Skip to main content
Every Clarity process has two layers. The outputs are what Clarity generates: the Current Process, the Automation Proposal, the process summaries, and the Process Landscape. The captures are the sources those outputs were built from: voice interviews, screen recordings, documents, and images, each with its own transcript. An AI assistant working through Duvo can read both. That means you are not limited to the summary Clarity wrote. You can ask an assistant to go through every interview on a process, compare what different people said, find disagreements the map smoothed over, or pull exact quotes for a business case. This page covers when to use which layer and how to set up an analysis that reads the captures.

​
Outputs vs. Captures

The outputs are a synthesis. They merge every capture into one consistent process, which is exactly what you want for a map — and exactly what hides the differences between people. When the question is “who said that?”, “do the two warehouses actually do this the same way?”, or “what did people complain about that never made it into a step?”, go to the captures.

​
Where You Can Work with Captures

All of these surfaces respect your access. An assistant sees the processes and captures you can open in Clarity, and nothing more.
  • Clarity Chat — the Ask Duvo box on a process. It reads the generated process and can list and read that process’s captures. Best for questions about one process. See Edit and Ask with Clarity Chat.
  • Ask Duvo across the organization — organization-wide answers use process summaries. Open a process, or name it in your request, when you need its captures or interview evidence.
  • The Duvo MCP server — connect Claude Desktop, Cursor, ChatGPT, or another MCP host and ask in plain language. See Connect to the Duvo MCP server.
  • The Duvo CLI — the best choice for large analyses, because transcripts can be saved to files and processed in bulk instead of loaded into one conversation. See Clarity CLI.

​
Start from the Outputs, Drill into Captures

The most reliable analyses start from the generated process and use captures as evidence, rather than starting from raw transcripts.

Get the shape of the process

Read the overview and the live Current Process first. In the CLI: duvo clarity overview <process-id> and duvo clarity current <process-id>.

Find the claim you want to check

Pick the steps, gaps, or numbers that matter for your question. duvo clarity gaps <process-id> groups missing information, open questions, and assumptions by proposed step.

Trace it back to a capture

duvo clarity evidence <process-id> prints a citation ID for each generated step. Resolve one with --citation <citation-id> to see the step it supports, the attribution (who or what the evidence came from, such as “Anna (sales interview)”), and the quoted excerpt. A citation does not carry a capture ID, so match the attribution against the list from duvo clarity captures <process-id> to find the capture to open.

Read the capture

Read the full transcript of that capture to see what was actually said, in context.

​
Analyze Every Capture in a Process

To go through all the interviews on one process, the assistant first lists the captures, then reads each transcript. With MCP (the tools an assistant calls for you):
  1. getClarityProcess with captures: "lite" — returns the process and its captures without transcript text, so the list stays small. Each capture carries its status, who recorded it, and whether a transcript is available.
  2. getClarityCapture with the process_id and capture_id — returns one capture with its full transcript and video transcript. Call it once per capture.
With the CLI, write the transcripts to a file and let the assistant work from it:
Without --include-transcripts, capture output contains metadata only. Add the flag whenever you want the text.
Load transcripts one capture at a time rather than all at once. A process with many long interviews can fill an assistant’s context quickly, and a focused read per capture produces better notes than one huge input.

​
Analyze Across Many Processes

For a portfolio question — “what do people across all our finance processes say about manual re-keying?” — combine the two layers:
  1. Use listClarityProcessSummaries (or duvo clarity process-summaries) to scan summaries, SWOT, and projected impact across processes and pick the ones that matter. Both are paginated: the CLI returns 5 processes per page (up to 10 with --limit), so advance --offset until you have reached the total in the response. MCP callers page the same way with limit and offset.
  2. For each selected process, read the captures as described above.
  3. Ask the assistant to keep notes per capture first and synthesize at the end.
For dozens of processes, prefer the CLI. Run duvo clarity list --json to get the process IDs (50 per page, up to 100 with --limit; advance --offset until you have them all), save each process’s captures to its own file with duvo clarity captures <process-id> --json --include-transcripts, and point your assistant at the folder. In Claude Code, Cursor, or a similar tool the assistant can then read, search, and compare files without holding every transcript in one conversation. Process Landscape interviews live at the organization level rather than on a process. Read them with duvo clarity interviews org list --org <org-id> and duvo clarity interviews org get <interview-id> --org <org-id>, or list landscape captures with duvo clarity landscape captures --org <org-id> --include-transcripts --json.

​
Write Prompts That Produce Trustworthy Analysis

The assistant can only be as precise as the request. A few habits make the difference:
  • Name the scope. “Every completed capture on the Invoice approval process” is answerable. “Our interviews” is not.
  • Ask for citations. Require the capture (or interviewee) behind every finding, and a direct quote for anything important. That makes each claim checkable in seconds.
  • Separate what was said from what was inferred. Ask the assistant to label conclusions drawn from the process data as its own inference, not as something a person said.
  • Mind capture status. Ask for completed captures — one that is still recording or processing may not have a finished transcript yet. Excluded captures stay readable, but Duvo leaves them out of analysis and evidence selection, so call them out separately if you include them.
  • Synthesize last. “Summarize each interview in five bullets, then list where the interviews disagree” beats “summarize the interviews”.
Example requests:
  • “Go through every completed capture on process <process-id>. For each one, list the steps the person described, the systems they use, and the exceptions they mentioned. Then show where their description differs from the live Current Process.”
  • “Read all interviews on the three returns processes. List every pain point that was mentioned more than once, with the interviewee and a quote for each.”
  • “For each step in the Automation Proposal marked low readiness, find what the interviewees said about that step and explain what is still unclear.”

Clarity

Capture methods, generation, Clarity Chat, and the Process Landscape.

Clarity CLI

Every duvo clarity command, including captures, evidence, and exports.

Connect to the Duvo MCP server

Use Clarity tools from Claude Desktop, Cursor, or ChatGPT.

Available MCP Tools

The full list of Clarity tools and what each returns.