Filings agent
Reads 10-K, 10-Q, 8-K and announcements. Flags changes in guidance, risk-factor language, segment disclosure and accounting treatment.
Product
The reading layer above your existing sources.
Portfolio Intelligence reads what your team would read if it had the hours. It reports what changed, argues why that change matters to the thesis you wrote, and shows the passage it came from.
Agents
Four readers, one ranked view.
Source types don't behave alike. A price move and a footnote in a 10-K need different reading. Each agent is specialised for its own material, and their observations are reconciled before anything reaches you.
Earnings agent
Works through transcripts and prepared remarks. Tracks shifts in management framing between quarters and unanswered analyst questions.
Market-activity agent
Watches price and volume relative to sector and index, so a move is read in context rather than in isolation.
News agent
Reads licensed news for developments involving the company, its customers, suppliers and competitors. De-duplicates coverage of the same underlying event.
Materiality
A score you can argue with.
Each event is scored 0–100. The score isn't a black box number — it decomposes into the factors that produced it, and every factor points back at evidence.
Thesis contact
Does the event touch a stated assumption, and in which direction?
Source weight
A filed disclosure carries more than a secondary report of the same thing.
Novelty
Is this new information, or a restatement of something already priced and already read?
Position context
Weight in the portfolio, and whether the same driver affects other holdings.
Thesis contact
Direct — gross margin assumption, negative direction. +38
Source weight
Filed 10-Q, management discussion. +24
Novelty
Not present in prior quarter's guidance. +19
Position context
Top-five weight; shared driver with two other holdings. +8
Weights are configurable per portfolio and shown on every score.
Thesis tracking
Assumptions, not adjectives.
A thesis written as prose can't be monitored. A thesis written as a short list of assumptions can. You state what has to remain true for the position to work; the system reports on each one as new information arrives.
- Five to eight assumptions per position is the working range
- Status moves between intact, improving and weakening — with the event that moved it
- Status changes are timestamped, so the thesis has a history
- You can review and revise assumptions at any time; the record is kept
MSFTThesis · last reviewed 04 Aug
Enterprise AI attach rates drive durable seat expansion.
↑ Copilot seat growth Improving
✓ Azure capacity Intact
⚠ Capex intensity Weakening
✓ Enterprise pricing power Intact
Illustrative. Assumptions are written by you, not generated for you.
Boundaries
What it deliberately does not do.
Being clear about this is part of the product. The system is built for research workflows, and stops where research stops.
No buy or sell calls
It reports what changed and why that touches your thesis. The investment decision is yours.
No trading, no execution
There is no order path. Research pilots require no brokerage credentials of any kind.
No return forecasts
Materiality scores rank attention, not expected performance. They are not price targets.