Why Weavidence Lab separates computation, interpretation, and review

#Answer in brief

Lab keeps data-quality validation, computation, and interpretation as distinct, inspectable stages — rather than one pipeline that quietly turns raw data into a finished chart — because a reproducible result requires knowing not just what the number is, but which step it survived and which step it might not have.

#The problem: a chart cannot show you what it hid

An analysis that goes straight from raw data to a finished visualization answers "what does the data show" while quietly refusing to answer "what happened to the data before it showed that." If a station identifier was missing for four rows, if a remediation step silently dropped or imputed values, if a quality threshold was crossed and waived — none of that is visible in the chart itself. It is only visible if the pipeline was built to keep it visible.

#What Lab keeps separate, on purpose

Lab's three core capabilities are data quality and validation; exploratory and statistical analysis; and reports and visualizations that can be registered as artifacts rather than treated as disposable output.^[1] Keeping these as distinct stages — not folding quality checking invisibly into the computation step — is what makes each one auditable on its own terms.

Three consequences follow directly from that separation:

• Lineage survives the whole pipeline. A reported number can be traced back through remediation and normalization to the raw ingestion it came from, rather than starting its traceable history only at the point someone thought to save an intermediate file.
• Quality issues surface before analysis, not after. A validation finding — a missing identifier, a threshold breach — is something the analyst sees and decides about, not something that silently shapes a result no one gets to review.
• Outputs are reviewable, not ephemeral. A report or chart can be registered as an artifact with its lineage attached, rather than existing only as a screenshot someone happened to keep.

#Jana's bounded role in this pipeline

Lab includes Jana as an assistant that surfaces suggestions — for example, recommending a source manifest be reviewed before a remediation step is applied. Jana suggests; she does not act. The distinction matters specifically because remediation changes data: an automated system quietly correcting values without a visible suggestion-and-decision step would break the same lineage guarantee the rest of the pipeline is built to protect.^[1]

#Practical implications

For a team evaluating whether an analytical result is trustworthy, the useful question is not "does the chart look reasonable." It is: can you walk backward from the chart to the raw data, through every step that touched it, and see which quality issues were found and how they were handled? If the pipeline cannot answer that, reproducibility is a claim, not a property.

#Limitations and uncertainty

Lab is in active development. Its data-quality, lineage, and artifact- registry capabilities described here are real and implemented, but Lab is not yet publicly accessible, and it is not represented as production-ready or externally validated.

#Connection to Weavidence

Lab's lineage record is the analysis-stage counterpart to Journey's Decision Trace and Academy's source-versioned curriculum — three different mechanisms built on the same underlying commitment. See Why traceability matters in AI-supported evidence systems.