A governed research record is a firm's single, versioned system of record for its investment views. Any covered analyst or automation can propose a change to a view, only the accountable lead analyst can accept one, and every step is logged. It exists so that investment teams can put AI to work in research without losing control of what the firm actually believes.
- A governed research record is a system of record for conclusions, not another writing tool or document store.
- Proposing and deciding are separated roles: analysts and automation propose, the lead analyst accepts.
- Every accepted change creates a new version; prior versions and the log of conclusions are never overwritten.
- Governance is enforced in the database, not in AI prompts, so it holds for every client that connects.
- The model matters now because AI made research output cheap; the scarce thing is a view the firm can stand behind.
Why do investment firms need one?
Because the bottleneck in research has moved. Producing drafts, summaries, and screens is now cheap: an agent can read a quarter's filings overnight and write a competent brief before the desk sits down. What has not become cheaper is accountability. When a portfolio manager asks what the firm thinks of a company, the answer cannot be "whatever the model said this morning." It has to be a view a named analyst stands behind.
Most firms hold that view in scattered places: decks, notes, chat threads, and the heads of senior analysts. That was tolerable when humans produced everything, because production was slow enough to police informally. Once automation writes too, the informal system breaks. A governed research record is the formal replacement: one place where the current view lives, with rules about how it changes.
How does a governed research record work?
Three mechanisms do the work.
One living view per company. The record holds a single current view for each covered name: thesis, ratings, assumptions, risks. It is versioned like source code. An accepted change produces version eight; version seven is retained, not overwritten.
Propose and accept are separate powers. Anyone on coverage can propose a change, and so can automation: an overnight intel job that spots a guidance cut can draft the update and attach its sources. But a proposal waits in a queue until the lead analyst accepts it. Acceptance is atomic: the new version is committed, the research log is appended, and the audit row is written in one step.
The rules live in the database. Role separation is enforced with row-level security in the database itself, not in application code or model prompts. A contributor who tries to accept their own proposal is rejected by the database. That matters because the record has many clients: a web console, AI assistants, scheduled agents. Rules enforced in prompts hold only for the client that carries the prompt; rules enforced in the database hold for all of them.
How is it different from a research management system?
A research management system (RMS) stores research: notes, documents, models, tags. It answers "where is our work on this company?" A governed research record governs the conclusion itself. It answers "what do we believe about this company, who decided that, and what did we believe before?" The two are complements, not substitutes, but the record is the part an RMS does not provide: versioned conclusions with role-separated change control.
AI copilots sit on the other side of the same gap. They generate research but hold no record: the output lands in a chat window and evaporates. A governed record is what makes their output usable at firm level, because it gives the generated work somewhere accountable to land.
Where does AI fit in a governed record?
Inside it, with a hard boundary. In Alphasyn's implementation of the model, AI connects over the Model Context Protocol and can do real work: query the firm's collective view, read any company's record, and draft evidence-grounded proposals. What it can never do is conclude. Ratings and conviction stay the analyst's call, and acceptance stays with the lead. The short version of the whole model: automation proposes, the analyst decides.
Frequently asked questions
Is a governed research record the same as a knowledge base?
No. A knowledge base stores documents; a governed research record governs conclusions. It holds one versioned, accountable view per company, with rules about who can change it, while documents and notes serve as the evidence behind that view.
Who can change a governed research record?
Anyone on a company's coverage, including automation, can propose a change with evidence attached. Only the accountable lead analyst can accept a proposal into the live view. Readers can query the record but cannot propose or accept.
What happens to old versions of the view?
Every accepted change creates a new version and every prior version is retained. Nothing is overwritten or deleted, so the firm can always answer what it believed at any point in time and why.
Does a governed research record replace analysts' writing tools?
No. Analysts keep writing wherever they write. The record holds and governs the firm's collective conclusions, and connects to the tools around it rather than replacing them.
Can compliance audit a governed research record?
Yes. Every mutation writes a row to an append-only audit log, and the research log of conclusions is append-only as well. The audit trail is a property of the database, not a report generated after the fact.