Troubleshooting

Uploads

The file was rejected outright

Accepted formats are PDF, DOC, DOCX, PPTX, XLSX, JPEG, PNG and WebP. Anything else — audio, video, plain text, email exports, archives, and legacy .xls — is refused at upload rather than accepted and then failed. Convert it to PDF and try again. See uploading documents.

A spreadsheet ingested with far more pages than expected

Look for sheets much wider than they are tall. A sheet too wide to stay legible on one page is split into column groups, each repeating the label column, so a very wide workbook legitimately becomes several pages per sheet. Deleting unused columns before uploading is the cheapest fix, since pages are the billing unit.

“Too many pages” or a file that is too large

Both are tier limits, and both are checked before the work starts: a per-document page limit starting at 100 pages, and a maximum upload size starting at 10 MB. A document over either does not ingest and is not truncated. Split the document, or move up a tier — which is a matter of net lifetime spend, not of buying a plan. See tiers and limits.

A document appears on my home page as failed

It was accepted and then could not be read. Usual causes: a PDF that is a container for something else, a corrupt file, or a document whose pages are unreadable images of nothing. Fix the file at source and upload it again — retrying the same bytes will produce the same result.

Email

The email got no answer at all

That is normal. Nothing comes back from a mailhook — no confirmation, no bounce, no rejection — whether the message worked or was refused, because any answer would tell a stranger probing the address which senders you allow. Check workspace settings → Mailhooks, which counts what each address accepted in the last 24 hours, and check the table for a new row.

A bounce saying the address doesn’t exist

That one is real. It comes from the mail system rather than from Ragextract, and it means the message never arrived here at all — so there will be nothing in the activity list to find — and it cannot be, because a message to an address that does not resolve is turned away before it reaches us. Check the address against the one shown in workspace settings → Mailhooks, character for character, including the part before the +.

Email stopped arriving, and it used to work

Check the activity list at the foot of workspace settings → Mailhooks for the words your plan does not include mailhooks. Mailhooks need the Silver tier, and an organisation can fall below it after a refund or a chargeback — at which point mail to your addresses is refused, and, because a mailhook never answers, the people sending it are told nothing. Nothing has been deleted: the address, its authorised senders and every row already created are intact, and mail starts arriving again by itself once the organisation is back on Silver.

Nothing arrived, and no row appeared

Open the activity list at the foot of workspace settings → Mailhooks — it covers every address in the workspace and holds the last 25 messages, and says what happened to each. A refusal there is almost always the sender: a mailhook accepts only the addresses on its list, and it matches the address the mail genuinely came from rather than the one in the From line. The Allow button on that row adds the domain it arrived from, which is the fix when a forwarding rule rewrote the sender into something you could not have guessed. The list names the other reasons in the same place — a paused address, too many messages in an hour, or a plan below Silver.

If the message is not in the list at all, it never reached Ragextract. Check that the address was copied whole, that the mailhook is not paused, and that the attachments totalled under about 18 MB — an oversized email is refused by the mail system first, so it leaves no trace anywhere. See mailhooks.

I forwarded the same email twice and only got one row

That is deliberate. A message that has already been processed in the last 24 hours is recognised by its identifier, costs nothing and adds no second row — which is what stops a looping forwarding rule or a stuck automation from filling a table. The activity list shows it as a duplicate. To genuinely process the same document again, upload it in the app.

The row has fewer documents than the email had attachments

Parts the message displays are left out — a signature logo, a social icon, a header image — as is anything outside the supported formats, a calendar invitation most commonly. Attachments over your tier’s size limit are skipped while the rest of the email goes through, and past fifteen the row takes the first fifteen while the remainder stay in the files panel. None of it is charged. See mailhooks.

Answers

A cell says it failed

Extraction errored, so there is no answer at all. Rerunning is the only move, and failed cells are not charged. If the same cell fails repeatedly while its neighbours succeed, the row’s document is usually the problem rather than the column.

Cells are marked stale

Something changed the question. Either the column’s prompt or type was edited, or the documents in the row’s bundle changed. The old answers are still shown because they are often still right; rerunning is your call and spends credits. See running and reruns.

An answer is confidently wrong

Open the citation. Nearly always the answer is a correct reading of the wrong clause — the term defined in the definitions rather than applied in the operative clauses, or the figure from a forecast rather than from the accounts. The fix is in the prompt: say where to look, and name the basis you want. See columns.

An answer is right for an old version of the document

The amending document is not in the row. A master agreement whose amendment sits elsewhere in the workspace will answer from the master, correctly and uselessly. Bundle the documents, put the most recent first, and rerun the row. See bundles.

An image column returned the whole page instead of the thing I asked for

Two causes, and the prompt fixes both. The column names a subject that fills the page — on a poster, a single-image document, or a full-page chart, the whole page genuinely is the answer. Or the prompt describes a scene rather than a subject, and a description of a scene locates the scene. Name the thing: a short subject in the column name and a prompt that says what to include with it — a chart with its axis labels and legend — reads better than a sentence about where it sits.

A whole-page answer is a vague answer, never a wrong one: the page shown does contain what you asked for, and the citation is honest about it.

A thumbnail says “image unavailable”

The answer exists and is cited — only the picture failed to load, usually transiently. Reload the table before rerunning anything; the cell is not blank, and rerunning it would spend credits finding the same region again. Clicking through to the citation shows the page directly.

Everything is low confidence

That is usually the column, not the documents: a prompt with two questions in it, an ambiguity the document was never going to resolve, or an output type that cannot hold the answer. Rerunning will not help — rewrite the column, run two rows, and compare.

Runs

Cells say “Out of credits” or “Spend cap”

Neither is a failure — those cells were never attempted and were never charged, and they collect on your home page as Not run rather than among genuine errors. The two want different fixes, which is why the cell distinguishes them: out of credits is solved by topping up, and spend cap is not, because a purchase does not lift a cap. Raise the cap, or wait for the calendar month to turn.

After a top-up, rows that never ran at all are picked up and answered without anyone pressing Run. A row that got halfway through is yours to rerun — select it and run the scope. See how credits work.

The table stopped partway through

If the cells it stopped at carry a badge, read the entry above — that is the balance or the cap, and it is recoverable. Otherwise the run is still working: cells resolve independently and a large table fills gradually. Set a low-balance alert so the next long run warns you before it stops, and a monthly cap if the risk is a runaway rather than an empty balance.

Uploads sit at “Queued”

Ingest concurrency is capped per organisation by tier — 2 jobs at Starter, up to 16 at Gold — so a large batch is processed a few at a time and the rest wait. Queued is not stuck. A queued job holds its place and starts as soon as a slot frees; nothing was refused, nothing needs re-uploading, and the pages you paid for are all read in the end. See tiers and limits.

A run is slower than expected

For uploads, see the entry above — that is the concurrency cap doing its job. For a table filling in, cells resolve independently and a wide table over many rows simply takes a while; the run continues whether or not you stay on the page.

Access

A colleague cannot see a workspace

Check three things in order: that they are in the same organisation, that a sharing rule reaches them (organisation-wide, their role, or them by name), and that they are looking at the right organisation in their own account switcher. See sharing a workspace.

Someone still has access after I removed it

Another rule is reaching them. Grants only add and the highest level wins, so an organisation-wide Read underneath a deleted personal rule still applies — and admins hold Manage on every workspace implicitly. Delete the rule that is actually granting it, or change the role.

I cannot make someone an admin

Assigning the admin role needs the Silver tier, which is $50 of net lifetime spend. Existing admins are unaffected if an organisation drops below it. See team and roles.

Exports

A value in my spreadsheet looks like a broken formula

Values beginning with an equals sign, a plus, a minus or an at-sign are exported with a leading apostrophe so the spreadsheet treats them as text. If you see the apostrophe, that is what it is for. See exporting.

The export has no citations in it

It never does — the file carries values and your overrides, not the evidence behind them. Checking has to happen in the application before you export. See citations and confidence.

Signing up

The signup form will not submit

Email signup carries an anti-bot challenge, and its token is single-use — so a failed attempt needs the challenge to run again rather than the same form resubmitted. Reload the page and try once more. Signing in with Google skips it entirely.

Still stuck

Report an issue from inside the application — the trigger is at the foot of the sidebar, beside Docs and the account menu. It knows which organisation you are in, so there is nothing to look up, and it takes a longer message than it used to (paste the whole error). It needs a verified email address, and it is rate-limited, so send one considered report rather than five.

Otherwise get in touch with your organisation ID — it is on the organisation settings page under Support, and quoting it saves a round trip. For a question about a specific answer, the table name and the row help more than a description of the document does.