A research system that shows its working
A live research workspace keeps every conclusion connected to its method, sources, model settings, evidence and caveats.
Project outcome
Every conclusion can be replayed from the evidence, method, model settings and caveats behind it.
Investment
Not published
Timeline
Ongoing
Status
Live frontier research
Reach
The investment research team using the workspace
The value of a research system is not that it sounds autonomous. It is that the judgement it supports remains attributable.
This project created a live research workspace for a large institutional investor. A lead analyst role locks the question and method, bounded specialist roles complete defined tasks, and the execution record preserves the evidence, model settings, provenance and caveats behind the result.
Project brief
Institutional research combines qualitative judgement with quantitative comparison. The team needed to evaluate companies, factors and periods without changing the method halfway through or losing the source behind a conclusion.
A final report was not enough. Another person needed to understand what was asked, which plan was used, what sources were allowed, how the analysis ran and what limitations applied.
What we delivered
- A research workspace with a lead analyst control point.
- Bounded specialist research roles inside a locked plan.
- Evaluation across a company-by-factor-by-period matrix.
- A managed lifecycle for creating, revising and validating factors.
- Deterministic peer comparison and backtesting aggregation.
- A reporting layer that uses completed source-backed results only.
- A persistent execution record containing the objective, plans, methods, sources, model configuration, evidence, provenance and caveats.

What changed
The team can inspect the question, method, sources, model settings, result and caveats behind a conclusion after the research run is complete. A later reviewer does not have to rely on the original analyst's memory or search through private working files.
The same method can be applied across the comparison matrix. Peer comparison and backtesting follow explicit rules instead of asking a model to combine persuasive paragraphs. The result is more consistent and easier to challenge on the terms the team defined.
Reports are prepared from completed results. The reporting layer cannot introduce a new finding while writing the document, so the final report remains a clear presentation of the research rather than a second place where unsupported reasoning can appear.

Engineering detail
The execution record is the central control. It connects the objective to the plan, the plan to the method, the method to the evidence and the evidence to the result. Plan revisions remain visible. Source constraints, exact model configuration and caveats remain part of the record instead of disappearing after publication.
The software roles are bounded by design. The lead analyst role owns the research question and interpretation. Specialist roles complete assigned work. The reporting role presents completed results. Professional judgement remains visible throughout the process.

The project in one sentence
A live research workspace made consequential analysis reviewable by preserving the method, evidence, provenance and caveats behind every conclusion.