SystematicEdge

@cheatyyyy everyone says the claude code harness sucks though 🤷‍♂️

why haven't i seen anyone try gpt-5.5 in claude code yet? it's most claude-shaped model yet

@nahcrof @aiCrof Por favor fix this good sirs I believe the model is a big bugged ( haven't tested the other ones yet ) - should I restart the session?

@SithHaters i believe you’re talking about look ahead bias or is this something else?

For starters: Try not creating a backtest that relies on "impossibly known data" In other words, it looks back 1 candle and decides if it was in a trade, on data this is known after the fact, but not before entry.... something that is not possible in real time. It's the most common culprit of outsized returns and over 80% wins.

absolutely kicking myself - I shoulda applied damn

@big_duca for the love of the game brother

Holy shit. I just saw someone in a coffee shop writing hand written code. No AI agents. He was typing into the editor directly. I almost wanted to go up to him and ask him why.

@nahcrof might ask you here - is reasoning effort/thinking configurable? I notice theres a field for the Python implementation but if using in opencode is there an equiv?

Why not crof?

@mweinbach you run /goal when your 5hr limit is at say 1%, and it will keep chugging through without hitting your weekly

Fun fact, you can run through 60% of a weekly ChatGPT Pro codex rate limit by using goal mode with GPT 5.5 xhigh fast in ~18 hours And frankly, that’s pretty generous

@SystematicIRE 8 Filters is insane, assuming each filters has 2 params / d.o.f indicators can have a lot of flexibility

Exactly what I would expect. This is a nice illustration of overfitting.

@fabianstelzer a date is injected - I presume only if the "date" tool itself is called otherwise if its in each user message and someone keeps an extra long thread, context may be largely dates then

why isn't Anthropic injecting a detailed date into each user message to give Claude a sense of time progression? This can't be a cacheing issue, each user message is a cache write anyway Seems like they are slotting a daily date into the system prompt instead? don't get it

@0xSero thoughts on quantizations of the above models - or do you run them as is normally?

Weekly best models for your hardware: ~~ 8 to 16gb ~~ Granite models are amazing: [NEW] - huggingface.co/ibm-granite/gr… Gemma-E4B is a good general QA model - huggingface.co/google/gemma-4… Qwen3.5-9B is the best at this level imo - huggingface.co/Qwen/Qwen3.5-9B ~~ 16 to 64gb ~~ Another larger Granite: This is a general chat model, really dense with world knowledge. [NEW] - huggingface.co/ibm-granite/gr… - Undisputed kings: The Qwens at various precisions: (Higher ceiling) - huggingface.co/Qwen/Qwen3.6-2… - huggingface.co/Qwen/Qwen3.6-2… The Gemmas at various precisions: (More efficient) - huggingface.co/google/gemma-4… - huggingface.co/google/gemma-4… ~~ 64 to 128gb ~~ - Ling is a new 100B~ contender decent agent [NEW] huggingface.co/inclusionAI/Li… - Mistral medium: from my experience their models have been the most consistent! [NEW] huggingface.co/mistralai/Mist… ~~ 128gb - 256gb ~~ Undisputed king: DeepSeek-V4-Flash [NEW] huggingface.co/deepseek-ai/De…

@tonbistudio codex : proceeds to create a file called “kimi 2.6 1T” and runs it , i swear sometimes model takes the biggest shortcuts

I’m just going to tell codex /goal get kimi k2.6 1T param model to run locally on my RTX 3060 and see what this madman cooks up

@banteg I thought if you prompt it/add a new msg as it is doing a goal, that will count as a turn/usage and you'll hit the limit?

im at 0% weekly limit, but i have one codex session that is still running. now i need to fit all my work there and constantly distract it so it doesn't reach the goal. if it reaches it, the world ends, and i have to rest. what a funny situation to be in.

@TheRealBadFish No NBA player I know advertises: "Buy my course and you'll be in the NBA too within 6 months!" It doesn't matter if they get paid by brands or by customers themselves tbh

Basketball stars do endorsements because brands pay them. But it’s flawed LeBron or Curry aren’t selling “How I shoot 3s” courses to retail investors their actual performance is opaque. Their endorsements are transparent so let see broker stats or STFU

@JoeyMorganNQ prob coz the promises / outcomes promised as a result of the course. personally find it scummy when people say do my course to make xxx a day!

Idk why people hate on people that sell courses. I got started as a trader because of a course. Sometimes people need it to give them a push and would have never started trading had it not been for that course. If it’s done with transparency and honesty I see nothing wrong with them.

@Bandz_Trading their 12hr usage is sending one hello every hour 😂

Complete lie and marketing

@nahcrof absolute baller move - no one else in the game would allow coding plans for this but you

This is at least partially my fault I can't speak for other services since I only really use them for testing and personal use, so I haven't read anyone else's ToS but I personally do not mind if a user resells crof I would generally prefer that if a user is doing it, it isn't through the subscriptions since I have more guarentee on making my money back from paygo pricing but they've been very profitable so far My personal view on these things is if you pay for my service, and you're not harming anyone or doing anything illegal using it, I feel like you are entitled to the usage you paid for. All that being said, the terms did say "without authorization" and @inlovewithgo did ask for my permission to resell my service As long as you ask me, I'll probably say yes so long as you're not currently expecting any crazy usage and if that changes I'll always talk to you going forward (plus, if a user wants to resell a sub, they're gonna resell someone's, I'd rather them spend the money reselling mine than someone else's)

@ism_sol Either codex, or spend a quarter ($5) and get yourself a @aiCrof sub + opencode , cheapest sub in the game for 500 requests a day

woi developer, lu ada $20 lu bakal beli yg mana? stress bat gua antara dua itu – Claude – Codex

@Mazi_Capital this would absolutely destroy me man - head up you've got it

April 2026 Results 📊 Wins: 1 | Losses: 8 | Break-even: 0 Return: -4.9% ❌ (including commissions) Payouts Secured: 1 | Payout Total: $594.16 Abomination of a month - glad it’s over. Never expected a month with a win rate like this, but trading has a way of humbling you. Took a lot of hits, but accounts are still intact - that’s what matters. Confidence took a hit too, but backtesting has helped rebuild it. A lot of lessons taken from this month. Moving into May with a few key adjustments: - Removing USDCAD - spread differences across accounts negatively impact expectancy, and the win rate isn’t high enough to justify it. - Adjusting execution - adding a +1 pip buffer to account for spread, costs, and copier latency. - Returning to previous risk model - better aligned with the sequence of wins and losses. Focus now is on executing, managing risk, and letting the model play out. Let's work our way out of this drawdown 🫡 Account Progresses: @the5erstrading $100,000 Funded: -5.7% @the5erstrading $20,000 Funded: -5.25% @the5erstrading $5,000 Funded: -6.2% @AlphaCapitalUK $100,000 Funded: -5.7% @AlphaCapitalUK $100,000 Challenge P2: -8.75% @BrightFunded $100,000 Public Challenge P1: -2.7% @BrightFunded $100,000 Challenge P2: -9.4% @DarwinexZero: +2.4% (Silver)

@tradertheory win rate is irrelevant without RR lmao - some profesh traders have a legitimately high RR too

Most professional traders have a win rate near 40 - 50%. But you believe that clickbait thumbnail guy saying he has a 90% win rate strategy.

@AzFlin and you'd be stick largely with claude's expensive models

you don't need an openclaw/hermes just open up a new terminal tab with claude code and do what you needa do