Reports come from completed evidence, not fresh model opinions

A polished research report can still be weak if the writing step introduces a conclusion that the analysis never produced.

The investment team needed a reporting process that respected the boundary between evidence and explanation. Completed research should be turned into a clear report. The reporting layer should not become a second model opinion that quietly adds a new finding because it makes the story sound better.

We delivered a reporting agent with a deliberately narrow role. It prepares the report from completed, source-backed results. It cannot introduce a new finding during the reporting step. That makes the final document easier to review because the report remains connected to the research that already passed through the team's method and evidence controls.

Client background

Our client is a large institutional investor with an investment research team that needs to move from complex analysis to decision-ready communication. The team combines model-supported research, source material and expert judgement. Its reports must be useful without losing the caveats that limit the conclusion.

The research workspace is live frontier work. The reporting capability is one part of a wider system that preserves the question, method, evidence, provenance and model configuration behind a result.

Business challenge

Report writing is often treated as presentation, but it can change the perceived meaning of research. When a writer or model fills a gap, strengthens a sentence or adds a new interpretation, the final document may no longer represent the completed analysis.

That makes review harder. A decision owner may ask:

• Is this sentence supported by a completed result?
• Which source is behind the claim?
• Was this caveat present in the analysis or lost in editing?
• Did the reporting process create a new finding?
• Can the report be traced back to the execution that produced it?

The team needed a reporting layer that improved clarity while preserving the boundary around what had actually been found.

Implementation

We separated analytical execution from reporting. Research roles complete bounded tasks inside a locked plan. Their results remain attached to evidence, provenance, methodology and caveats in the persistent execution record.

The reporting agent reads those completed results and prepares the report. Its job is to organise, explain and present what exists. It does not conduct a fresh research pass or add a conclusion that was not part of the completed result set.

This separation gives the team a practical review point. The analytical result can be checked before the document is written. The document can then be checked for clarity and faithful representation of that result. If a claim is not in the completed evidence, it does not belong in the report simply because it would make the narrative stronger.

The report remains connected to its execution context. The team can inspect the original objective, plan, source constraints, model configuration, evidence, provenance, methodology and caveats behind the material being presented.

Why it was difficult

Good reporting requires interpretation, but interpretation can cross the line into new analysis. The system had to support a readable report without pretending that presentation is neutral by default.

The control also had to be practical. A report that includes every internal trace would not help a decision owner. The solution needed to preserve the detailed evidence trail behind a concise output, so a reviewer could move from the conclusion to the working when needed.

Value delivered

• Reports are built from completed, source-backed results.
• The reporting step cannot introduce a new finding.
• Evidence, provenance and caveats stay connected to the material being reported.
• The team can review the analytical result before the report is prepared.
• Decision owners receive a clear document with a replayable research trail behind it.
• The boundary between analysis and presentation becomes explicit and governable.

The value is trust in the handoff. The report can be concise without becoming untraceable, and the team does not have to choose between usable communication and defensible evidence.

What this means for your business

If an analyst, model or reporting tool can introduce a new conclusion at the last step, your review process is still open after the evidence work is done. Separate the completed result from the writing layer and make the reporting role narrow enough to govern.

That applies to compliance summaries, underwriting packs, policy briefings and board reporting. The reader should be able to understand the decision and, when necessary, follow it back to the evidence that earned its place in the document.

Related: Applied research · Data and analytics · Specialty lenders