Punchcard Excel imports updates

Bug Fixes

• CoAudit's activity feed now shows consistent, readable steps when helper agents are at work: CoAudit sometimes delegates parts of a task to specialized helper agents. Their activity now appears in the same clean format as the main agent's steps — showing actual file names, collapsing repeated actions into a single grouped count, and displaying a compact summary once the work is done. Tapping the summary expands a scrollable view of every step. Previously, this activity could show garbled internal codes instead of file names, repeat the same step dozens of times and flood the sidebar, or disappear from the chat entirely after a page reload.

• Cited workpaper cell references in Excel exports no longer fail on large workpapers: When CoAudit builds an Excel export that cites specific workpaper cells, it can now anchor those citations directly from the workpaper data it already scanned — without going back to re-read source documents a second time. This resolves a failure that previously blocked cited Excel exports when working across a large workpaper.

New in CoAudit

• Clickable citations in AI-produced Excel workbooks: When CoAudit builds an Excel workbook for you, individual cells can now include citations that link directly to the source document pages — the same click-to-open experience you get from a workpaper exported in Punchcard. Ask CoAudit to build you a workbook and it will wire up the citations automatically.
• CoAudit can now delegate complex tasks to worker agents: For requests that require reading many files and authoring detailed evidence across a large workpaper, CoAudit can now hand off that exploration to a focused worker — keeping your conversation responsive while the heavy lifting runs. This capability was previously available only to administrators; it is now part of every task-mode CoAudit chat.
• Filter workpaper rows by review state: CoAudit can now scope its reads and actions to only the rows you've flagged, validated, or haven't reviewed yet. Ask it to "look at only the flagged rows" or "re-run only unreviewed items" and it will target exactly that subset.
• Automation re-runs preserve your reviewed work: CoAudit now correctly understands that re-running a workpaper automation never overwrites cells you've already validated or flagged — so you can safely ask it to refresh the rest of the workpaper without worrying about losing your review work.
• Upgraded AI model: CoAudit chat now runs on an upgraded model with noticeably better accuracy on complex routing and multi-step tasks.

Chat & UI

• Drag folders and ZIP files into the chat: You can now drop a folder (or several folders at once), or a ZIP archive, directly into the CoAudit chat composer — all files inside are collected and uploaded as individual attachments. Previously only individual files could be dragged in.
• Scroll position stays put when the AI finishes responding: When the AI completes a reply, your reading position is no longer snapped to the bottom of the conversation. If you scrolled up to read an earlier message, you stay there. The scroll-to-bottom button becomes highlighted with a bouncing arrow to signal that a new reply is waiting — clicking it or scrolling down clears the indicator.
• Redesigned AI loading indicator: The indicator shown while the AI is generating a response is now a slim, solid bar pinned at the bottom of the conversation with an animated Punchcard mark and status text. It no longer lets transcript content bleed through behind it.

Requests

• Simplified request status labels: Client-facing request statuses are now Outstanding, Returned, Submitted, and Accepted. The button clients use to submit work is now labeled Submit Request; when a request is already submitted, it shows Return to Outstanding instead. Both auditors and clients will see a one-time announcement on the board summarizing the label changes side-by-side.

Bug Fixes

• Testing attribute name saves reliably on the memo page: Typing a name and then quickly reloading the page or closing the tab no longer silently discards the change — it now saves the moment you click away from the field.
• ZIP file previews work correctly in CoAudit chat: Files extracted from an uploaded ZIP now preview correctly in the document pane instead of showing a fetch error.
• PDF black-page recovery improved: Black or blank PDF pages are now detected and recovered from the moment a document opens, not only when navigating between pages.
• Collective workpaper columns compute correctly across multiple files: A column configured to produce one shared result across all files in a datasource group now generates a single synchronized value and applies it to every file, rather than running independently per file.
• Excel workbook delivery is more reliable: In rare cases, generating an Excel artifact could fail due to an internal step unrelated to the workbook's content. That step now fails gracefully so your workbook is still delivered.
• Task breadcrumb segments no longer collapse: Breadcrumb text in the task header now maintains a visible minimum width instead of occasionally collapsing to nothing.

Features

• Import Deep Analysis questions into Skills: Each existing Deep Analysis question now has an "Import to Skill" button that opens the Skills editor pre-filled with the question's content and name. The option to add new Deep Analysis questions has been replaced with a link to the Skills library — existing questions continue to run, edit, and delete as before.

Reliability Fixes

• Files still processing after upload no longer appear as failed: When a large file takes longer than expected to finish text recognition, the workpaper now shows a "still processing" notice rather than marking it as failed. The status updates automatically once processing completes, so there is no need to re-upload.

• Failed saves now revert to the last confirmed value: Across status changes, memo fields, review states, task names, and descriptions, a save that doesn't go through now reliably restores the field to the last value the server confirmed, instead of leaving it stuck in the unsaved state.

• Switching tasks no longer mixes up edits between tasks: A save started on one task will no longer affect the task you navigate to afterward. Each save stays in the context of the task where it was made.

• Review status changes now update all related items immediately: Accepting or clearing a review state on one workpaper item in a linked group now immediately reflects on the other items in that group in the same view, without requiring a refresh.

• Narrative card review badges stay in sync: Review status badges on narrative cards now consistently reflect the latest confirmed state, even when a save and a concurrent change arrive at the same time.

Performance

• Large Excel uploads no longer freeze the browser tab. Uploading large spreadsheets previously caused the tab to lock up for a minute or more and sometimes left a file that appeared in the list but could never be downloaded. Uploads now complete in under a second regardless of file size.

AI & Sandbox

• Co-Audit Agent sandbox steps now show what they're doing. When the agent runs code to build a report or analyze data, each step is labeled in plain English — for example, "Building Excel workbook" — instead of a static spinner. The underlying code is still available behind a "View code" toggle for those who want it.
• The Stop button now immediately halts sandbox work. Pressing Stop while the agent is executing code interrupts all further steps right away rather than letting them run to completion.

Workpaper

• Circular-reference detection covers more scenarios. Punchcard now catches circular dependencies that arise when pasting data with a different column mapping, when a match relationship is auto-linked, or when a match mode is switched. When a loop is found, the warning banner displays the exact chain as clickable column pills so you can jump straight to the involved column.
• Batch run approvals now name each column. When the Co-Audit Agent queues several column-level runs for approval at once, each entry in the approval card now shows the specific column name and its document group. Previously all entries showed only the document group, making a batch of runs on the same group indistinguishable.
• Editing a deleted column now shows a clear message. If you try to update a column that a teammate has already deleted, Punchcard now explains that the column no longer exists instead of showing a generic error.

Reliability

• Extraction progress no longer flickers. The Review & Corrections panel in the chat sidebar stays steady while an extraction is running instead of flashing repeatedly.
• Financial Statement Review consistency checks are more stable. A rare scenario where two consistency rules targeted the same entry at the same time could cause the analysis to stop partway through. This is now handled gracefully.
• Logging in from multiple locations no longer produces an error. Opening Punchcard in a second tab or browser at the same time as an existing session previously triggered a server error in some cases; it now resolves cleanly.

CoAudit Chat

• Grouped action approvals: When the AI requests several actions at once — such as moving multiple files into document groups — all pending approvals are now consolidated into a single grouped card. A "Preparing N approvals…" indicator appears while the AI is still working, then a single Approve all / Reject button handles the entire batch. After approval, the card shows a result summary inline (e.g., "Moved 5 of 6 files — receipt_03.pdf couldn't be moved").
• Reject with reason: Declining an approval now reveals an optional text field where you can explain why. The AI receives your note and can adjust its next steps accordingly.
• Clearer choice prompts: When the AI asks you to choose from a list of options, the card now uses distinct, clearly labeled Submit and Cancel buttons — making it easier to respond without ambiguity.

Financial Statement Review

• Sync scroll preserves your alignment: When reviewing current-year and prior-year PDFs side by side, enabling Sync Scroll now locks in whatever alignment you've already set manually, rather than snapping both panes back to identical page positions. Scroll together from wherever you left off.
• Prior-year tick mark editing: Clicking an existing tick mark annotation while in Prior Year Consistency mode now opens the full context menu — giving you the same edit and delete controls available in other review modes.
• Shift-constrained proof marks: Hold Shift while drawing a line, arrow, or double-line to snap it to horizontal, vertical, or 45° diagonal. Useful for drawing precise accounting double-underlines and tie-out arrows.

Workpaper

• Document group file uploads fixed: Creating a new document group from a client request and then uploading files into it now works reliably. Previously this could fail silently, leaving the group unable to accept files.
• Datasource rename conflict explained: Renaming a datasource to a name already used in the same task now surfaces a clear message explaining the conflict. Previously the rename would silently revert with no feedback.

Reliability

• PDF viewer recovery: The PDF viewer no longer gets stuck with unresponsive zoom controls and dead page navigation when first opened inside a panel that was collapsed or hidden. The viewer now recovers automatically when the panel is expanded.
• Excel viewer stability: Opening the Excel viewer on a file whose AI citations reference a sheet that no longer exists no longer crashes the viewer.
• Task chat stability: The CoAudit chat panel no longer goes blank when navigating between tasks in the same session.
• Team and client invite error messages: When adding a team member or client user to an engagement fails, the notification now explains the specific reason (for example, "identity provider temporarily unavailable — try again shortly") instead of showing a generic error.

New Features

• Private engagements: Mark any engagement as private from the engagement settings page options menu (⋯). A private engagement — and all its tasks — is hidden from firm colleagues who aren't explicitly named as team members. Only named members and admins can see it. The toggle updates immediately and reverts automatically if saving fails.

• Excel viewer renders cell colors and comments: The Workpaper Excel viewer now faithfully recreates background fill colors and font colors from the original file, so color-coded headers, highlighted rows, and styled tables appear exactly as they do in Excel. Cells with comments or notes show a small red triangle in the corner; hovering reveals the comment text.

• TEUR-aware math validation: The Financial Statement Review calculator now detects when a document is presented in thousands or millions (TEUR, MEUR, "in thousands") and automatically adjusts. Operand values display as integers matching the statement convention, a quiet unit label confirms the active scale, and the verdict reads "≈ Validates on rounding" when a footing passes within the tolerance band but isn't an exact tie. Column totals that appear before their operand columns — common in some European reporting layouts — are now also validated correctly.

• Plain text and RTF files can now be uploaded: .txt and .rtf files are now accepted for upload and AI extraction, joining PDFs, Excel files, and CSVs.

• Real-time identifier availability when creating an engagement: The short identifier field now checks availability as you type and shows an inline hint if the chosen identifier is already taken — for example, "ACME" is already in use. This engagement will be created as ACME2. — so you know the final identifier before submitting.

• CoAudit can link workpaper columns as engagement-wide variables: When working with the CoAudit agent on a Workpaper task, you can ask it to mark a column's extracted output as an engagement-wide variable, making those values available for reference across all tasks in the engagement.

Bug Fixes

• Excel citations now highlight the correct cell: Citation highlights in the Excel viewer now land on the right cell even when the spreadsheet has leading blank rows or columns before the data area.

• Citations on repeated values highlight one cell, not a span: When a cited value (such as "$0") appears in multiple cells of the same table row, the citation now highlights a single cell rather than drawing an incorrect band across unrelated cells.

• Page no longer freezes after closing a confirmation dialog: Clicking Cancel on a confirmation dialog — such as a file delete confirmation — could leave the entire page unresponsive to clicks. This is fixed.

• Workpaper automation correctly populates records for shared datasources: Running automation on a task that shares a datasource no longer produces an empty grid. Each task now generates its own records correctly.

• Financial Statement Review handles partial document processing more gracefully: If one section of a document encounters a problem during processing, the successfully processed sections are now preserved so the review can continue rather than failing entirely.

• CSV exports with title rows now import correctly: CSV files that begin with a report title line or blank line before the column headers — common in GL and payroll exports — are now recognized and imported correctly.

• Office lock files are rejected at upload: Temporary Office files (those beginning with ~$) are now immediately rejected with a clear unsupported-file error rather than being silently submitted for processing.

Performance

• Excel viewer opens large files faster: Cell color scanning is now limited to the visible portion of the sheet on open, so large workbooks no longer cause a noticeable delay when first viewed.

AI & CoAudit

• Files uploaded in CoAudit task chat are now auto-sorted into the correct document group as soon as they finish processing. A persistent trail in the conversation records what happened — "Uploaded invoice.pdf", "Sorted 3 files into Invoices · Bank Statements" — so you always know where your files landed, even after leaving and returning to the chat.
• Ask CoAudit to move a misrouted file: if a file ends up in the wrong group, tell CoAudit ("move invoice.pdf to the Invoices group") and it will move it without leaving the chat.
• Excel cell background colors and cell comments are now captured during file processing. Cells with color-coded classifications or embedded auditor notes in Excel workbooks are now visible to AI extraction columns.
• Excel files with formulas now show correct values instead of blanks. Formulas that previously computed to empty — common in programmatically produced spreadsheets — are now evaluated correctly.
• Chat messages with file attachments now reliably receive AI responses. A bug where certain messages appeared to send successfully but never produced a reply has been resolved.

Workpaper & Automations

• Testing attribute automations now pick up every file, including those added after the first run. Previously, files uploaded to a shared document group after you first ran an automation were silently skipped. All files are now included on every run.
• "Fill empty" reruns now correctly process all rows, including rows that were added since the previous run.
• Automations and Match Agent runs stay isolated to the correct workpaper when a document group is shared across multiple workpapers. Results no longer bleed between workpapers sharing the same document group.

Engagements & Clients

• Engagement identifiers no longer block creation or editing when already in use. The identifier field now adjusts automatically (e.g. SAND → SAND2) instead of returning an error that requires you to try again.
• Document groups created from PBC request imports are now available in the document group picker across other workpapers in the same engagement, so evidence collected for one workpaper can be reused without re-uploading.
• Creating new client organizations is faster. The New Client form now responds immediately after clicking Create instead of freezing on firms with many existing clients.

New Features

• Copy data directly from another request: A new "Copy from Request" option in Import Table Data lets you populate a workpaper table by copying data from another task in the same engagement. Pick the source task, choose a sample datasource or a tabular file, then configure the header row and columns using the same picker as a direct Excel import. Both source groups ("Sample Selections" and "Table Files") are always shown with a clear indication when a group is empty.

• Pick which sheet to import from multi-sheet Excel files: When importing an Excel file that has more than one worksheet, a sheet selector now appears in the column picker so you can choose exactly which sheet to use. Previously, imports always used the first sheet with no option to change it.

Performance

• Copy from Request source picker opens faster for large files: Row counts, column counts, and column headers in the source picker now load immediately instead of re-reading the file each time the picker opens. Files with tens of thousands of rows that previously caused a multi-second delay now appear instantly.

Bug Fixes

• All worksheets are now selectable when importing older Excel files: In some Excel files uploaded before a recent update, worksheets beyond the first appeared grayed out in the import sheet picker. Those sheets are now always selectable and always show the correct preview data for that sheet.

• Citation pills scroll to the right location on the first click: Clicking a citation pill in cell reasoning now navigates the document viewer to the correct passage immediately. Some citations previously required two clicks before landing in the right spot.

• Viewer header and toolbar stay visible after clicking a citation: Clicking an inline citation pill in cell reasoning no longer causes the document filename and toolbar buttons to disappear. The header stays intact while the citation is highlighted in the document.

• Toolbar buttons stay fully visible in a narrow viewer: When the document viewer is at a constrained width, toolbar action buttons (lasso, search, and others) now remain fully visible — the filename truncates before any buttons are hidden.

• "Informed by" tree: clicking a value node navigates the workpaper fully: Clicking a value link in the "Informed by" provenance tree now moves the grid selection, cell-detail panel, and document viewer to the referenced cell together. Previously the document viewer could fail to update or show the wrong content.

• "Informed by" tree: citation pills show one highlight and scroll vertically: Clicking a citation pill inside the "Informed by" tree now shows a single highlight on the document (no duplicate outline) and scrolls straight to the cited passage without any sideways movement.

• Wide preview tables scroll horizontally in Copy from Request: The data preview on the Copy from Request confirmation step now displays a horizontal scrollbar for tables with many columns, so all column data is visible without being squeezed.

Features

Workpaper

• See what informed each AI result: When Punchcard AI fills a workpaper cell using values from other columns, an "Informed by" section now appears in the cell detail panel. Expand it to see a tree of every upstream value the AI referenced, the reasoning behind each connection, and inline citations you can click to jump straight to that location in the source document. Click any node in the tree to navigate directly to that cell in the workpaper grid.
• Excel and CSV file import now available to all users: Uploading an Excel or CSV file when adding selection data is no longer a limited rollout — every user now has access. This is especially useful for very large populations that are impractical to paste from the clipboard.

Bug Fixes

Chat

• Send button no longer stays disabled after attaching a file: After uploading a file in a chat conversation, the Send button could remain unclickable even after the file finished processing. This is now fixed.
• Chat messages containing links now display correctly: AI responses that included certain link formats could cause the entire message to fail to load. This is now fixed.

Workpaper

• OCR status indicator always visible on the Files tab: The badge showing files with pending or failed text recognition was being partially hidden behind the adjacent tab. It is now correctly positioned and fully visible at all times.
• Spurious error on workpaper open no longer appears: A "Failed to load available headers" message that sometimes flashed briefly when first opening a workpaper no longer appears.

Notifications

• Activity summary emails are no longer sent twice: Some users were receiving their activity summary email twice in quick succession. This has been resolved.

Performance & Reliability

Large Data Paste Handling

• Pasting huge Excel tables no longer freezes the application. Previously, importing journals with hundreds of thousands of rows would hang the browser for many minutes. The new optimized parser and incremental processing keep the workpaper responsive throughout the entire import.
• Massive improvements to memory usage for large data imports. The system now samples data intelligently rather than processing every cell upfront, preventing the browser from running out of memory on very large pastes.
• Smart confirmation dialog for very large pastes. Pastes exceeding 200,000 rows now show a recommendation to use the file upload feature instead, which handles that scale more efficiently without blocking the UI.

Data Integrity

• Fixed cell splitting when pasting Excel data with line breaks. Previously, cells containing quoted text with embedded newlines would incorrectly be split into multiple rows during import. This now works correctly.
• Single-column table pastes no longer split valid wrapped cells. Previously, wrapped cells in single-column tables would be broken into extra rows during import.

Authentication

• Users can now log in with the updated organization authentication domain. The organization's recent authentication system update is fully supported, so all accounts can access the platform without interruption.

Shared Document Groups

• Smarter import flow when client requests are involved: Switching a document group to a different shared source now handles every combination of client-upload ownership with the right confirmation experience. When both the current and target groups are owned as client requests by different tasks, a dedicated dialog appears — showing which task currently handles client uploads, what files will be affected, and what happens to client upload routing — before you confirm.
• See exactly which files will be permanently deleted before switching: When switching to a different shared document group would permanently remove files, the panel now shows the exact file count upfront. The confirm button stays disabled while this check runs, so you can never accidentally confirm before the full impact is visible.
• "Import document group" toggle: The toggle in the document group editor panel is now labeled "Import document group" when the group is standalone, and "Shared document group" once sharing is active, making its current state clearer at a glance.

EU and German Document Accuracy

• Fixed: EU financial values ending in zero cents were being dropped: In EU-format tables — such as German balance sheets — values whose cents ended in zero (for example, 15.375,50 or 16.991,00) were being silently excluded during extraction. This affected roughly 30% of numbers in affected tables, causing footings to fail and numbers to appear missing. These values now extract correctly.
• Fixed: prior-year comparisons could mis-route for round amounts: A related issue caused the year-over-year comparison logic to silently treat all values as equal for certain round figures, which could misdirect reconciliations across periods. This is now resolved.
• German section references and legal citations no longer treated as financial numbers: German document markers — such as "Anlage 3", "Kapitel 5", "§ 322 HGB", and numbered headings — are now correctly excluded from financial number extraction, reducing noise in AI analysis on German audit reports.
• German fixed asset schedules no longer incorrectly combined: The system was sometimes merging separate sections of a fixed asset schedule (Anlagespiegel) that document different financial events — for example, acquisition costs on one page and accumulated depreciation on another — into a single table, dropping data in the process. These sections now remain separate.
• Fewer false inconsistency flags on comparative-period columns: The inconsistency checker was occasionally raising exception marks on comparative-period columns based solely on shared row position, even when the values are expected to differ across periods. This has been corrected.
• Improved extraction for EU headerless and multi-block schedules: Table extraction quality improved for EU-format financial statements that span multiple sections without standard column headers.

Processing Reliability

• PDFs with complex graphics now process reliably: PDFs containing many vector-graphic elements — such as diagrams or heavily formatted pages — were causing OCR processing to fail entirely for that file. These files now process normally, with all text and numbers extracted as expected.

Improvements

• Applying a template is now instant. The intermediate document group mapping screen has been removed from the template apply flow. Selecting a template and clicking Apply creates the task immediately — no extra confirmation step required.

• Email invites accept any paste format. When adding team members or clients to an engagement, you can now paste addresses in any common format: space-separated, comma-separated, one per line, or directly from Outlook ("First Last <email@firm.com>"). Invalid addresses are flagged inline before the invite is sent, so you can fix them on the spot instead of hitting a generic error after the fact.

• Foot/Crossfoot calculator lists operands in document order. Values in the F/CF calculator are now shown in the order they appear in the PDF — page by page, top to bottom — making it straightforward to trace each number back to its location in the source document.

Bug Fixes

• "Apply Valid" / "Apply Exception" no longer fails after a tick mark is removed. Re-applying a footing result to a number that previously had its tick mark removed now succeeds as expected, instead of showing a "Failed to finalize calculation" error.

• AI chat panel no longer goes blank during streaming. An issue that caused the chat panel to crash and stop rendering while a response was being generated has been fixed.

• Workpaper page is stable on long-running engagements. An issue causing the workpaper page to crash on engagements with a long activity history has been resolved.

New Features

• AI-produced files in task chat. The Co-Audit agent can now build custom Excel spreadsheets, Word documents, and PDFs directly in the chat — not just describe what to do, but hand you the finished file. Ask for things like "build a loan schedule with a separate sheet per vendor" or "fill this form from our workpaper data" and a colored download chip appears at the end of the response. The agent can also read evidence and reference files already attached to the task when building its output.

• Move and Copy matched evidence. Right-clicking a matched row in the workpaper now offers a Move or Copy option under "Move to":

  • Move — removes the original match and creates it under the chosen parent row, with a preview of all downstream records that will be affected before you confirm.
  • Copy — adds a second association without touching the original, useful when the same evidence belongs under multiple parents.

• Run on empty match cells. Right-clicking an empty match column cell on a parent row now shows a Run option alongside "Pick document," so you can ask the agent to fill the match automatically rather than picking a file manually.

Improvements

• Template reconciliation is faster and clearer. When applying a template whose document groups share names with groups already in your engagement, the reconciliation screen now:

  • Lets you type a custom name when creating a new group (formula references update automatically).
  • Shows a "Client request · ID" chip next to any reuse option that's already shared with clients, and warns you that client uploads will continue going to the original task — so you know the upload ownership consequence before confirming.
  • Skips the screen entirely when there are no naming conflicts, jumping straight to a one-click Apply summary.
  • Highlights inline when two rows would resolve to the same name.

• First-run modal is simpler. When you open the Run dialog on a scope that has never been run, you now see just the scope description and a single Run button. The breakdown table and duplicate Re-run button only appear after something in the scope has been run.

• Remove Match is consistent everywhere. The consequence preview — showing how many downstream matched rows will be removed — now appears regardless of whether you use the grid right-click menu, the cell-detail panel, or the document-detail panel.

• Document group picker shows more context. The "Use Existing Document Group" panel now displays a file count badge on each group row, and task labels include the short task ID for faster scanning.

• AI artifact sessions start faster. The Co-Audit agent's file-building sessions now reach a ready state noticeably faster than before.

• Better AI model for EU users. The default AI model for auditors in the EU region has been upgraded to match the model used in the US.

Bug Fixes

• Co-Audit sidebar no longer splits in two. Opening "Fix with Co-Audit" from the Problems tab previously rendered a duplicate strip of recent chats below the chat panel. The chat panel now takes the full sidebar height as expected.

• Download chip appears once on saved chats. Reopening a chat that contained an AI-produced file used to show the download chip twice. It now appears exactly once.

• Template column setting for "run on no match" is now preserved. A per-column setting that controls whether the agent runs when no evidence match is found was being silently dropped when saving or applying templates. It now round-trips correctly.

Workpaper

• Bulk accept/reject now covers all row types: Validate, Flag, and Clear operations in the range selection panel now apply to consolidated cells (cross-datasource inline rows) and top-level non-group rows — these were previously silently excluded. The accepted/rejected/pending counts in the range panel also update in real-time as review states change from any surface (right-click menu, live sync), without requiring you to re-select the range.
• Deep analysis spinner appears immediately: Starting a deep analysis run — whether a single question or a bulk run — now shows the progress spinner right away in the toast notification, instead of briefly displaying an empty circle before the run is registered server-side.
• Crash fixes in the workpaper editor: Two stability issues were resolved: one that could crash the workpaper when collapsing the chat panel via ESC in certain edge cases, and one triggered by mention transforms in the text editor.

Chat

• PDF-to-Excel exports work in follow-up messages: You can now ask the AI to export a PDF to Excel in a follow-up message, even after the original attachment has scrolled out of the latest turn. Previously the export tool was only available in the same message where the file was attached, making common workflows like "now export that to Excel" fail unexpectedly.
• Reorganized task chat sidebar header: Navigation controls (back to board, chat history) and action controls (new chat, expand/collapse, more options) are now split into distinct groups in the task chat header, improving discoverability and reducing visual clutter.

Formulas

• More reliable collective formula reductions: Formulas that aggregate multiple AI-retrieved values now handle mixed types correctly — numeric comparisons, boolean logic, and currency symbols (including $ prefixes) are preserved across collective reductions in AI retrieval formula paths.

Improvements

• PDF to Excel on large documents: The extraction pipeline that cleans and fills converted tables now processes pages in batches. Large PDFs that previously failed mid-conversion due to size limits will complete successfully. Pages that are exceptionally large are escalated to a higher-capacity model; pages that still exceed limits are skipped gracefully rather than aborting the entire conversion.

Bug Fixes

• OCR retry reliability: If an OCR job had to retry after partially completing, subsequent attempts could fail with a conflict error — leaving the file stuck in a failed state. Retries now cleanly replace any partial results, so files recover correctly without manual intervention.

Copilot Chat

• Citation continuity across conversation turns: when the AI cites document evidence with numbered references like [1] or [2], those source links are now carried forward into follow-up exchanges — the AI can correctly refer back to previously cited content in multi-turn conversations without losing context
• Copy and download controls on AI tables: markdown tables in AI responses now include a toolbar to copy or download tabular data directly, eliminating the need for manual selection
• Built-in product coaching: ask the copilot where to find any platform feature and it will describe what it does and guide you to the right page — the full platform layout is now part of its context in every chat mode

Engagement Board

• Filters persist across engagement switches: search queries and active filter selections now carry over when switching between engagements in the same browser session; assignee and category filters are automatically reconciled against the new engagement's valid values, with unrecognized values quietly dropped

Review

• Calculate-First mode for foot/crossfoot validation: draw all operands on the PDF first, then apply the result to a cell — an alternative to the standard flow; footer action buttons (Save, Reset, Apply Valid/Exception) stay pinned and accessible while scrolling the calculator panel
• Assignment focus mode: when the system cannot automatically link a drawn operand to its row and column in the table, the calculator collapses and a focused card guides you step-by-step to draw the row and column headers in the PDF to complete the link; the calculator re-expands once the link is resolved

Task Overview

• Task properties update live from AI chat: status, priority, due date, assignee, and category now reflect changes made via AI chat in real time — no page refresh needed
• Description stays current without refresh: the task description editor automatically remounts when AI chat updates the content externally, so edits appear immediately

Workpaper

• OCR is resilient to server interruptions: OCR processing now runs as a durable background job — if the server restarts mid-upload, the job is automatically picked up and completed on another node; duplicate OCR requests for the same file within 5 minutes are deduplicated; files no longer get stuck in a failed state due to transient server issues
• Formula cells export as live Excel formulas: cells whose value starts with = are written as real Excel formulas in all .xlsx exports — values calculate automatically when the file opens; number formatting (currency, date, number) is applied based on the column's data type
• Testing attribute columns are visually distinct: columns used for testing procedures now display with a blue header background and bold centered label, always appearing as the rightmost columns within their datasource group for quick identification
• Automation output reliability improved: AI-generated text in match and testing attribute automations is sanitized before saving, preventing rare write failures caused by edge-case characters in LLM outputs