Finding the gaps
Why small, unobvious gaps in the record can undo a decision or an investment, and why whoever sees them first holds the opportunity. Two worked examples.
Big decisions rarely fail where everyone is looking. They fail at a small gap nobody connected: a reviewer who disagreed, a study that did not replicate, a requirement that was never met, a company that needed money faster than it said. These gaps sit in the public record, spread across documents, agencies and years. Whoever connects them first can avoid the risk, or take the other side of it.
Why small gaps matter#
Every approval, deal and system rests on a chain of evidence. When one link is weak, everything built on it inherits the weakness, usually without anyone noticing, because the weak link is one sentence in one document among thousands.
- For a business, a gap is a risk to price in before it becomes a write-off, a recall, a failed launch or a lawsuit.
- For an investor, a gap is where the market's view and the record disagree, which is where returns and losses come from.
- For anyone building on a decision (a partnership, a formulary, a supply contract), a gap is the question to ask before signing.
The kinds of gaps the graph shows#
Because the record is a knowledge graph, gaps show up as structure, not as a feeling:
| Gap | What it looks like in the graph |
|---|---|
| Contradiction | Two positions in the same decision that point opposite ways, such as two reviewers disagreeing |
| Thin support | A decision linked to a single line of evidence, or to evidence its own committee voted against |
| Drift | A position that departs from an earlier one on the same question (see the ApoE ε4 example) |
| Missing link | A requirement with no fulfilment, a question with no position: [no position recorded on this thread] |
| Record vs. reality | An event the company reported that the agency's record does not show |
| Timing | A signal in one domain (a flight, a filing) that comes before the announcement in another |
Example: the gap Berkshire filled at Occidental#
In April 2019 Occidental was bidding against Chevron for Anadarko. Its gap was simple and unobvious from the outside: to win, it needed $10 billion, fast. On Sunday, April 28, Occidental's jet (N550XY) flew to Omaha and back; on April 30, Berkshire committed the $10 billion. The record shows who priced the gap, and how:
| From the record | |
|---|---|
| Berkshire's terms | $10 billion of 8% preferred stock, $800 million a year in dividends, paid ahead of common shareholders, plus a warrant for 80 million shares at $62.50 (8-K, May 3, 2019) |
| Occidental's shareholders | −10.3% from the close before the flight to the day Occidental signed for Anadarko; −16.2% by July 31, 2019 |
| Afterwards | Dividends paid partly in Occidental stock in 2020; the preferred bought back from 2023; in 2025 Berkshire agreed to buy Occidental's chemical business, OxyChem, for $9.7 billion in cash, completed January 2, 2026 |
The opportunity was the gap between what Occidental needed and what it could raise on its own terms. The signal that someone was about to fill it, a jet in Omaha on a Sunday, was public two days before the announcement. See the private-jet example and the full cited case on underlying.ephemerisx.com.
Example: the gaps in Aduhelm's approval#
On June 7, 2021, FDA granted Aduhelm accelerated approval for Alzheimer's disease. Every one of the gaps below was in FDA's own documents on approval day, in different reviews and hundreds of pages apart:
| Gap | What the record says | Source |
|---|---|---|
| Thin support | Approved on a surrogate: “based on reduction in amyloid beta plaques”, not on a clinical benefit | Label, p. 2 |
| Contradiction | The statistical review: Study 301 “failed for the high dose (p=0.83)”. The clinical review: Study 302 was “robust and exceptionally persuasive” | Statistical review, p. 9 · Clinical review, p. 14 |
| Committee against | Asked whether Study 302 alone gave strong evidence: 1 yes, 8 no, 2 uncertain. On its key question: 0 yes, 10 no, 1 uncertain | Clinical review, pp. 133–134 |
| Missing link | Approval required a new randomized trial “to verify the clinical benefit”; none had reported | Approval letter, p. 6 |
Any one of these is a footnote. Together, on one screen, they describe an approval resting on a surrogate, a failed twin study, a divided review team and a committee that voted against it, with the proof of benefit still to come.
What followed#
- Biogen discontinued Aduhelm in 2024. That is the company's report; it is not in FDA's record. As of September 28, 2026, Drugs@FDA still lists Aduhelm's marketing status as Prescription, and no withdrawal notice appears in the Federal Register. The Alzheimer's record shows this as an event “not in the FDA record”, itself a record-vs-reality gap.
- The later antibodies in the class, Leqembi and Kisunla, reached traditional approval after committees voted in favour (6–0 and 11–0). See Committee vs. approval.
How seeing the gaps could have changed a decision#
Seeing the gaps together would not have predicted what happened. It would have changed the questions asked before committing to Aduhelm:
- A partner or supplier would ask what happens to the deal if the confirmatory trial fails or is never finished.
- An investor would ask how much of the value depended on an approval its own committee opposed.
- A health system or insurer would ask whether to commit before the clinical benefit was verified.
- A competitor would ask what FDA would need to see from the next drug in the class, and the later record answers it.
From gaps to decisions#
Pro is being built to find gaps like these across every domain in the record, turn them into risk assessments with their evidence cited, and alert you when a new document opens or closes one. Team plans deliver them into your own risk and tracking systems. See Pro overview.