Working with Claude-generated Artifacts
This guide provides an important update regarding the dashboard artifact generated by Claude, summarizing your analysis in Petavue-Claude MCP. While the core analysis of your data is robust and accurate, we've identified some inconsistencies with the dashboard-style (.html) artifacts.
Our commitment is to provide you with 100% accurate and reliable outputs. In this document, we will explain the identified issues, provide an interim solution for generating fully accurate reports, and outline our plan for a permanent fix.
Identified Issues with Dashboard (.html) Artifacts
We traced problems to Claude occasionally making assumptions when rendering dashboards. The core analysis remains correct, but secondary assumptions can affect the visual layer:
1. Claude adds extra metrics with flawed assumptions
The primary issue is that Claude sometimes adds a new metric to the dashboard that was not part of the original core analysis. It invents this metric by creating its own formula using other correct, pre-existing data points.
Example: The dashboard shows five widgets. Four of them (Total Deals, Win Rate, etc.) display the correct numbers because they came directly from the Petavue analysis.
However, the "Total Revenue" widget shows an incorrect figure of $8.8M. This metric was not in the original analysis. Claude invented it by making an assumption and multiplying Total Deals by the Average Deal Size (1,188×$7,380). This calculation is incorrect because accurate revenue figures must be based on won deals only, not the total number of deals.
2. Inconsistent Chart & Graph Data
Some charts are being populated with estimated data points that were not part of the original, verified analysis.
- Hallucinated Data Buckets: Charts like the "Deal Size Range" graph may show data grouped into buckets (e.g., $0-25K, $25K-50K). The number of deals within these buckets is an estimation by Claude, not an actual count from your data done by Petavue.
- Incomplete Trend Data: In rare instances, trend analysis graphs have been generated using only a small sample of the data (e.g., the first 20 records), failing to provide a complete and accurate view.
3. Visual and Layout Errors
Some dashboards have experienced presentation glitches, including console errors (Uncaught Error: Script error) which can break the page layout and result in an unresponsive, continuous scroll.
Our Recommendation: Use "Document-Style" Reports for 100% Accuracy
To ensure you are working with completely reliable data, we strongly recommend using the document-style artifact feature for now.
These reports are text-based summaries, similar to a Google Doc, that can be downloaded as a PDF. Most importantly, the numbers, stats, and findings in these documents are pulled directly from the core analysis done by Petavue. These documents do not involve the secondary (Claude-induced) AI assumptions that can affect dashboards, guaranteeing their 100% accuracy and reliability.
With these reports you will have better control and confidence. In fact, you may also follow up on adding a new metric or make changes to existing metric to add to the report. If you request a new metric, the system will perform the analysis again and seamlessly update the report with the new data.
How to Generate “Document-Style” Report:
It's very straightforward. After the analysis is complete and the initial findings are displayed, simply ask Claude to create a "report" or a "document". The system will then generate a comprehensive and accurate summary for you. As shown below, on left in the main conversation we are asking the system to generate a document based on above analysis, on the right we can see the report document generated.
How to generate a document-style report:
After your analysis finishes, prompt Claude with: “Create a document-style report based on this analysis.” You will then receive a complete summary you can download and share with full confidence.
Our Path Forward: A Flawless Dashboard Experience
We are already working on a permanent solution. We are building a new dashboard workflow inside the Petavue platform. This will give us direct generation and control over visualizations and remove the need for Claude’s interpretive assumptions. The goal is to deliver:
- Every metric in the dashboard matches the verified analysis.
- Interactive visuals that are 100% reliable without layout errors.
- A trustworthy experience you can use in executive reviews and board reporting.
Thank you for your understanding. Your feedback is crucial, please try the above and reach out to us in case you have any questions or face any issues.