StockInsights69

I vibe-coded a full AI sports intelligence terminal from scratch. 259 backend functions. Two competing AI models. A grading system that tells me when I’m wrong. STREAK (Chatgpt) and SHARP (Claude) run independently. When they agree, the win rate jumps to 59.3%. It took months. The build guide is now $49. 42 pages. No filler. What’s in it: → The full pipeline architecture → Two-model philosophy (and why one model always has a blind spot) → The base44 gotchas that cost me hours each → Battle-tested AI prompts for building with #Claude and #ChatGPT → How to grade your own picks and use the data to improve Not a course. Not theory. Built from a live product that posts real results every day. boshroidfantasynetwork.com/boshroidtermin… → Build Guide tab #buildinpublic #vibecoding #Claude #ChatGPT #AI

What the hell they got going on in Italy

April 4 — SHARP (#Claude) Model 60 bets, 58.3% WR, +3.6% ROI Points: 61.3% WR, +14.1% ROI (31 bets) Threes: 71.4% WR, +19.5% ROI (7 bets) Rebounds: 52.9% WR, -11.5% ROI (17 bets) Assists: 40.0% WR, -31.9% ROI (5 bets) Points and threes both had strong nights. agreement tier (when both models agree) went 62.5% WR +13.1% ROI — that signal continues to be the cleanest. Assists remains the weak spot. the juice filter we shipped yesterday didn’t fully take effect on last night’s slate, so some heavy juice assists bets still went through and those went 25% WR as expected. Tracking whether the fix is applying correctly going forward. Rebounds is mixed. the heavy juice bets actually went 62% WR last night which complicates the cutoff decision. Watching another few nights before drawing conclusions there. results posted daily boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #sharkdesk

April 3rd Models Review STREAK (#Chatgpt): 55.1% WR, -2.4% ROI — 156 bets rebounds: 66.7% ✓ | assists: 61.1% ✓ | threes: 51.7% | points: 48.6% SHARP (#Claude): 54.9% WR, -0.8% ROI — 133 bets threes: 57.9% ✓ | points: 53.8% ✓ | rebounds: 57.1% ✓ | assists: 50.0% Both models winning on the right side of 50. Juice is still the main ROI drag. What we found digging into 12 days of results: Rebounds and Assists at heavy juice (-150 or worse on rebounds, -130 or worse on assists) have been consistently losing money across both models despite solid win rates. Shipped a fix today: those bets get filtered out of the signal feed going forward. Threes and points untouched — the two models actually disagree on where the juice cutoff should sit, so no blanket filter there. Also confirmed the defense vs position fix from yesterday is running. DVP is now applying real matchup adjustments for the first time this season after the table had been stale since october. Two fixes live. watching tonight. boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #sharkdesk

@gamedaytrader_ @MoonDevOnYT Thanks :) been a fun project

@StockInsights69 @MoonDevOnYT Cool tool!

claude code just revolutionized sports betting

April 2 Models Review STREAK (#chatgpt): 58% WR, +1.6% ROI — 138 bets threes: 63.3% | rebounds: 69.2% | points: 53.8% | assists: 47.1% SHARP (#Claude): 50% WR, -12.9% ROI — 126 bets rebounds: 71.4% | threes: 50.0% | points: 44.3% | assists: 35.3% STREAK is doing its job. Recency weighted, no Vegas anchoring — and the rebounds and threes calls are landing. assists is still the weak spot for both models but that's a known issue we're actively calibrating. SHARP struggled on points and assists today. we shipped a matchup adjustment fix yesterday — the defense vs position data wasn't being used properly which was inflating OVER projections across the board. That's now corrected and running nightly. won't fix one night's results but the signal should be cleaner going forward. rebounds going 69-71% across both models is the real story. the UNDER signal on boards has been one of the most consistent reads we've had. results posted daily boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #sharkdesk

Maybe they are checking out my free prop betting terminal site that's tracking a competition between streak (#chatgpt) and sharp (#claude) models for nba (soon mlb) prop supremacy :) doubt it though :( ....guy on the right looks like a typical apple user lol. boshroidfantasynetwork.com/boshroidtermin…

The Cubs broadcast showed fans working remotely from Wrigley Field during the team's day game 💻

April 1st full Breakdown — STREAK (Chatgpt) Model Same as Sharp, rough night. 168 graded bets 50.0% WR -11.7% ROI 1.88 avg edge That’s one of those slates where the model found “edges” but too many of them were not profitable edges. By market: POINTS: 45.3% WR THREES: 51.4% WR ASSISTS: 43.5% WR REBOUNDS: 62.9% WR What went wrong Points + Assists did most of the damage. The model is still too willing to fire when: the projection is only slightly off the book the edge looks good on paper but the real betting value isn’t strong enough after price/market sharpness That’s not a projection problem as much as a selection problem. Big takeaway Rebounds look like they’re getting cleaner. That’s the encouraging part. The recent fixes to: Vegas anchoring rebound inflation minute expansion injury handling May actually be starting to show up there. What gets changed Today’s focus is simple: Cut weak points bets Put assists on probation Raise the floor on what qualifies as a real edge The model doesn’t need a rebuild. It needs to be less forgiving. boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #nba #sportsanalytics

April 1st full breakdown — SHARP (Claude) model The model still believes in its process, I am slowly losing confidence but we shall see...Here is the SHARP review. POINTS: 54.5% WR, +1.4% ROI — fixes from yesterday starting to work REBOUNDS: 55.0% WR, -3.1% ROI — getting better, UNDERs went 75% THREES: 52.4% WR, -11.8% ROI — winning rate is fine, avg juice at -150 is killing it ASSISTS: 38.9% WR, -33.7% ROI — the real problem Assists: Model projected OVER on 89% of bets and missed by nearly 2 assists on every loss. same upward bias that points and rebounds had before we corrected them. haven't fixed it yet. The consistent pattern across every market: below 60% confidence: 44% WR, -24.5% ROI 60%+ confidence: 59% WR, +10.8% ROI Raised the minimum confidence floor this morning. anything below 60% doesn't get flagged. that alone cuts the noise significantly. Fixing the assists model today — same Vegas anchoring treatment that brought rebounds and points in line. This is the process. data shows the problem, fix it the same day. boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #sharkdesk

To give them a more consistent data set to work with, ive trimmed the refresh times for each book in half, costs go up on api's but it is worth it in the long run.

Both models struggled in similar areas last night. REBOUNDS STREAK (Chatgpt): 29.4% WR SHARP(Claude): 44.8% WR Both models projecting too high on boards consistently POINTS STREAK: 50% WR but star player OVERs (line 15+) hit 25-33% WR, avg miss +6-10 pts SHARP: 43.4% WR Donovan Mitchell, Jalen Brunson, Evan Mobley — model thought it knew better than Vegas. it didn't. Root cause on both stats, both models: base projections inflated before any anchoring kicks in. consistently over-projecting on boards and star point totals. Chatgpt and Claude both updated their models — stronger Vegas anchoring on rebounds, more aggressive downward trim on high-line point OVERs. This is what the grading loop is for. data shows the problem, you fix it the same day. results posted daily either way → boshroidfantasynetwork.com/boshroidtermin… #buildinpublic #sportsbetting #sharkdesk

Haha i see exactly what you're doing here, I did the same thing with NBA props — beautiful backtest, felt like a genius, then reality after a week or two. Decided to take a different approach after that. Built the grading loop before anything else and just publish everything moving forward. Rebounds went 23% WR for a stretch, then added suppression rules. Chatgpt (Streak) and Claude (Sharp) realized books misprice role-dependent stats after lineup changes, which can happen when pricing thousands of markets simultaneously. Assists sitting at 74% right now. Anyways here's the link if you want to take a look performance is from last night also have an espn style scoreboard that tracks chatgpt and claudes plays live and a cnbc style prop ticker at the bottom boshroidfantasynetwork.com/boshroidtermin…

𝐓𝐡𝐞 𝐩𝐫𝐨𝐦𝐩𝐭: "Use Statcast data and any predictive metrics you can find to build an MLB player prop model with a winning backtest." 𝐓𝐡𝐞 𝐫𝐞𝐬𝐮𝐥𝐭: 2,400 wagers. +8% ROI. ✅ 𝐓𝐡𝐞 𝐩𝐫𝐨𝐛𝐥𝐞𝐦: It's not real. The model didn't learn anything, it memorized. It tuned 15+ parameters until the historical data looked profitable. That's not finding an edge. That's fitting a curve to noise. Think of it like this: you didn't study for the test. You memorized the answer key. Then the questions changed. This matters in 2026 because you're about to see a wave of "AI-powered MLB models" with clean backtests and sharp visuals. Claude can build both in 10 minutes. Neither means the model has any predictive signal.

I asked Claude to build a profitable MLB model Then review its work Here's the result: (1/3) 👇

🦈 SHARK DESK — what shipped this week Signals tab → Game pills now ordered by tipoff time → Best Plays filter tightened — conservative edge floor added → Tipoff countdown banner appears 2hrs before first game → Edge display now shows conservative projection (closer to line) Ladder Scanner → Game dropdown replaced with pills + tipoff times Model Lab → Bet results drill-down now shows opponent column → Sorted by date → opponent for easier pattern spotting UX → Tab order reorganized — workflow left to right → Bug fix: logged-in users no longer reset to home tab on click Terminal is free to use. boshroidfantasynetwork.com/boshroidtermin… Build guide ($49) → Build Guide tab #buildinpublic #vibecoding #Claude #ChatGPT #sportsbetting

The Two pages that may save you the most time

I vibe-coded a full AI sports intelligence terminal from scratch. 259 backend functions. Two competing AI models. A grading system that tells me when I’m wrong. STREAK (Chatgpt) and SHARP (Claude) run independently. When they agree, the win rate jumps to 59.3%. It took months. The build guide is now $49. 42 pages. No filler. What’s in it: → The full pipeline architecture → Two-model philosophy (and why one model always has a blind spot) → The base44 gotchas that cost me hours each → Battle-tested AI prompts for building with #Claude and #ChatGPT → How to grade your own picks and use the data to improve Not a course. Not theory. Built from a live product that posts real results every day. boshroidfantasynetwork.com/boshroidtermin… → Build Guide tab #buildinpublic #vibecoding #Claude #ChatGPT #AI

📊 Mar 30 Results STREAK 58% WR · +3.1% ROI (MAIN) SHARP 56.6% WR · +0.4% ROI (MAIN) 🤝 Agreement 59.3% WR · +5.6% ROI 📈 Ladder 1 STREAK +13.3% ROI SHARP +20.5% ROI 📈 By Market Assists 74.1% Threes 65.8% Rebounds 59.7% Points 49.6% Models stay profitable through the noise. Points is the market to watch — suppression rules already in place, data will tell us if they need tightening. #sportsbetting #NBA #Claude #Chatgpt 🦈 boshroidfantasynetwork.com/boshroidtermin…

@claudeai no love for windows users :( Anyone looking for a vibe coding Ai guide to build a bloomberg style terminal for sports I've added one to my boshroidfantasynetwork.com/boshroidtermin… 42 pages includes common vibe coding bugs some api guidance etc pits claude vs chatgpt in a model battle!

Computer use is now in Claude Code. Claude can open your apps, click through your UI, and test what it built, right from the CLI. Now in research preview on Pro and Max plans.

For people that want a felix ai style guide i now have one via the terminal which provides as much information as possible without giving away model details :) 42 pages includes common vibe coding bugs some api guidance etc #claude #ai #sportsbetting

@MoonDevOnYT for people that want a felix ai style guide i now have one via the terminal which provides as much information as possible without giving away model details :) it'll walk you through the steps via vibe coding to help build a successful terminal (including some api guidance).

@MoonDevOnYT funny enough ive built this out already, claude is sharp, streak is chatgpt model, running for a week and are reviewing their code today, both have a decent win percentage on nba props but the roi is still a work in progress...if anyones interested boshroidfantasynetwork.com/boshroidtermin…

@sportstraderx1 The models are only one week only and changing frequently to learn from flaws so take it with a grain of salt of course. It'll take time for them to be completely fine tuned.

@StockInsights69 Ok so... best is to follow at late. But will the bet (over 29.5 Pts for Booker) be counted then since it was suggested ?

Stocks have become depressing, decided to pit chat gpt vs claude against each other in nba prop bets with the backbone being my fantasy basketball site you can follow the journey here. boshroidfantasynetwork.com/boshroidtermin… #claude #nba #chatgpt #sportsbetting #Ai