Look for what could invalidate an investment thesis.
Bring a thesis and its supporting evidence. Ask reviewers to challenge the assumptions, construct an opposing case, and identify facts to verify.
Two free Refinement reviews each month · No credit card required
A convincing narrative can leave out the downside.
Provide the time horizon, data sources, valuation assumptions, and risks already considered. Separate observed facts from forecasts and preferences.
Use the review to prepare questions and further research. It does not verify market data or make an investment suitable for your circumstances.
What to look for in the review.
Name the assumptions
Ask what must hold for the thesis to remain plausible.
Build an opposing case
Consider a credible downside scenario using the same evidence.
Check the evidence
Verify dates, calculations, and source claims before relying on a finding.
A thesis review should test the causal chain
The panel should not merely produce a longer risk section. It should connect each claim to evidence, a failure mechanism, and a decision consequence.
| Thesis element | Confirmation-biased treatment | Adversarial treatment |
|---|---|---|
| Market | Large category and favorable trend | Who captures value, what constrains adoption, and what changes the pool? |
| Product | Differentiated feature narrative | Durability, switching cost, substitutes, and replication path |
| Economics | Headline growth or margin | Cohorts, cash needs, sensitivities, and hidden operating leverage |
| Catalyst | Event expected to unlock value | Probability, timing, observability, and failure alternative |
| Valuation | Upside case against selected peers | Scenario range, base rates, dilution, and required expectations |
| Risk | Generic disclaimer list | Mechanism, leading indicator, impact, mitigation, and trigger |
| Decision | Buy or pass conclusion | Position conditional on evidence, uncertainty, and portfolio context |
Prepare a useful review.
Give reviewers a clear question and enough context to examine it.
Separate facts from estimates
Mark what comes from audited statements, primary research, management claims, third-party estimates, and your own model.
Build scenarios from mechanisms
Optimistic, base, and adverse cases should differ because customer behavior, competition, financing, execution, or regulation changes—not because a spreadsheet applies arbitrary percentage shifts.
This is thesis stress-testing, not investment advice or a forecasting oracle.
Models may use stale, incomplete, or incorrect market knowledge. They cannot verify a filing, private data room, channel check, or management statement unless you provide reliable source material and a qualified person validates it. Never treat model consensus, confidence, or a chairman conclusion as a recommendation to transact.
Protect confidential information and follow securities, employment, privacy, and contractual obligations. Do not upload material non-public information or restricted research. The panel can improve the structure of diligence and surface neglected questions; suitability, valuation judgment, portfolio risk, and final decisions remain human responsibilities.
What to know before you run the panel
What should I provide for an investment thesis review?
Provide the thesis, evidence sources, model assumptions, valuation framework, catalysts, time horizon, alternatives, known risks, and the decision rule. Separate verified facts from estimates and confidential information. State what would cause you to change or abandon the position.
Which format is useful for investment research?
Devil’s Advocate is strong for attacking a preferred thesis. Oxford Union works for a clear invest-or-pass motion. Delphi helps protect independent initial views. Steelman is useful when the counter-thesis has been dismissed too casually. None replaces regulated advice or primary diligence.
Does agreement mean the answer is correct?
No. Models can repeat the same mistake or rely on the same weak source. Read the objections and check important claims against evidence.
Why use other models instead of asking one to critique itself?
Self-critique can help. Other models and specific review roles provide additional responses to compare. The format determines whether reviewers begin independently or see earlier answers.
Does this replace human review?
No. Use the findings to improve a draft or prepare questions for a responsible reviewer. The models do not approve a decision or verify every claim.
Bring your next draft.
Create a free account to draft or import your work. Choose the reviewers and settings before starting a review.