Pinned Tweet
How to use PerpsWatchdog, page by page, in 4 minutes.
Funding and trading costs, pre-IPO perps, stock perps after hours, insider buys, the early radar, and our own measurements of Hyperliquid's trades.
Edge is open to everyone until 27 November.
perpswatchdog.com
🤖 Made with AI
2
5,675
We replayed six months of our engine's own published funding alerts, plus a backward replication over 2020–25 (16,575 posts). Verdict: failed, no edge. Directional funding alerts looked like random entry. Link in our profile.
5
Pre-IPO perps on Hyperliquid this week: 3 markets on 3 companies that have not listed yet, $103.5M open interest.
Anthropic: $96.5M across 1 market
OpenAI: $6.4M across 1 market
Each builder states what its price means; we compare like with like only.
18
We tested whether the Korean 'kimchi' premium and the Coinbase premium work as sentiment gauges for returns. In our data they did not: the study failed and sits on our public failed-ideas list. Link in our profile.
4
One of 19 ideas we tested and dropped: Fibonacci, Gann, zeta zeros, Tesla 3-6-9 and 13 universal constants. Verdict: failed. No special number beats random numbers, and the zeta-zero control caught a placebo-design flaw.
Shorts pay 25% a year to hold NEAR perps on Hyperliquid, while the price rose 2.0%: the crowd is leaning against the move.
Context: lowest in 20 days.
6
Study failed: pullback continuation, cross-sectional momentum and time-series momentum on crypto perps. Lesson: the pullback "edge" was market direction. Time-series momentum stays a paper experiment. Link in our profile.
5
Funding-spread carry (cross-venue) failed. We tested it on the 12 months before the period where it looked good. It did not hold. Lesson: the edge came from one recent Hyperliquid-vs-exchange regime. Link in our profile.
6
Hyperliquid vaults: in 95 of 139 vaults with outside money, depositors as a whole earned less than the return the vault shows.
Citadel: strategy +4009%, depositors +247%
Brob: strategy +1312%, depositors +322%
Hyperliquid public data, approximate.
15
We tested textbook chart patterns and 13 classic indicators on crypto perp returns. Verdict: failed. Patterns showed up as often in random series, and the triangle top failed its holdout (unseen data). Link in our profile.
Chart patterns and 13 classic indicators: we tested pattern detection and an indicator zoo. Patterns appeared as often in random series, and the triangle top failed its holdout. One of 19 ideas we dropped. Link in our profile.
9
What a $1M market order costs, buy and sell (round trip, before fees). Median of 14 days, weekdays, Hyperliquid vs dYdX:
BTC: 1.3 bp vs 19.2 bp
ETH: 2.7 bp vs 25.5 bp
SOL: 7.4 bp vs 35.4 bp
XRP: 12.4 bp vs 85.5 bp
Spread plus depth, read from each venue's public order book.
21
We tested whether spot ETF flows and stablecoin mints predict crypto returns. Verdict: failed. ETF flows follow price: +0.43 same-day vs +0.07 next-day correlation. Flows react on the same day and do not lead the next. Link in our profile.
4
We tested Fibonacci, Gann, zeta zeros, Tesla 3-6-9 and 13 universal constants against random placebo numbers. Verdict: failed. No special number beat random, and the zeta-zero control caught a placebo-design flaw.
17
Did the weekend stock perps call the US open? 2026-11-16
Hyperliquid perps at 09:25 New York: -0.85%
Market at 09:45: -1.02%
Same direction.
Running record: 12 of 13 days.
5
We tested whether Bluesky's public firehose carries enough market chatter for a trending coins board. It failed. Over 2.1 h we read 304,023 posts (~147k posts/h), but only 181/h named a market instrument. Link in our profile.
1
On 32 weekends between 2 Feb and 19 Oct 2026, the Monday 09:00 New York perp price on Binance's TradFi perps pointed the same way as the move to the 09:35 reference price 83% of the time. Descriptive only, not a trading edge. Link in our profile.
4
Wall Street is shut. These perps are not. Biggest moves on Hyperliquid's 24/7 stock perps since Friday's close:
Strategy -9.72%
PURRDAT -9.71%
Circle -6.80%
Coinbase -6.30%
Direxion Semis 3x -5.99%
One of 19 ideas we dropped: 34 indicators, combinations, ML, regimes. Verdict: no tradable edge. Short-term reversal is real (rank IC ≈ 0.04) but dies to costs within hours; residualising doesn't help. Link in our profile.
We tested 34 indicators, a 523-trial combination study, machine learning and pairs trading. Verdict: failed, no tradable edge. Short-term reversal is real, but costs erase it. Residualising doesn't help. Link in our profile.
5






