both.
• Say so when you want the wide pass. A description like "full review of everything" tells the generator not to narrow.

If you describe a task that is not a coding review at all (summarizing, translating, building a chronology), the generator does not apply this narrowing, because the shape of the request is different.

Selected documents are too longCopy Anchor Link

What you see: Building Case Context from the documents you selected stops with a message that the documents are too long to process together.

Why it happens: Claira reads the full extracted text of every selected document at once, and the AI model can only take in a limited amount of text per request. Very long documents, or documents whose extracted text is garbled or in a non-Latin script, use up that limit much faster than ordinary text. When the documents' native files are what goes over the limit, Claira tries again once with the extracted text alone, so you see this message only when the text itself is too long.

What to do:

• Remove one of the selected documents and try again. Removing the longest document usually works.
• If you still need that document's content, run it on its own afterwards. Case Context keeps what it already has and adds to it.
• Check in Nuix Discover that the documents have readable extracted text. A document with garbled text adds little to Case Context and uses up much of the limit.

A document was left out, or read from its text onlyCopy Anchor Link

What you see: After building Case Context from the documents you selected, Claira says that some documents were left out and names them. Or, under Source documents linked to Case Context, a scanned PDF or an image shows Read from extracted text rather than Read from native file and extracted text.

Why it happens: Claira sends a document's native file only when it can: the file must be a PDF, image, audio or video file of 15 MB or smaller, the files of all the selected documents must fit together (about 20 MB), Nuix Discover must let Claira download it, and the AI model must be able to read it. A document whose file cannot be sent is read from its extracted text alone. A document that also has no extracted text gives Claira nothing to read, so it is left out and not added to the context's source documents. On a Claira Private case, Claira reads extracted text only.

What to do:

• Build from the left-out document on its own, or with fewer documents, so its file has the room to itself.
• If the file is larger than 15 MB, have it processed for extracted text (OCR or transcription) in Nuix Discover, then try again.
• Check in Nuix Discover that you can open the document's native file.

When none of the selected documents can be read, nothing is changed:

Context not appearing in prompt generatorCopy Anchor Link

What you see: You have created a Case Context, but the prompt generator does not seem to use it.

Possible causes:

1. The context is still a draft. The prompt generator only offers approved contexts. Check the status badge.
2. You are in the wrong case. Case Context is scoped per case. Make sure you are working in the case that has the completed context.
3. The section relevant to your prompt is empty. The prompt generator uses specific sections depending on the task. For example, a privilege prompt draws on the Privilege Indicators section. If that section is empty, the generator cannot use it.

What to do:

• Verify the Case Context status is Approved.
• Confirm you are in the correct case.
• Open Case Context and check that the relevant sections are filled in.

If you have confirmed all of the above and the prompt generator still does not reflect your Case Context, contact support for further assistance.

Need help? Contact support@claira.to