How to write a falsifiable investment thesis

Most investment theses are constructed backwards. The investor decides to own the asset, then builds a story that justifies the position. The story is designed to accommodate any outcome — if the stock goes up, the thesis is validated; if it goes down, the thesis needs more time or the market is being irrational. A thesis built this way cannot fail, and a thesis that cannot fail cannot teach you anything.

A falsifiable thesis is one that specifies, before you invest, exactly what evidence would prove it wrong. It is not about being pessimistic or adding caveats. It is about making a concrete prediction with clear success and failure states, so that six months later you can look at what happened and know whether your reasoning worked.

Every falsifiable thesis has three components. First, a directional claim: a specific thing you believe will happen, not a vague statement that the asset is interesting. Second, a mechanism: the causal chain that connects some observable condition in the world to the asset's price. Third, and most important, at least one condition that would prove the thesis wrong — a pre-committed trigger tied to the mechanism, not to the price.

Here are two rewrites from unfalsifiable to falsifiable.

Unfalsifiable: "NVIDIA is a leader in AI and demand for compute will keep growing." Falsifiable: "NVIDIA data center revenue grows faster than 40% year-over-year through calendar 2026 because hyperscaler capex is structurally shifting toward inference-dense workloads. I am wrong if revenue growth drops below 30% for two consecutive quarters while hyperscaler capex continues to grow above 20%."

Unfalsifiable: "Renewable energy is the future and NextEra Energy is well-positioned." Falsifiable: "NextEra's regulated utility base rate case approvals will outpace inflation by 200 basis points through 2027 because Florida's load growth from data center demand gives regulators political cover to approve above-inflation returns. I am wrong if the next two rate cases come in at or below CPI plus 100 basis points."

Notice what changed. The falsifiable versions name a specific metric, a causal mechanism, and a disconfirming condition. They can be proven wrong by events. The unfalsifiable versions can absorb any outcome.

Tensile requires every thesis to declare its invalidation condition before the position can be entered, making unfalsifiable theses structurally impossible to record.