> For the complete documentation index, see [llms.txt](https://docs.docbits.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.docbits.com/overview-and-basics/release-notes/incremental-updates-12-25-august-2026.md).

# Hotfixes 12–25 August 2026

*What changes in the DocBits production upgrade on 25 August 2026, covering everything since the 12 August release. Each service lists the version being deployed, then what's new or fixed in plain language. Services not listed had no customer-facing changes.*

***

## Highlights

* **Stricter organisation isolation.** A security sweep closed several places where data from one organisation could be read or written from another: document scripts, sub-organisation user lists, group memberships and the processing token a document carries through the pipeline are now all checked against the caller's organisation. Approvals also enforce four-eyes properly: the second approver must be a different person than the first.
* **Documents stop getting stuck.** Four separate causes of documents hanging forever were fixed: exports that stayed in "Exporting" after being denied, restarts that froze when a processing step crashed, barcode splits that never reported back, and the accounting screen hanging on "Preparing…". In each case the document now either finishes or shows a real error you can act on.
* **Credit notes are recognised as credit notes.** XRechnung 3.0, 3.0.1 and 3.0.2 credit notes in CII syntax, pure CII credit notes and ZUGFeRD 2.4 / Factur-X 1.08 documents are all classified correctly now, with the total read from the right field. Scanned documents that mention both "invoice" and "credit note" are resolved by which keyword sits closer to the document type, and amounts turn positive again when you reclassify a credit note back to an invoice.
* **PO matching does arithmetic you can trust.** Tolerances are compared as exact decimals instead of floating-point values, are based on the purchase order value, and invoices that reference several purchase orders are matched against all of them. Columns you never mapped no longer distort the line amount check, and when required columns are missing, the error names them.
* **Workflow runs keep their work.** A workflow that writes a field value now writes it onto the document in a way a later export can't silently revert. Retried triggers no longer discard what the run already did, and two triggers hitting the same document queue up instead of stealing each other's lock.
* **Password-reset e-mails send again.** They were silently never leaving the server. The reset form also shows real feedback after you submit, and the response no longer reveals whether an account exists.

***

## Web App — `10.55.0`

### Signing in and accounts

* Password reset works end to end again: the mail arrives, the form confirms the submission, and the answer is the same whether or not the address has an account.
* If your organisation requires two-factor enrollment, the login screen now says so instead of failing without a message.
* Administrators can no longer switch on organisation-wide MFA enforcement before login enrollment is available, which previously could lock people out.

### Validation screen

* The zoom slider now goes up to 150% (it used to stop at 80%), and zooming into a table works past the container width instead of doing nothing.
* Empty amount fields count as 0 instead of firing an error toast, and a double click on the document image is ignored when no field is selected.
* The banner shown when another session holds the document lock had no text; it explains itself now. Tagging a table no longer triggers a false "document was modified externally" warning about your own change.
* In the AI table, a column remap that would unmap another column asks for confirmation first, and values that aren't numbers are flagged in AMOUNT and NUMBER columns.
* The "Extracted table" tab links to manual table training again when it is empty.
* Item numbers in the Compare line-item table are shown as identifiers, not rounded like amounts.
* Approver fields resolve user and group ids to names, so they never show a raw id or sit empty. Task deadlines are converted through one UTC-aware path, so every viewer sees the same date.
* Documents sent back to validation show a loading spinner instead of a dead screen while they are prepared.

### Accounting

* Splitted line items keep their % sign after pressing Enter, and 0 % is accepted as a value.
* In the account filter, Enter commits the first matching account instead of doing nothing.
* Flexdimension characters are mapped by dimension id, so dimensions land in the right column even when the order differs.
* A failed accounting preparation recovers with an error message instead of hanging on "Preparing…" forever, and reopening a document no longer serves stale data from the previous one.

### PO matching

* Opening PO Matching without every mandatory column mapped is possible again; when something needed is missing, the message names the exact columns.
* Columns that aren't mapped to anything are hidden when the screen opens, after asking you once, and they no longer flow into the line amount calculation.
* The matched quantity refreshes after saving, and the missing-column popup routes you to Field Validation where you can fix it.

### Dashboard and search

* Dropdown-based columns (invoice type, status and similar) show their label in your interface language instead of the raw stored value.
* Free-text search accepts parentheses as plain text; it used to reject the query. The "not equal" filter operator stays selected, and editing a filter by hand no longer corrupts the field name.
* Selecting a sub-organisation in quick search inserts its name, not its uuid, and the sub-org autocomplete no longer lists duplicates.
* The dashboard can now fetch up to 10,000 documents per search window, so large result sets page correctly.
* The duplicate-document panel shows the same resolved columns as the main list, and multi-word supplier filter values survive pressing Enter.

### Tasks

* The assignment e-mail goes out when a task is assigned, once. Editing a task or marking it done no longer re-sends it, and the "assigned on" date stays the date of assignment. Task e-mails also render properly in Outlook.

### Workflow Builder

* Search, sort order and pagination in the workflow list stay consistent while you filter.
* The "run workflow on change" toggle in the layout builder now gates the run, and enabling it requires picking a workflow.

### Settings and administration

* The WatchDog download link and setup command point to the environment you are in, not always production.
* Decision Trees: the selected document field stays highlighted when the picker reopens, truncated labels get a tooltip, and user names (not raw ids) are shown when adding a line.
* The System Admin checkbox is editable when editing a user.
* Analytics: Core Web Vitals render from the real measurement data, and the logs service view works.
* "Use Default Template" in the layout manager copies the default layout as intended.
* Custom field labels no longer override the bundled translations of standard fields.
* Supplier portal quotes: submitting a quote with a REF1 value outside the allowed list is blocked.
* MediOrder gets duplicate document detection on its validation screen.

## API Service — `12.82.3`

### Security and organisation isolation

* Switching the acting organisation is validated against your actual membership and fails closed, and an internal test endpoint that could be abused to cross organisations was shut.
* Document scripts can no longer be read or overwritten across organisations, neither via the apply-to-document call nor via a foreign version id on save.
* Sub-organisation user lists and group member lists only return people from the caller's organisation, and adding several users to a group at once no longer drops all but the first.
* A credential from the wrong organisation is refused before it can become a document's processing token, and full-text search queries run as the calling user rather than a service identity.
* Four-eyes approval is enforced: the second approver must differ from the person who approved first.
* The live PO Dashboard list is scoped to the user's sub-organisations.

### Document pipeline

* Documents denied for export no longer sit in "Exporting" forever, and export errors always carry a message instead of an empty one.
* When a processing step crashes, the document goes to an error state instead of being stuck in "restart in progress" with no way out.
* A barcode split that fails or times out marks the document as Error instead of silently showing "Running", and a split that produces no children keeps the parent and flags it instead of deleting everything.
* A failed retry can no longer overwrite a document that meanwhile finished processing.
* Documents restarted without user interaction and split children now run under a durable organisation token, so long-running processing doesn't die with an expired session.
* An empty layout-template response is no longer cached for six hours, which used to make layouts vanish until the cache expired.

### Extraction and e-documents

* Amounts written with a trailing minus ("100,00-") are parsed as negative instead of being dropped.
* Swiss documents are detected as Swiss (CHF, CHE VAT numbers, CH IBANs) instead of defaulting to German conventions, and dates written with typographic dashes parse correctly.
* XRechnung 3.0, 3.0.1 and 3.0.2 credit notes in CII syntax are classified as credit notes with the total read from the grand total field; the same goes for pure CII credit notes. A declared ZUGFeRD 2.4 / Factur-X 1.08 version wins over the generic profile identifier, and bare XRechnung types resolve to their UBL or CII twin instead of failing.
* Dropdown (list-of-values) fields such as Tax Country and Tax Code keep their value through field transformation; they were being emptied.
* Table extraction: a failure in a numbers-only column stays in that column instead of killing the whole table, AI table extraction gets a timeout that survives multi-batch runs, and two crashes on unusual table shapes (rows without page positions, ragged column counts) are fixed.
* Source rule patterns match case-insensitively.

### Export

* A tax check that fails during export preview returns a readable error instead of a server error, on both preview endpoints.
* SFTP export can send the original document alongside the converted one.
* When export configurations exist on several levels, the most specific one wins consistently.
* BOD exports can carry column type attributes via mapping.

### Import and master data

* The e-mail Import Log is complete: rejected and failed inbound e-mails always get a log row with an accurate reason. No more silent drops.
* Purchase order BOD imports keep sub-lines attached to the right line; a carried-over flag used to attach them to the wrong one.
* Importing a CSV with several new suppliers works (their generated ids no longer collide), cash discount term aliases import and honour the "on conflict" setting, and the on-conflict IGNORE choice applies beyond suppliers.
* The supplier suggestion (TF-IDF) keeps its supplier id when a preference is updated, so suggestions stop pointing at nothing.

### Other fixes

* Dashboard rows resolve dropdown labels in the user's language, without blocking the request.
* After editing fields, the PO match status updates instead of showing the pre-edit state.
* Purchase Order Change documents get five Purchase-Order-parity fields and a default field validation layout.
* Error responses across 152 endpoints return readable messages instead of raw exception objects, and the logs analytics page no longer answers with 502 for organisations without a log index.

## Auth Service — `1.77.9`

* Password-reset e-mails were silently never sending; fixed, together with the thread-safety issue underneath it.
* A replayed refresh token is rejected: the authoritative database check runs every time now instead of being skipped on a cache hit.
* Two-factor authentication: an authenticator app can be enrolled alongside e-mail codes, and removing the last passkey or regenerating backup codes requires a fresh second factor first.
* A valid sub-organisation id is no longer rejected with "Organization not found", and an API key created in a sub-organisation resolves its technical user from that sub-organisation.
* Editing an organisation validates the partner id and no longer resets the organisation type as a side effect.
* "Remaining tokens" in the subscription view is anchored to the contract year, not the calendar year.

## Auth Bridge Service — `0.5.7`

* Account replication between the EU and US regions recovers on its own. A dropped replication stream reattaches in place, replication keeps flowing while a reconciliation runs, and reconciliation memory is bounded so the service stops crash-looping on large tables.

## Barcode Service — `1.18.7`

* Barcode reading runs under a time limit and reports a timeout instead of hanging, which used to leave the document stuck in processing.

## Docflow Service — `2.9.8`

* Field values written by a workflow card land on the document in both stored representations, so a later export no longer reverts them.
* A retried trigger keeps the work the run already did, contended triggers on the same document queue up instead of stealing the lock, and an escalated retry is ranked first in the queue.
* Purchase order comparison cards: tolerances compare as exact decimals and are based on the purchase order value, reversed comparison directions are available as options, a group assignee is reported as a group instead of failing a user-id comparison, assignment ids compare correctly as UUIDs, lines with empty numeric values are skipped, and a "received" comparison without any received data reports missing data instead of pretending to match.
* The Apply Decision Table card has been retired.

## Email Service — `1.41.0`

* Gmail imports pick up each attachment exactly once; duplicates from overlapping fetches are gone.
* The import read-cursor only advances after an import is confirmed, so a crash mid-import can no longer skip e-mails.
* When an import configuration is deactivated because a similar one exists, that deactivation is visible and notified instead of silent.

## Extraction Service — `1.54.5`

* Whether a document is a credit note or an invoice is resolved by which keyword sits closer to the document type mention, instead of first match wins.
* When several tax interpretations are within tolerance, the exact reconciliation is preferred over a near miss.
* After a forced re-OCR the document type and locale are restored, so table extraction and training work on re-OCR'd documents again.
* Documents without a document type no longer crash table rule lookup.

## FTP Service — `1.32.8`

* Folder scanning does one listing round trip per folder with a bounded depth, so imports from large FTP directories are much faster and stop timing out.

## Fulltext Service — `1.42.3`

* Documents whose stored search payload had no extracted fields are re-indexed from the database, so they show up in dashboard search again.
* The dashboard search window supports up to 10,000 documents.
* Facet searches no longer fail when semantic search is active.

## OCR Service — `1.10.7`

* The OCR time budget is sized by real per-page cost, so long documents finish instead of hitting the pipeline limit.

## PO Match Service — `1.59.8`

* Table lines with zero quantity are skipped in mismatch checks instead of producing false mismatches.
* When required PO-match columns are missing, the result names them.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.docbits.com/overview-and-basics/release-notes/incremental-updates-12-25-august-2026.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.