Betta Protocol

CA: Er899oJCB7pKAf9rPwrqkEeLwfQQz5YYVU6LoNYTpump x.com/i/broadcasts/1…

Betta Protocol - Today's Updates 🐟 New System: Activity Logger: - Logs every single fish detection with full context (position, motion, confidence, timestamp) - Stores 10,000+ entries with behavioral data - Provides hourly/daily activity summaries - Peak activity tracking and behavior pattern analysis - In-memory caching for fast queries Enhanced Daily Reports: - Rich summaries from activity data: detection counts, peak activity hour, active percentage - Behavioral insights (high energy 💪 / chill vibes 😌 / explorer mode 🗺️) - Twitter-formatted output (<280 chars) - Replaces old basic detection counts with detailed analytics Claude Reporter Upgrades: - Now receives 60 minutes of behavioral context (not just current moment) - Updated prompts to use YOLO detection box as visual guide for accuracy - Handles hiding fish gracefully (acknowledges when out of sight) - More precise location descriptions (left/center/right, top/middle/bottom) Main.py Integration: - Activity logger runs alongside existing systems - Logs every detection automatically - Passes rich context to Claude every 5 minutes - Daily rollover triggers summary generation Bug Fixes: - Activity logger: Fixed file initialization issues with missing 'entries' key - Activity logger: Added error handling for corrupted JSON files - Claude vision: Fish was being mislocated (said "center-bottom" when actually left side) - Claude vision: Now explicitly uses YOLO bounding box for accurate positioning

Claude, Organisms, and Memecoins: Signal vs Noise A new trend has emerged on pump.fun: memecoins paired with Claude AI and a living organism, funded by token communities rather than venture capital. The idea is compelling. The execution across most projects is not. Out of 100+ "by Claude" projects, only 4 are real. Here's how I verified them. Where It Started: The origin is @TrophyTomatoSol - a tomato plant kept alive by Claude AI making real-time decisions about water, light, and nutrients. Day 57 and counting. The system is connected to physical sensors and actuators. The plant lives or dies based on AI decisions. Built by @d33v33d0 after months of infrastructure development before any token existed, the project centers on a single question: Can AI sustain life? The critical detail most imitators miss: the system existed before the coin. Why This Matters: What made Sol important wasn't just the technology - it was the funding model. VCs fund predictable outcomes and proven markets. Weird experiments? Questions nobody asked? They won't touch it. Instead of VC capital with strings attached, Sol's trading fees funded better equipment and increased scientific rigor. The project recently onboarded a molecular biologist, compensated via a percentage of fees, to formalize methodology. This model funds curiosity, not predictability. It turns genuine interest into sustainable funding for projects traditional finance won't support. As Martin put it: "VCs want returns. This model wants answers." The Problem: The model was immediately copied - superficially. Search "by Claude" on pump.fun: 11 pages of results. "Chicken by Claude", "Banana by Claude", "Driving by Claude" - plants, animals, and activities all claiming AI involvement. After verification, the majority are either misleading or outright scams: fake livestreams, generic AI text detached from footage, vague claims, and no verifiable integration. They borrow Sol's credibility while delivering nothing but exit liquidity. What's Actually Real: After filtering for verifiable, continuously operating systems, only four projects show evidence of genuine experimentation: 1. Sol The Trophy Tomato (@TrophyTomatoSol by @d33v33d0) - AI-managed plant life with real sensors and actuators 2. Claude Fish Market Maker (@ClaudeFishMM) - AI trading informed by live fish behavior 3. Edward by Claude (@Edwardbyclaude by @imandrei_fyi) - AI-assisted behavioral monitoring of a Syrian hamster named Edward 4. Betta Protocol (@BettaProtocol) - Betta fish observation using custom YOLO computer vision and AI-generated observational analysis Four real projects across eleven pages of listings. How to Spot the Difference: Red flags: - No continuous livestream, or an obvious looped feed - No consistent iterations by developers - Generic observations anyone could write - "Coming soon" features instead of live systems - Anonymous teams with no technical footprint Green flags: - Verifiable 24/7 uptime - Specific tooling disclosed (YOLO, OpenCV, named APIs) - Vision-based observations tied to live footage - Visible hardware integration - Public builders, open iteration, documented failures The Reality Check: Out of 100+ "by Claude" projects, only four demonstrate genuine technical integration. The rest are fake streams, and cash-grab imitations. This isn't about ambition or early-stage constraints. It's about falsifiability. Real experiments accept the risk of being wrong. Most failed projects are designed to avoid that risk entirely. The Sol model proved collective funding can support meaningful research. It also attracts grifters at a 25:1 ratio. The mechanism that enables genuine inquiry also enables exit scams at scale. The Choice: This trend can fund serious questions about AI and life; questions traditional institutions won't support. Or it can devolve into another memecoin graveyard littered with rugpulls. The difference isn't technology. It's standards. Builders must prove legitimacy. Communities must demand proof. Verify the tech. Demand live operation. Require the possibility of failure. Don't fund performance masquerading as research. Be safe. DYOR. Don't get rugged❤️

Implemented OMLE.v3! By using the @ultralytics hub, we refined our existing OMLE.v2 with new references created from our @Pumpfun vods. With OMLE.v3, we get detection that is more consistent, provide more accurate data, leading to better daily summaries for @AnthropicAI 's Claude Haiku to generate for our community.

@GetBiffed Will be adding a snail very soon actually! This helps with the leftover food that Elmo doesn’t eat :)

@BettaProtocol can you add a crab?

Live January 20 x.com/i/broadcasts/1…

TEMPERATURE SENSOR IS NOW LIVE!!!! Should I also make a GitHub and share all the files I’m using to show the gearworks of this stream?

Live January 20 x.com/i/broadcasts/1…

This alone took me 30 minutes 😭 I’m having fun but why is crimping hard

We're here to announce that we will be doing our best to participate in the first Pump Fund Build in Public Hackathon by @Pumpfun! The timing honestly couldn't be better. We started Betta Protocol a few days ago and have been constantly iterating, migrated from a single script to a modular 12-file system, implemented YOLO tracking to cut down false positives, set up MJPEG streaming for OBS, just added a night vision camera so we can track Elmo even in the dark, and we're working on Arduino support for more hardware control. We've been building in public from day one. Daily AI summaries, livestream running, motion tracking, assisted feeding schedules all while Claude provides commentary on what our betta is up to. This hackathon is basically validating what we're already doing: ship fast, iterate openly, let people see the process. Join us on the $BETTA journey!

@Pumpfun What if the dev has locked 5%? Would that count?

Eligibility requirements - Launch a token - Build a project - Own at least 10% of the token’s supply - Build in Public: Posting on X, forming a Community & Streaming on Pump fun are HIGHLY encouraged -Apply via the form: forms.gle/UwuNvov75Tx1m2…

Introducing the $3,000,000 Build in Public Hackathon Brought to you by Pump Fund - pump fun’s New Investment Arm It’s time to completely reimagine how early-stage projects are built and funded. Learn more 👇

A little behind the scenes when I boot up the Betta Protocol system

Below is an example of a post that Claude has generated. These reports were originally scripted to post at exactly 10PM PST, however there are some obstacles ($200/month), so for now will be manually posting these generated reports. Day 3: The Still Day • Activity Level: Significantly reduced—zero detections recorded (100% below baseline) • Inactivity Window: Single 120-minute period of no movement; peak observation window at • Behavioral Pattern: Minimal to absent throughout 24-hour cycle Observation: After two days of normal activity, our subject has entered a notably quiet phase. The complete absence of detections is striking, though the single extended rest period rather than scattered inactivity suggests this may be consolidated resting behavior rather than systemic concern. Given the species' known capacity for extended periods of stillness—particularly in the evening hours when we recorded peak inactivity—this warrants observation but not alarm. I'll be watching closely over the next 48 hours to see whether activity rebounds toward baseline or if we're seeing a genuine shift in behavioral patterns.

On the second day, we: - upgraded to a camera with supported night vision - implemented OMLE.v2 (better fish recognition) - added non-copyright music to the stream! We keep building!