Track record
What our catalyst-timing signal was validated on, what we tested and threw away, and the live record as upcoming events resolve. Nothing on this page is a projection of future returns. · not financial advice
What the backtest actually showed
We tested one question across every past catalyst we could independently confirm: when does the anticipation run-up peak? It peaks before the event, and usually well before — which is what makes catalyst timing the thing worth signalling.
Read it as a distribution, not a promise: the peak lands somewhere in this spread, and the tail past 45 days is the single biggest bucket (34.7%). The signal places a catalyst on that curve — it does not claim to know the exact day, and a distribution this wide is exactly why we publish the live record below rather than a headline win rate.
The outcome model failed, so we stopped selling it
The obvious product is predicting which way a readout goes. We built it, graded it honestly, and it did not work — so it is not part of this product.
Direction (does it pop?) and magnitude (how much?) both came back null at n=497 confirmed binary events— correlation ≈ 0 with confidence intervals centred on zero, on two independent models and on their ensemble. An earlier “+8pp edge” turned out to be an artifact of broken date labels plus a query that silently truncated at 1,000 rows and graded only half the data. We found it, killed it, and rebuilt the labels.
That failure is why the timing signal makes sense. If the outcome of a binary event were forecastable, you would hold through it. It is not — so the only defensible edge is the anticipation that builds before the coin lands. Anyone selling you a readout-direction call is selling you noise.
How we grade
Most published hit rates never prove the event happened at all. Being strict about what counts is the whole point — it is also why our numbers accumulate slowly.
- 1We call it before the event.
Every graded signal was recorded while the catalyst was still upcoming — the prediction date is strictly earlier than the event date. No look-ahead, no retro-fitting.
- 2We wait for the catalyst to actually drop.
A scheduled date is often a bucketed guess — “H2 2026” becomes a single day on a calendar — and readouts routinely slip. So we confirm the real drop from an 8-K filing, a ClinicalTrials.gov results posting, or the headline that broke it.
- 3The confirmation window scales with the event.
A Phase 1 gets about 35 days, a large Phase 3 about 95, and a PDUFA about 100 — an FDA extension is a standard three months. Trials also get a 30-day early window, because interim and first-cohort data often lands ahead of any completion-derived date.
- 4Only then do we score it.
The call is compared against where the run-up actually peaked, split at 10 days before the event.
- 5No confirmation, no score — ever.
If the date passes and we never detect a drop, the prediction stays “awaiting confirmation” and counts as neither a hit nor a miss. Scoring against a date we cannot verify would grade us on fiction.
- 8-K filing1,723
- Headline that broke it157
- ClinicalTrials.gov results23
- Confirmed & scoreable1,903
- 1,011No drop detectedExcluded from every hit rate
34.7% of scheduled catalysts never produced a detectable drop. Those are thrown away, not counted as wins.
The live record, accumulating in public
The timing signal only fires on upcoming catalysts, so nothing can be graded until those events pass and we confirm they happened. This section is the honest state of that ledger today.
We started recording forward timing calls on 880 upcoming catalysts. Not one of them has both passed and been independently confirmed yet, so the graded count is zero. We could fill this space by scoring against scheduled dates we never verified, or by quoting the backtest as if it were a live result. Both would be a lie, so the number stays at zero until events earn it.
The 851 pending signals above are the proof the pipeline is running: they are timestamped before their events, and they will be graded whether they turn out right or wrong. The first ones resolve in the table below.
What resolves next
The soonest pending signals. Each was recorded before its event date — when these land and we confirm the drop, they become the first entries in the live record above.
| Event date | Ticker | Company | Signal state | Called on |
|---|---|---|---|---|
Jul 23, 2026 today | ARGX | — | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | AVXL | — | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | BNTX | BioNTech SE | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | ICLR | — | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | NVS | Novartis Pharmaceuticals | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | NVS | Novartis Pharmaceuticals | Priced in | Jul 22, 2026 |
Jul 23, 2026 today | SMMT | — | Priced in | Jul 22, 2026 |
Jul 24, 2026 in 1 day | JNJ | Janssen Research & Development, LLC | Priced in | Jul 22, 2026 |
Jul 27, 2026 in 4 days | AZN | — | Priced in | Jul 22, 2026 |
Jul 27, 2026 in 4 days | AZN | AstraZeneca | Priced in | Jul 22, 2026 |
Signal states describe timing — where a name sits on its anticipation curve — not a recommendation to buy or sell anything. Showing 10 of 851 pending; the full calendar lives on the catalyst board.