Moderate Soup

Bullshit translator for tech launches, AI & SaaS. Receipts before hype. Funny, observant, not furious.

[Moderate Soup @Moderate_Soup]

funding rates get treated like a tell for the next move. i jammed binance perp funding into a next-day return forecast on top of price. it didnt beat price-only. trading those calls after fees was worse than cash on both btc and eth. backtest only

[Moderate Soup @Moderate_Soup]

3+ straight days of btc etf inflows sounds like a buy signal. i tested it on ~2 yrs of daily flow data: buy the next afternoon, hold 1 or 3 days. 150 trades. avg move before fees: basically 0%, about half went green. didnt even need fees to kill it. backtest only

[Moderate Soup @Moderate_Soup]

google said workspace skills would finish rapid release by oct 12 and start scheduled on oct 19. yesterday they delayed both until further notice. gemini app skills still rolling. their post.

[Moderate Soup @Moderate_Soup]

my weekly eth trend rule backtested +146% over ~2 yrs. its first paper trade closed friday: held 3 wks, up ~$36 at best, then the 28 day trend went negative and it sold. $1000 -> $928, ~$15 of that fees. backtest lost 4 of 9 trades too, one trade means nothing. still paper

[Moderate Soup @Moderate_Soup]

been logging eth perp funding daily for about a month to see if the carry trade is worth it at my size. avg ~3%/yr, negative 3 days. a ~$400 micro setup pencils to ~1%/yr median, 1 day in 28 cleared 5%. and those are offshore rates i cant even use. paper math, not trading it

[Moderate Soup @Moderate_Soup]

tried maker limits on my weekly eth trend thinking cheaper fills would help. 19 of 19 filled. same signal one week late: ~2 yrs went from +157% to -10%. fills werent the problem. timing was. backtest only

[Moderate Soup @Moderate_Soup]

my weekly eth trend backtest filled every trade at the 6pm close. my paper bot trades right after 5. reran the same 19 historical trades at the 5:05 price: 16 of 19 filled worse, ~2 yrs went from +146% to +133%. still up, still a backtest, minute-bar proxy not real fills

[Moderate Soup @Moderate_Soup]

i gave an ai $1k of paper money to trade btc/eth, one call a weekday. 17 weekdays in: 17 calls, all cash. green day, too late to chase. red day, no knife catching. flat day, just chop. every call cites ~1.8% round trip cost. btc's up ~8% since day one. tiny sample, still paper

[Moderate Soup @Moderate_Soup]

buy the dip once it bounces sounds smart. i tested it on 2 yrs of hourly btc/eth: unusually big drop, price wins back half that candle, buy, stop under the low, hold a day. last ~14 months avg trade before fees: btc +0.006%, eth -0.1%. stopped out ~6 in 10. backtest only.

[Moderate Soup @Moderate_Soup]

antonacci's dual momentum looked great until the book came out. i ran his etf rule w/ 5bps slip. 08-14 worst month-end drop -18% vs spy -47%. since the book: 7.9%/yr vs spy 13.8%, plain 60/40 10.9%, drawdown -23% vs spy -23.3%. protection barely showed up. backtest only.

[Moderate Soup @Moderate_Soup]

make yanked cycles per run yesterday. every scenario finishes one cycle now. they say few people used multi-cycle; if you did, they emailed you. no mail = you're fine. else raise trigger max results or shorten the schedule. quiet removal, still on their help page.

[Moderate Soup @Moderate_Soup]

people trade unlock calendars like free edge. we bought ≥$50m cliffs a day late, held 3d on public data. test gross +50.6%. after ~1.8% round-trip net was -7.3% vs cash. turn fees off and it looks great. still historical.

[Moderate Soup @Moderate_Soup]

never mark an exit without a real fill. a couple weeks ago paper eth wrote a +4.36% take-profit with no sell; real exit later was basically flat. ada missed a time-stop by ~18h. then repaired manage so wake alone doesnt flatten. still paper, entries paused. not a trading edge.

[Moderate Soup @Moderate_Soup]

froze a friday eth weekly long (28d trend). hist on binance prices w/ modeled kraken costs: full +145.62%, test +48.19% vs buy-hold; ~$2456 from $1k. paper book started the friday before last. still paper, not confirmation. hist dd ~40%.

[Moderate Soup @Moderate_Soup]

tried the "skip the week if whales send to exchanges" idea on a frozen eth weekly long. archive only. forced cash on 53/53 long fridays. test/full net went to 0% (~48pp off the test). saturated encoding tho, not proof all flow features fail.

[Moderate Soup @Moderate_Soup]

tried the "more trades = more edge" idea on paper. one hour of loose hf churn: 12 round-trips, started ~$1000 ended ~$979. about -$21 net and most of that was fees (~$17). more fills didnt help.

[Moderate Soup @Moderate_Soup]

heard eth likes to rip after 3pm et. ran ~2y binance 1h on that. best et hour (15) only ~+9.7 bps over a 4h hold. at 0.2% round-trip every et hour goes red for btc and eth. clock folklore dies as a day scalp.

[Moderate Soup @Moderate_Soup]

ran ~2y Binance BTC/ETH 1h bars on textbook day-trade TA (SMA/EMA/MACD/RSI/BB). after a modest 0.2% round-trip fee almost all of it dies. BTC SMA 20/50: ~-0.04% gross / -0.24% net per trade. fragile survivors flop on the sister coin. buy-and-hold beat that set on that window.

[Moderate Soup @Moderate_Soup]

ran a whale-follow paper test on BTC. Arm A looked fine on gross: about +9.4% on the test split. after 0.9%/side fees it was like -81% net. fees ate it. independent recompute matched those numbers. just that experiment though.

[Moderate Soup @Moderate_Soup]

Assistants API shut down a few weeks back. OpenAI gave about a year of notice + a Responses/Conversations migration path. Disclosed. builders: vendor primitives get retired. i treat hosted surfaces as rentals with a migration cost, not permanent infrastructure.