Is BTCBuzz financial or investment advice?

No. BTCBuzz is a research and educational project built by Cloud First Consulting to study how machine learning, validation gates and AI agents behave on a hard, noisy problem. It does not give financial, investment or trading advice. All of its trading is simulated paper trading with no real money, every figure in this article comes from simulation or from software measurements, and nothing here is a recommendation to buy, sell or hold Bitcoin or any other asset. Cloud First Consulting is not authorised or regulated by the Financial Conduct Authority to give investment advice.

BTCBuzz is a Bitcoin signal bot written in Python. Once a day, after the 00:00 UTC close, it collects price, derivatives, on-chain, macro and news data, runs an ensemble of RandomForest, XGBoost, LightGBM and a PyTorch LSTM, and writes a BUY, SELL or HOLD report with its reasoning. It paper trades its own signals and posts the report to a Telegram group. A background research loop proposes changes to the strategy, and a set of validation gates rejects the ones that only look good in a backtest.

What is Jev from TypeSafe.ai?

Jev is a model from TypeSafe.ai that the company calls a System One model. You send it a piece of text and a set of typed questions, and it returns structured answers: a probability for a yes or no question, a probability for each option in a multiple choice question, or a probability for each level of an ordered scale. It does not write prose. At the time of writing the listed price was $0.042 per million input tokens with no charge for output, and the model version we pinned was jev-1.13.0.

Is Jev a large language model like ChatGPT or Claude?

Jev works differently from a chat model. A chat model writes text, and your code then has to read that text and decide what it means. Jev writes no prose. It returns a probability for each typed question you ask, so the output is already a number or a label your code can compare with a threshold. TypeSafe lists poor text generation among the known limits of jev-1.13, which fits its purpose: fast, narrow decisions over text, with the reasoning and arithmetic left to your own software.

How do you call the Jev API?

Send an HTTP POST to https://api.typesafe.ai/v1/systemone with an Authorization bearer key. The JSON body has three fields: model (we pin jev-1.13.0), state (the text to read) and questions (a map of question ids to a type of noul, choice or score, plus instructions and, for choice and score, the options or levels). The response holds one typed answer per question id and a usage block with the input token count. TypeSafe publishes Python and JavaScript SDKs. BTCBuzz uses plain HTTP so that it needs no new dependency.

How much does Jev cost to run in BTCBuzz?

Each headline costs about 1,120 input tokens for our set of 11 questions. At roughly 50 headlines a day that is about $0.0024 a day, or less than $1 a year. Scoring our whole 28-day archive of 1,508 headlines took 1,451 API calls and 1,632,955 tokens and cost $0.0686, with no failed calls.

TypeSafe's own documentation lists the limits for jev-1.13. It does not count reliably, it is poor at arithmetic, it reads dates as text so it cannot compare them or work out durations, it gets less accurate when the input contains a lot of irrelevant text, and it struggles with double negatives and multi-step reasoning. We designed around that. Jev never sees a price, an indicator or the trade ledger. It only reads short pieces of text, and every count, average and threshold is computed in our own code.

Does Jev make BTCBuzz more accurate?

We do not know yet, and we have not claimed it does. Across 1,499 archived headlines Jev and our keyword scorer flagged almost the same number of adverse events (282 and 273) but agreed on only 99 of them. Nobody has yet judged which reading was right. We wrote the pass and fail criteria before running anything: 150 headlines labelled blind by a person, then per-class F1 for both scorers with confidence intervals. Until that is done, Jev runs in shadow mode and cannot change a signal.

What does shadow mode mean?

The new component runs on live data next to the old one, and its output is written to a log beside the old output, but only the old output is used. In BTCBuzz nothing Jev returns can change a signal, a position size or a published number. Each behaviour that could change something, such as dropping a suspicious headline or withholding a research experiment, has its own switch that defaults to logging only.

Why not judge Jev by whether the bot makes more money?

Because the sample cannot support it. The bot makes a few dozen simulated trades a quarter, and each trade is held for up to 30 days, so the outcomes overlap heavily. A small improvement in how headlines are read would take a year or more to separate from noise. If we used simulated returns as the test we would end up crediting or blaming Jev for market movements. We judge it on whether it reads the news correctly, which we can measure in a day.

Can a small business use a model like Jev?

Yes, for narrow decisions over text where the answer is one of a fixed set: routing a support ticket, deciding whether an email is a complaint, checking whether a supplier message mentions a delay, or screening text before it reaches a larger AI model. The pattern that carries over from BTCBuzz is to keep the questions small and literal, do all counting and thresholds in your own code, run the new model beside the old rule before trusting it, and write down what a pass looks like before you see any results.

Jev by TypeSafe.ai, Tested: A $1-a-Year AI Model Inside a Bitcoin Signal Bot

Jev is the first System One model from TypeSafe.ai. You send it a piece of text and a set of typed questions, yes or no, multiple choice or an ordered scale, and it returns a probability for each one instead of prose, at a listed $0.042 per million input tokens with no charge for output. We pinned jev-1.13.0 and put it inside BTCBuzz, an educational Bitcoin signal bot built by Cloud First Consulting to study machine learning, validation gates and AI agents. BTCBuzz paper trades only, with no real money, and nothing it or this article publishes is financial or investment advice. The bot is about 24,900 lines of Python with 481 automated tests on a two-core server, and once a day it collects price, derivatives, on-chain, macro and news data, runs a RandomForest, XGBoost, LightGBM and PyTorch LSTM ensemble, and posts a BUY, SELL or HOLD report to Telegram. Three of its rules read news by counting keywords, which let "Dropbox Security Breach" score as a crypto security event and let a returned hack count as a fresh one. Jev now answers 11 literal questions about each headline in shadow mode, so it cannot change a signal. Across 1,499 archived headlines the keyword scorer flagged 273 adverse events and Jev flagged 282, with 99 in common, and Jev judged 106 of the keyword scorer's flags not material to crypto. The article reports the API call shape, measured costs and latencies (about $0.0024 a day, 1,451 calls and $0.0686 to score the archive, median latency 711 ms), the published limits of jev-1.13 on counting, arithmetic and dates, four bugs a human review caught after the tests passed, three readings where we think Jev or our own code is wrong, four more jobs for the same model including prompt injection screening, and the pass and fail criteria written before the first API call: 150 headlines labelled blind, per-class F1 with confidence intervals, four weeks of judged disagreements, and operating limits on failures, latency and cost.

Frequently Asked Questions

Is BTCBuzz financial or investment advice?: No. BTCBuzz is a research and educational project built by Cloud First Consulting to study how machine learning, validation gates and AI agents behave on a hard, noisy problem. It does not give financial, investment or trading advice. All of its trading is simulated paper trading with no real money, every figure in this article comes from simulation or from software measurements, and nothing here is a recommendation to buy, sell or hold Bitcoin or any other asset. Cloud First Consulting is not authorised or regulated by the Financial Conduct Authority to give investment advice.
What is BTCBuzz?: BTCBuzz is a Bitcoin signal bot written in Python. Once a day, after the 00:00 UTC close, it collects price, derivatives, on-chain, macro and news data, runs an ensemble of RandomForest, XGBoost, LightGBM and a PyTorch LSTM, and writes a BUY, SELL or HOLD report with its reasoning. It paper trades its own signals and posts the report to a Telegram group. A background research loop proposes changes to the strategy, and a set of validation gates rejects the ones that only look good in a backtest.
What is Jev from TypeSafe.ai?: Jev is a model from TypeSafe.ai that the company calls a System One model. You send it a piece of text and a set of typed questions, and it returns structured answers: a probability for a yes or no question, a probability for each option in a multiple choice question, or a probability for each level of an ordered scale. It does not write prose. At the time of writing the listed price was $0.042 per million input tokens with no charge for output, and the model version we pinned was jev-1.13.0.
Is Jev a large language model like ChatGPT or Claude?: Jev works differently from a chat model. A chat model writes text, and your code then has to read that text and decide what it means. Jev writes no prose. It returns a probability for each typed question you ask, so the output is already a number or a label your code can compare with a threshold. TypeSafe lists poor text generation among the known limits of jev-1.13, which fits its purpose: fast, narrow decisions over text, with the reasoning and arithmetic left to your own software.
How do you call the Jev API?: Send an HTTP POST to https://api.typesafe.ai/v1/systemone with an Authorization bearer key. The JSON body has three fields: model (we pin jev-1.13.0), state (the text to read) and questions (a map of question ids to a type of noul, choice or score, plus instructions and, for choice and score, the options or levels). The response holds one typed answer per question id and a usage block with the input token count. TypeSafe publishes Python and JavaScript SDKs. BTCBuzz uses plain HTTP so that it needs no new dependency.
How much does Jev cost to run in BTCBuzz?: Each headline costs about 1,120 input tokens for our set of 11 questions. At roughly 50 headlines a day that is about $0.0024 a day, or less than $1 a year. Scoring our whole 28-day archive of 1,508 headlines took 1,451 API calls and 1,632,955 tokens and cost $0.0686, with no failed calls.
What can Jev not do?: TypeSafe's own documentation lists the limits for jev-1.13. It does not count reliably, it is poor at arithmetic, it reads dates as text so it cannot compare them or work out durations, it gets less accurate when the input contains a lot of irrelevant text, and it struggles with double negatives and multi-step reasoning. We designed around that. Jev never sees a price, an indicator or the trade ledger. It only reads short pieces of text, and every count, average and threshold is computed in our own code.
Does Jev make BTCBuzz more accurate?: We do not know yet, and we have not claimed it does. Across 1,499 archived headlines Jev and our keyword scorer flagged almost the same number of adverse events (282 and 273) but agreed on only 99 of them. Nobody has yet judged which reading was right. We wrote the pass and fail criteria before running anything: 150 headlines labelled blind by a person, then per-class F1 for both scorers with confidence intervals. Until that is done, Jev runs in shadow mode and cannot change a signal.
What does shadow mode mean?: The new component runs on live data next to the old one, and its output is written to a log beside the old output, but only the old output is used. In BTCBuzz nothing Jev returns can change a signal, a position size or a published number. Each behaviour that could change something, such as dropping a suspicious headline or withholding a research experiment, has its own switch that defaults to logging only.
Why not judge Jev by whether the bot makes more money?: Because the sample cannot support it. The bot makes a few dozen simulated trades a quarter, and each trade is held for up to 30 days, so the outcomes overlap heavily. A small improvement in how headlines are read would take a year or more to separate from noise. If we used simulated returns as the test we would end up crediting or blaming Jev for market movements. We judge it on whether it reads the news correctly, which we can measure in a day.
Can a small business use a model like Jev?: Yes, for narrow decisions over text where the answer is one of a fixed set: routing a support ticket, deciding whether an email is a complaint, checking whether a supplier message mentions a delay, or screening text before it reaches a larger AI model. The pattern that carries over from BTCBuzz is to keep the questions small and literal, do all counting and thresholds in your own code, run the new model beside the old rule before trusting it, and write down what a pass looks like before you see any results.

Our Services

Contact Cloud First Consulting

Email: info@cloudfirstconsulting.com

Location: London, United Kingdom

Hours: Monday-Friday, 9:00 AM - 6:00 PM GMT

Book a Free 30-min Discovery Call