How to Use Supply-Chain Data to Try to Disprove an Investment Thesis
July 29, 2026
Altsets
Research by Altsets Research
Use supplier, customer, and network evidence as a falsification tool by defining what relationship facts would weaken or contradict a stock thesis before searching for confirmation.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- Existing relationship data can challenge common assumptions about diversification, supplier importance, customer concentration, and market sensitivity without requiring a bearish conclusion.
- A stronger workflow defines the relationship evidence that would contradict the thesis first, then uses network data and public sources to test whether the thesis survives.
Supply-chain data is most useful when it can prove an investment thesis wrong. If the network is used only to find facts that support a stock idea, it becomes another confirmation tool. The better workflow is to state the thesis first and then ask what relationship evidence would contradict it. Several supplied Altsets examples show how this works.
Thesis: "Different sectors and countries mean the portfolio is diversified"
A portfolio can contain ASML, Micron, Nvidia, and Quanta Computer and still sit inside one connected economic chain. The supplied network contains a path from ASML to Micron, Micron to Nvidia, and Nvidia to Quanta. Those companies span different business functions and countries.
The network does not prove the portfolio is badly diversified. It disproves the simpler claim that sector and country labels are sufficient evidence of independence. The supply-chain diversification guide turns that contradiction into a portfolio workflow.
Thesis: "A small cost-share supplier cannot be a major risk"
Shin-Etsu Chemical is associated with only 1.34% of Micron's cost base in the displayed relationship data. A cost-only screen could rank it as relatively unimportant. Public product information shows Shin-Etsu sells semiconductor silicon wafers and photoresists.
Micron warns that some material categories can have limited or sole-source supply and can take time to qualify. That evidence does not prove Shin-Etsu is a sole-source Micron supplier. It does disprove the assumption that a small cost share automatically means low operational criticality. The financial materiality versus operational criticality guide handles that question directly.
Thesis: "One concentrated customer means one end-market bet"
Tesla is associated with 19.03% of LG Energy Solution revenue in the displayed data. That is meaningful customer concentration. LG Energy Solution also publicly describes a Tesla energy-storage relationship and operates manufacturing capacity capable of serving both EV and ESS demand.
So the relationship does not support the simple conclusion that Tesla concentration equals a single vehicle-demand exposure. Customer concentration and end-market concentration are different. The customer-versus-end-market guide decomposes those layers.
Thesis: "A relationship percentage predicts the stock reaction"
Micron's mapped Nvidia revenue exposure is 17.62%. That is economically important. It does not determine how Micron shares react to Nvidia news.
If the thesis relies on that percentage as a stock beta, the thesis contains an unsupported step. The exposure-versus-stock-beta guide shows where operating exposure ends and market sensitivity begins.
Thesis: "A company's filing contains the full dependency picture"
Nvidia reports major customer concentration without naming its largest direct customers. Micron warns about limited-source supply without quantifying every named supplier relationship. The filings are authoritative for what the companies report.
They are not designed to map every named cross-company dependency. The filings-versus-network guide shows how each evidence source fills a different gap.
Falsification needs a specific failure condition
A weak thesis says: I like this company because AI demand is strong.
That is difficult to disprove. A stronger thesis says: Micron's upside depends materially on Nvidia-linked demand remaining strong, and I expect the relationship to remain economically important.
Now the investor can look for contrary evidence: Nvidia exposure falls, another customer becomes larger, product mix shifts, contracts change, capacity moves elsewhere, and the relationship disappears in point-in-time data. The thesis becomes testable.
The graph can generate disconfirming questions
For any stock idea, ask:
- Which supplier would make this thesis fail?
- Which customer is more important than the story assumes?
- Which shared dependency makes diversification weaker than it appears?
- Which structural-only edge challenges the narrative?
- Which public disclosure contradicts the assumed product function?
- Which historical snapshot shows the relationship is not stable?
Those questions are often more valuable than asking the graph to confirm the original story.
Do not manufacture contradiction for its own sake
Falsification is not automatic bearishness. A thesis can survive the test. The purpose is to identify what would have to be true for the investment idea to be wrong. If the relationship data supports the thesis after a serious attempt to challenge it, confidence becomes more meaningful.
LLMs can help if the prompt is adversarial to the thesis
An LLM connected to relationship data can be asked to search specifically for evidence against a claim. For example: Assume my thesis is wrong. Which supplier, customer, network path, or missing metric should I investigate first?
The model still needs the evidence rules described in the LLM grounding guide. The goal is not creative contradiction. It is structured retrieval of contrary evidence.
A repeatable thesis-falsification workflow
- Write the investment thesis in one sentence.
- Identify the relationship assumptions inside it.
- Define what data would contradict each assumption.
- Map suppliers, customers, and second-order dependencies.
- Check directional economic metrics.
- Add public product and filing evidence.
- Search deliberately for conflicting evidence.
- Keep missing data as unknown rather than favorable.
- Record whether the thesis survived, weakened, or changed.
- Re-run the test when the network or public evidence changes.
The workflow turns supply-chain research from a collection of interesting relationships into a way to test whether an investment story deserves to survive. For relationship definitions and evidence limits, read the Altsets methodology. The Altsets documentation covers the interfaces available for visual, LLM, and programmatic research. Browse Supply-Chain Data Use Cases for narrower questions that can be used as individual thesis tests.