Our research, including the tests that found nothing
We test every claim the Terminal makes on our own recorded data first. We publish the tests that found nothing alongside the ones that found something. Methods and scripts are added to each note when it's written up.
- 13 September 2026
Cross-venue Bitcoin arbitrage is dead, and zero-fee venues have manufactured tapes
We took 7,199 samples of the top-of-book gap between two USDT venues. The median gap was 0.79 basis points, against a cheapest possible round trip of about 2. Across 258 liquid altcoins on seven venues, walking $10,000 through the book, nine in ten candidates came out negative. On every venue that advertised zero fees, we never found a real counterparty and a zero fee together.
- 12 September 2026
Residual lead-lag between venues is real and smaller than one tick
Cross-venue residual reversion on BTC measured +2.18 basis points. That is statistically real but smaller than one tick, and half the minimum cost on the cheapest execution venue we tested. Overlapping signals made the edge look 2.5 to 6.6 times bigger until we removed the duplicates.
- 8 September 2026
No level type beats a placebo over 365 days
We scored hand-placed and automatic levels (volume, time and structure) against random levels over a year of one-minute Bitcoin bars. None did better than the placebo. Based on that measurement, levels carry zero weight in our scoring.
- 12 September 2026
The divergence detector was broken, so its null result is withdrawn
Its pivots were only two bars wide, and it didn't require the oscillator to be in any particular zone. The earlier study that found no divergence edge was measuring the broken detector, so it is void. We are re-testing the rebuilt detector.
Full write-ups with methods are being prepared from our working notes. Each number is as we measured it on the date shown with its note.