How to Build a Falsification Framework for Any Earnings Report
The Durability Curve
Subscribe to The Durability Curve on Substack — free weekly analysis on AI infrastructure, verification, and durable systems.
A repeatable system for making falsifiable investment predictions before earnings. Includes specific thresholds, observability checks, and pre-commitment protocols. Worked examples from the NVDA May 20 earnings framework.
Step 1: Identify the Structural Claim
Every stock thesis rests on a structural claim about how the world works. For NVIDIA, the claim is: 'AI compute is the binding bottleneck in the infrastructure stack, and NVIDIA owns it.' That claim makes specific, testable predictions about revenue growth, margin stability, and competitive position.
The structural claim is the thing you are actually betting on — not the quarter, but the trajectory. A good structural claim survives a single bad quarter. A bad one collapses on the first miss.
Step 2: Define Specific Thresholds
A falsifier must have a number. 'Demand slows' is not a falsifier. 'Data center revenue below $72B' is. The threshold comes from the supply chain: when six independent optical companies all report demand growing 27-90% YoY, a data center number below $72B would contradict that signal. The threshold is grounded in observable data, not arbitrary.
Set the threshold before the print. If you set it after seeing the number, you are rationalising, not falsifying.
Step 3: Ensure Observability
A falsifier must be observable within the earnings call or 8-K filing. For NVDA: data center segment revenue (explicit line item), Q2 guidance (explicit number), non-GAAP gross margin (standard disclosure), and CEO commentary on optical interconnect (transcript). Every falsifier must be checkable within 24 hours of the print.
If you cannot observe the falsifier, it is not a falsifier — it is a belief dressed up as a thesis.
Step 4: Pre-Commit to Interpretation
Decide now what each outcome means. Write it down. 'If falsifier 1 fires, the bottleneck has migrated from compute to power/construction. The photonics thesis takes a temporal hit but does not break. If falsifier 2 fires, demand is genuinely flattening — revise conviction downward.'
The act of writing the interpretation before the data arrives is what makes the framework work. It prevents you from moving the goalposts after the fact.
Why Falsification Works
Falsification frameworks protect against two cognitive biases: confirmation bias (seeking evidence that supports your view) and anchoring (failing to update from your initial position). By pre-committing to specific thresholds and interpretations, you make it harder to rationalise away disconfirming evidence.
The framework does not tell you what to buy or sell. It tells you when your thesis needs revision. That is a different, more durable kind of edge.
See also: The Five Layers of the AI Infrastructure Thesis — the five-layer framework applied to NVDA's May 20 print.
See also: Three Signals the Market Is Missing on NVDA — three signals ahead of May 20.
Subscribe to The Durability Curve on Substack — free weekly analysis on AI infrastructure, verification, and durable systems.
Browse research products on Gumroad — in-depth reports, frameworks, and data packs.
All Telegraph articles — index page.
For deeper analysis: The Five Laws framework explained — read the full article on Substack | Commission custom research — the AI Research Commission on Gumroad
Subscribe: Free weekly analysis at free weekly analysis on Substack
Research: AI Research Commission - custom research — buy on Gumroad