How an Investing Agent Can Catch a Thesis Change Before Earnings
September 11, 2026
Altsets
Research by Altsets Research
Store the dependency assumptions behind a stock thesis, then compare new relationship data and public evidence with that original state so an agent can flag changes that actually matter.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- A thesis-change log should preserve the relationship state and falsifiable dependency assumptions that existed when the position was opened, then compare new evidence against that baseline.
- Current Micron and SK Hynix announcements show how product and partnership evidence can make an existing Nvidia relationship more strategically specific even when the graph edge itself is not new.
An investing agent should not only summarize what happened today. It should remember what had to remain true for the stock to be owned in the first place and flag when the underlying dependency structure stops matching that thesis.
That creates a different kind of research workflow: a thesis-change log. When a position is opened, the investor records the important customer, supplier, bottleneck, and product assumptions behind the decision. The agent then compares new relationship data and public evidence with that original state instead of treating every new headline as equally important.
The starting point should be a snapshot of the thesis
A useful thesis log begins with a small set of falsifiable statements. The investor might believe that a supplier has durable exposure to a major customer, that a customer relationship is expanding into a new product, that no single upstream dependency is large enough to threaten production, or that several portfolio holdings are economically independent.
The agent stores those assumptions alongside the relationship state that supported them. The point is not to preserve a long investment memo. It is to preserve the claims that would matter if they changed.
Without that baseline, an agent can summarize new information but cannot tell whether the information actually changes the reason the stock is owned.
Relationship changes can matter before the financial statements show the result
A new customer, a disappearing supplier, a changing relationship magnitude, or a new shared dependency across the portfolio can alter the structure of the thesis before the effect is obvious in reported earnings.
That does not mean every relationship update is material. The agent should compare the new state with the original investment logic and ask whether the change affects demand visibility, supply resilience, bargaining power, event concentration, or portfolio overlap.
The result is a thesis update, not a generic news alert.
Product evidence can strengthen or weaken the same relationship over time
Micron's current HBM4 production for Nvidia Vera Rubin is an example of public product evidence that can make a customer relationship more strategically specific. SK Hynix's multi-year Nvidia partnership does something similar by connecting the relationship to next-generation memory and AI-factory development.
For an investor exposed to either company, those developments can strengthen the case that Nvidia is not merely a historical customer relationship. They tie the dependency to a current technology roadmap.
A thesis-change log would record that shift in evidence quality rather than simply filing the announcements under "positive news."
The agent should distinguish a changed fact from a changed interpretation
Sometimes the relationship does not change but the investor's interpretation should. A large customer may become more risky because the customer's own demand outlook deteriorates. A supplier may become safer because an alternative source is qualified. A structural edge may become more important after a product announcement clarifies what the companies are actually doing together.
The agent should therefore log both relationship changes and evidence changes. The graph provides one layer of state, while filings, product announcements, contracts, and management commentary provide another.
That separation helps prevent the model from turning every new document into an invented relationship change.
Point-in-time data keeps the agent from rewriting history
A thesis log is much more useful when the agent can retrieve the relationship state that was actually known at the time of the original decision. Otherwise, current network knowledge can leak backward and make an old investment thesis look more informed than it really was.
Point-in-time relationship snapshots allow the investor to ask a harder question: given the data that existed when the position was opened, what has genuinely changed since then?
That is valuable for post-mortems, strategy improvement, and avoiding hindsight bias.
The most useful alert is "your assumption changed"
Most investing alerts are about prices, news, or earnings. A thesis-change system can generate a much higher-value alert: one of the assumptions behind your position is no longer supported by the same evidence.
The alert should explain the old assumption, the new evidence, and which part of the thesis may need to be reconsidered. It should not automatically label the change bullish or bearish.
That keeps the investor in control while making it much harder for a stale thesis to survive simply because nobody revisited the original reasoning.
MCP makes the workflow portable without changing the investment logic
Model Context Protocol gives AI applications a standard way to invoke external tools and data sources. That makes it possible for the same thesis-change workflow to retrieve relationship data inside different agent environments without rebuilding the investment logic around one chat interface.
The important part is still the data discipline. Tool access does not give the model permission to invent missing metrics, collapse company entities into securities, or treat a structural relationship as quantified exposure.
The agent should preserve direction, date, evidence state, and missingness each time it updates the thesis log.
The conclusion is continuous falsification
The best use of an investing agent is not to produce more summaries. It is to keep asking whether the assumptions behind the portfolio are still true.
A thesis-change log turns supply-chain data into a persistent research memory. The investor can see when an important dependency appeared, disappeared, became more strategically relevant, or created new overlap elsewhere in the portfolio. The agent is no longer just watching the market. It is watching the logic behind the investment.
The agent monitoring guide explains how to define the network around a holding after purchase. The connected-company filings guide shows how relationships can determine which outside documents deserve attention.
For relationship definitions and evidence limits, read the Altsets methodology. The Altsets documentation covers the available interfaces for agent and MCP workflows.