Ledger
How many hypotheses we registered and how many we archived as empty, by the kind of mechanism we were looking for. Registered means formalized in the research log; archived means closed as empty for any reason (no signal, no edge after costs, not enough data). The share reflects which ideas we chose to test, not a failure rate of strategies in general. Updated October 20, 2026.
76
hypotheses registered
51
archived as empty waters
4
published log entries
25
other research, not disclosed
Why the published cases were empty
A barren result is not one result. Each published entry names the mechanism that emptied it.
| Reason | Meaning | Published entries |
|---|---|---|
| not_detected | No detectable effect in our test | 3 |
| negative_net | Negative or zero net of costs | 1 |
By kind of mechanism
Share of registered hypotheses archived as empty.
The numbers
| Mechanism | Registered | Archived | Archived share |
|---|---|---|---|
| event / calendar | 27 | 15 | 56% |
| trend / momentum | 11 | 10 | 91% |
| flows | 10 | 7 | 70% |
| microstructure | 7 | 6 | 86% |
| mean reversion | 7 | 4 | 57% |
| other | 7 | 4 | 57% |
| cross-sectional | 3 | 2 | 67% |
| volatility / options | 2 | 1 | 50% |
| carry / funding | 2 | 2 | 100% |
A high share of empty waters is not a flaw of the search: it is what a rigorous search looks like. Real costs, out-of-sample data, nulls and multiplicity leave very little standing.