Known Limitations

Real limitations of the product today — useful to know before you go looking for a workaround that doesn't exist.

Recording & sessions

• Screen recording requires the Linc Chrome extension. Without it, upload a pre-recorded video instead.
• "Just Questions" is only available when accepting an invitation (and only if your administrator allows it). A self-started session can't use it, and you can't switch a session to Just Questions after accepting another way.
• Recordings are capped at 2GB. A larger video is refused at the file picker — trim it, re-export it at a lower resolution or bitrate, or split the process into shorter recordings (join the later parts with "Part of an existing process").
• You cannot replace or re-upload a video once a workflow has started. Retries reuse the recording you already uploaded: a failed or stalled analysis offers Analyze again / Try again, and failed document generation offers Try again — and when the SOP was already written before the diagram step failed, the banner also offers View SOP, so the finished SOP is usable without re-running. The one step with no retry button is removing sensitive information ("We couldn't process this video safely") — for that you start over.
• Starting over: an invited workflow can be Reset from its page; a self-created workflow cannot — delete it from the Scout list and create a new one.
• A meeting recorded by the meeting bot only gets screenshots while a screen was shared. If nobody shared their screen, the SOP has no screenshots. A small webcam thumbnail that the meeting app shows next to a shared screen can still appear in a screenshot — remove that screenshot in the SOP editor if it shouldn't be there.
• You cannot change a workflow's name or description after creating it (you can move it to a different department from its "⋮" menu).
• A second screen is only recorded if you checked "Record two screens" before starting, and you can't record more than two screens.
• You can't type in the session chat while a video or documents are processing.

Shadow mode (recording your whole day)

Shadow mode is a pilot, and these limits apply to it:

• A recording day is captured by the browser extension only. A recording-day invitation offers no Upload, Meeting, or Just Questions tab, and a day you start yourself (New Session → Record my working day) likewise has no document upload and can't be marked as covering multiple processes.
• A day you start yourself has no schedule. It is one session for one calendar day, the day you created it. Once that day is past it can't be started or resumed — its page offers Record today, which creates a new session for today. It can't be linked to a library process, and nobody is emailed if your recorder goes quiet (that email goes to the person who sent an invitation, and there isn't one).
• A recording day can't be deleted while it is recording or uploading. The delete is refused until 15 minutes after recording stops.
• Quitting Chrome, or an extension update, ends capture for the day. Everything recorded up to a few seconds before it happened is kept and uploads automatically the next time you open Linc; the time after that is marked as a gap in the day's record, and Resume my day carries on. On extensions before 1.2.3 the same is true of closing the Linc tab (its start screen says "Keep this browser tab open for the day"). From 1.2.3 (its start screen says "You can close this tab") the recording runs inside the extension, so the Linc tab can be closed freely.
• A single piece is capped at 512 MB, and one stretch of recording at 2000 pieces. A new stretch starts each time you click Resume today. Each screen you share saves its own pieces, and they all count toward the 2000. In normal use a piece is about 50–150 MB and a full day is well under 200 pieces per screen, so these limits are only reached if something has gone wrong; a piece over either limit stays on your device and is not uploaded.
• One day per session. Once you click End day and Save, that day is over: it cannot be resumed or recorded again in the same session. A recording-day invitation creates one session per scheduled day: the first when you accept, every later one when you click Record today.
• Record today only works on a scheduled day. The person inviting you picks the days (first and last day, timezone — every day in between counts; an older invitation may also be limited to some weekdays); on any other day the session page names your next recording day instead, and a session for a day other than today does not offer Record my day. Only the person who accepted the invitation can create its recording days.
• The schedule can't be changed after it is sent. Re-sending the invitation from the invite dialog replaces its days while it is still pending; once accepted, the days are fixed.
• Two screens at most, each recorded separately. Tick Record two screens before Record my day and your browser asks for the second screen right after the first. Each screen is saved as its own video, so the day's footage is roughly twice the size (disk and upload).
• A screen can't be added mid-day. Resume today asks for as many screens as the day was started with. On a day started with Record two screens, if one screen stopped (or you skipped the second), stop sharing and click Resume today to pick both screens again. A day started with one screen stays one screen.
• On a Mac, the browser's two "Stop sharing" buttons can't be told apart. This is how the browser works, not a Linc setting. While you share two screens, the browser shows a "…is sharing your screen" bar with a Stop sharing button for each screen, but it puts both bars at the bottom of your main display, one on top of the other. Your second display shows no bar, and nothing tells you which bar belongs to which screen. Clicking Stop sharing may stop both screens: each click stops one, and the other screen's bar then appears in the same place. Linc has no button to stop one screen on its own. To keep your screens out of the recording for a while, use Privacy pause (it pauses both screens). To record only one screen, end the day and leave Record two screens unticked on your next recording day.
• Video only — no microphone. Nothing is narrated, and there is no mute button because there is no audio.
• No live questions. The interviewer doesn't ask anything during the day and no interview opens when you end it. Nothing is analysed straight away; the footage is reviewed later on a schedule, and the interview about your day comes after that.
• Sensitive information is not automatically blurred in this pilot. Use Privacy pause before opening anything personal or confidential.
• About 12 GB of free disk space is recommended for a full day. Footage waits on your machine until it has uploaded; if less than that is free when you start, a warning shows but recording still starts.
• You cannot discard or reset a recording day. Ending the day offers Save only, and a recording-day workflow can't be Reset from its page the way other invited workflows can — an invited recording day can only be removed by the person who invited you (a day you started yourself you can delete from your Scout list, like any other session of yours).

Importing existing processes

• Accepted formats are BPMN 2.0 XML, PDF, PNG/JPG/WebP, Word (.docx), PowerPoint (.pptx), Excel (.xlsx/.xls) and CSV, up to 30 MB each — 5 MB for images (PNG, JPG, WebP), so a large photo of a whiteboard may need shrinking first. Older .doc and .ppt files aren't readable — save them as .docx / .pptx first.
• A PowerPoint deck is read from its shapes, not its pixels. Boxes, arrows, groups, SmartArt, tables and speaker notes are all read — and because the connectors are read as the flow, a drawn slide has its map transcribed as drawn, the same as a PDF diagram. A flowchart that is a pasted picture of a diagram isn't read at all: export that slide as PNG or PDF and import it that way instead. Decks over 200 slides are refused — anywhere you upload a deck, not just here.
• Splitting a file into several processes is page-granular, and the survey is a reading. When one file holds several processes (see Processes → When one file holds several processes), each unit is read on its own pages — but two units drawn on the same page both get that page, and a chapter that starts mid-page carries the tail of the one before, so a step can occasionally land in the wrong unit; check the ladders of neighbouring units. The table of contents itself is Linc's reading of the document: it can propose a chapter that only summarises a procedure documented elsewhere, or miss that two chapters are variants of one process — that is what the confirm screen is for (untick, rename, regroup, or Read as one process). A single image or one-page PDF is split only when Linc is very sure; boxes that merely name other processes on one diagram are never split off. A file with more than about sixteen units is flagged as a likely process library.
• Variant groups created at import don't show in Architect's Variants comparison straight away. The grouping is saved with the processes, but that comparison lists processes from an Architect analysis, so the group only appears there once an analysis has put those processes on the map. It is the same group in both places — confirming or dismissing it in the Library or on the map answers it everywhere.
• A diagram is kept as drawn; a written procedure is not. A BPMN upload — and a PDF, image or slide whose flow Linc could follow — keeps the map as drawn: the arrows, the gateways, the start and end events and the notes on the page are transcribed, and generating the SOP doesn't redraw them. A written procedure (a Word document, or a page with no arrows to follow) gives you the step outline Linc could read, and its map is drawn from that outline. Transcribing is still a reading of a picture: an arrow Linc can't make out is filled in from the order of the steps, and a low-resolution or hand-drawn diagram may come out with missing or merged steps. Filling an arrow in lowers the read confidence reported for the file, so an import that says low read confidence — check the steps is worth opening: compare the Coverage ladder and the process map against the original. The original picture of the map is kept either way and shown in the SOP as an "As documented in …" figure, but it is a picture — it can't be edited in the BPMN editor.
• Pictures carried in from imports aren't checked for personal data. Screenshots taken from an imported document go into the SOP as they are; if a screenshot shows names, e-mail addresses or account numbers, crop or blur it in the source document before importing (or remove it from the SOP in the editor). At most 60 pictures are kept per file.
• Imported steps stay at "outlined" depth until someone records the process with Scout — an import doesn't count as evidence of how the work is actually done. Architect still analyses imported processes, but marks them and any gaps raised against them Documented, not observed: the findings describe what the documentation says, not what anyone was seen doing.

Encrypted Office files

• Files encrypted by a sensitivity label can't be read. An Excel, Word or PowerPoint file whose Microsoft 365 sensitivity label encrypts it (for example "Confidential") is refused when you add it. Change it to a label that doesn't encrypt, if your organization allows it, and add it again.
• Only Office 2010-or-later password encryption can be unlocked in Linc. For those files Linc asks for the password and keeps an unlocked copy. Older password-protected files — .xls workbooks and files saved by Office 2007 — must have the password removed in Office first.
• The check runs in the upload dialogs only. Files brought in by a SharePoint import, or attached inside an email, aren't checked; if one is encrypted it shows as failed to process.

Activity attributes

• Attributes are filled in only where a source states them. An import reads Responsible, System, Input, Output, Evidence and Manual/Automatic from whatever the document shows — a swimlane, a data store, a modelled task type, a named column — and leaves the rest blank rather than guessing. Blanks are filled in on the Activities grid.
• Recorded and interviewed steps don't populate the grid yet. A step a Scout session discovered arrives with no attributes, so fill it in on the grid.
• Editing the grid doesn't rewrite a deliverable that already exists. A newly generated SOP carries the Activities table and a newly generated diagram honours the lanes, task types, systems, data and evidence — but an SOP or diagram generated earlier doesn't change until you regenerate it.
• Diagram attributes land only on steps the diagram names. A step is matched to a diagram task by name; if the generator merged, split or renamed a step so that no task matches, that step's attributes stay in the SOP table but aren't drawn. The Activities table itself is exact regardless.
• The Activities table appears only in deliverables generated on a process page (from an imported outline or a merge with new interviews). A Scout session's own SOP follows the template's Activities section as written prose, since a session's steps don't carry attributes yet.
• The attribute set is fixed. You can choose which of the six are required (Settings → People, under Step requirements) but not add, rename or remove attributes.
• The .xlsx export round-trips steps, not gateways. A decision in the ladder exports as a row named "Decision: …" and comes back as an ordinary step of that name.

The in-interview process map

• The draft map shown during an interview can't be edited directly — no dragging, renaming, or deleting boxes. Corrections are made by telling the interviewer in the chat.
• It doesn't appear at the very start of an interview — it needs a few answers (or a processed recording) first.

Language

• Generated documents cannot be produced in another language. SOPs, diagrams, analyses, and roadmaps are always written in English, even when the interview or sources were in another language. System, product, and team names are never translated. When a draft comes back with words in the interview's language (most often Portuguese or Spanish), Linc rewrites it once automatically; a stray label or phrase can occasionally remain — Regenerate to get a fresh draft.

Downloads & exports

• SOPs: PDF, Word (.docx), or Markdown (.md). From a Scout interview's SOP preview, the PDF reproduces the branded preview, so it's only offered once a branded template is applied. On a library process's page, PDF and Word are always available — with a plain layout when your organization has no designed template. Markdown never contains the screenshots or the process flows — each becomes a line naming what it showed.
• Cropping or redacting a screenshot doesn't reach earlier versions. A crop or redaction in the SOP editor replaces the image in the version you're editing, so the cropped-away or blacked-out part is gone from that version and from everything downloaded from it — but every earlier saved version of the SOP still holds the original image. A crop or redaction can be adjusted only until you close the editor; after that the saved image is final.
• Process maps only appear when the version has a BPMN diagram. A template's process map slot and the flow fragments after steps are drawn from the process's own diagram; a version without one shows the written introduction alone. A very wide map is drawn as several numbered parts rather than one shrunken picture.
• BPMN diagrams: PDF or XML — no PNG or SVG image export. The PDF is the map as one sheet to zoom, or on A4 landscape pages to print (as in the SOP PDF), not an editable file; the diagram itself stays fully editable in the BPMN editor.
• Flowcharts and roadmap diagrams: PNG, SVG, or Mermaid source (.mmd) — no PDF.
• Chat conversations can't be exported or downloaded — see Chat below.
• Documents don't update themselves when a teammate answers after generation — the workflow is marked Needs update; the recording's documents are not regenerated, so take what the answer changes into the process's latest version from the process page.
• A session's as-recorded SOP, BPMN and flowchart are read-only, for the owner too — they are the record the process was built from. Edit the process on its page; only its latest version can be edited (older ones read Superseded by vN).

Search (⌘K)

• It doesn't search inside documents, transcripts, or messages — it matches names and titles only. Ask Chat about contents. The Library's own filter (press /) is separate: it narrows the open folder, including the text of written documents there.
• Individual files aren't listed — search finds the Library folder a file lives in.
• A person's page shows no processes they own (Linc doesn't record process ownership), and a department's page doesn't list its interviews.

Intended vs Actual

• It grades one department at a time, from that department's page. There's no organization-wide or Initiative-wide grade, and it isn't in Architect. A process with no department set is never graded.
• Only documented processes are graded — a process needs a published SOP or at least one interview behind it.
• Older .doc, .ppt, .txt, and .rtf priors are refused at upload, because their text can't be extracted. Save as .docx, .pptx or PDF first.
• A run grades at most 40 processes; the summary says how many were left ungraded.

Industry benchmark

• Only processes that resemble a known domain are compared — Internal Audit, Source-to-Pay, Order-to-Cash, Record-to-Report, Hire-to-Retire, FP&A, plus any process you imported as a reference model. A facilities or one-off workflow with no matching reference is skipped.
• A process is matched by its name, department and description — not by reading its steps. A process whose name doesn't resemble the reference is skipped even when the work itself is a standard flow.
• The comparison is not observed evidence. Aligned does not mean anyone was watched doing that step.
• There is no business-case presentation export of the benchmark yet (Architect Excel/CSV of Opportunities includes Origin and Reference columns).

Interview planner

• Inline questions can only carry open questions. A documentation gap always needs an Interview. If a gap is assigned to someone set to Inline questions, the plan won't send at all until you switch that person to Interview or move the gap to someone else.
• The planner can't tell someone not to record. It asks in one of two ways — Interview or Inline questions. On an interview, whether the person records a walkthrough, uploads one, or just talks it through is their choice once they accept.

Architect questions

• Only admins can forward questions; regular members see the Questions tab read-only.
• A forwarded answer doesn't change the current results in place — it's kept as context for the Initiative and picked up the next time you run the analysis. (Answers to questions asked outside an Initiative are recorded but not fed into the next run.)
• One open forward per question at a time — while a forward is still awaiting an answer, forwarding that question again is refused with "This question has already been forwarded and is awaiting an answer". Links expire after 7 days, and re-running the analysis doesn't cancel a forward that's still waiting.
• You can't challenge or edit a gap while the Overview is Preliminary — run a full analysis first.

Architect opportunities

• Opportunity dollars are work exposed to change, not measured savings. Linc has not compared before and after. The figure is the hours on that row × your Initiative Assumptions.
• The Opportunities headline does not add every row. Overlapping work (inbox triage that already includes confirmations and tracking) would be counted twice. Use the per-row hours and source tags.
• A re-run is needed for existing analyses to pick up measured recording durations and source tags. Older runs keep their original hours until you re-analyze.

Access

• To-Be surfaces (Opportunities, Roadmap, Designer, Analysis chat) are not on by default for members — they need To-Be access or an admin role. Guests can't open Architect or Designer at all.
• Sharing sessions works within your organization only, guests can't be added as collaborators, and people you share with can't delete the session, clear its conversation, move it between departments, or forward questions.
• "Ask the organizer" only exists on interviews someone else set up, goes only to the person who invited you, and is a single question-and-answer, not a thread.
• Each invitation creates one workflow — except a recording-day (shadow mode) invitation, which creates one recording day per scheduled day — and invitees see only their own workflows.
• Initiative admins manage only their own Initiatives — creating Initiatives and granting Initiative admin are org-admin actions.
• Only organization admins and Initiative admins/owners can start an Architect analysis. Other members can open Architect and read results but will not see Run Analysis, Run First Analysis, or Re-run architect.
• Reorganizing library processes (delete, group as variants, merge, split a version out) is for the process's creator, its Initiative's admins and org admins. Other members don't see those actions.
• Deleting a process that is on an Architect process map does not remove it from the map — the next analysis run lists the process in the Library again. Merging keeps only the merged process's current document, not its earlier versions.

Chat

• Chat cannot generate an SOP or a BPMN diagram, and it cannot start an Architect analysis. Use Scout for those deliverables and Architect to run an analysis. Chat answers questions about the documents you select and about analysis that already exists.
• Chat conversations can't be exported or downloaded, and you can't edit the AI's responses (editing your own message redoes the conversation from that point).
• Chat only knows a process's published SOP. An interview's own SOP is not searchable, and neither is a draft version (one added by an automatic match, or a version nobody has published) or an interview whose "looks like an existing process" suggestion is still unanswered. Publish the version, or answer the suggestion, and chat picks it up.

Company context the AI reads

• The background block has a size limit. The company context handed to Chat, Scout and the deliverable writers — organization description, departments in scope and their people, the Initiative's written documents — is capped at about 48,000 characters. Over that, the oldest written documents and short text documents are left out first, one at a time, until it fits; the organization and department descriptions, people, and the Learned from interviews document are never dropped. Nothing is deleted — the documents stay in the Library — and Settings → What Linc knows says how many were left out. Architect analysis runs read the full, uncapped block.
• Chat on another organization's Initiative reads no company context. When you work in an Initiative that another organization owns (you were added as an external member), Chat answers from the documents you select only; it does not read your own organization's description, nor the other organization's.
• What Linc knows is org-admin only, and only for your own organization's Initiatives.

Analysis chat attachments

• Attached files can't be opened by clicking their chip (no preview); limits are 20MB per file, 5 files per message, and only images, PDFs, spreadsheets, Word, PowerPoint, and text/Markdown/JSON files are accepted.

Miner log uploads

• On Miner, Upload logs stores spreadsheets and other tabular extracts (Excel, CSV, JSON, XES, and plain text) for the organization — up to 50MB each and 40 files. These are not a fixed set of filenames.
• Analysis chats copy those files into the chat workspace and can compute over them with Python (pandas). A file deleted on Miner is dropped from the chat the next time a message is sent. Chat attachments are separate and still use the 20MB / 5-file limit above.