Free preview · Chapter 2
What you are building, and the words you'll need.
Two full sections from the beginner path of the Kronos Field Guide, free to read. Launching soon.
No support of any kind is provided.
Educational only; not financial advice. Start with pretend money. You can lose money even when every rule works as intended. Paper results do not promise live results.
1. What you are building
Imagine three separate jobs. One part watches past market prices and draws possible future prices: that is Kronos. A second part follows rules about when to buy, hold, reduce or close a position: that is the Prometheus-style strategy and risk layer. A third part sends orders to a broker and checks what actually filled. A coding agent can help build these parts, but a confident explanation from an agent does not prove that the software works or that its strategy makes money.
A forecast is a guess about a future price. An order is a request to buy or sell. A fill is a completed purchase or sale. Those are different events. The most damaging bugs can happen between an order being requested and a fill being confirmed.
The first version should only observe. The second should replay saved prices. The third may submit orders to a broker's paper account after you explicitly choose and accept your settings. This package does not switch you to a real-money account.
2. The words people use
| Term | Plain meaning | Why it matters |
|---|---|---|
| Asset / symbol | The thing being observed, such as a stock or cryptocurrency | The same asset can have different symbol names at different providers |
| Position | How much of an asset you currently own | Several purchases can grow one position beyond your intended limit |
| Long | Owning an asset and benefiting if its price rises | Selling that asset closes or reduces your long |
| Short | Borrowing an asset to sell first and hoping to buy it back cheaper | This is not the same thing as a stop-loss, and is disabled in the template |
| Equity | Cash plus the marked value of positions, minus liabilities | It moves even when nothing is sold |
| Realized loss | Loss locked in through a completed sale | Different from a temporary mark on an open position |
| Unrealized loss | Current marked loss on something still held | Still economically important; may worsen before you can sell |
| Exposure | How much of the account is invested in a category | Multiple coins can all fall together |
| Drawdown | Decline from an earlier account peak | Recovering the same percentage does not restore the account |
| Candle / bar | A period's opening, highest, lowest and closing price and volume | An hourly bar and a daily bar describe very different horizons |
| OHLCV | Open, high, low, close, volume | The market-history inputs used by the forecasting system |
| Spread | Gap between the best quoted buying and selling prices | Crossing it creates an execution cost |
| Slippage | Difference between expected and actual execution | It can get much worse when markets move fast |
| Liquidity | How easily an order can trade without moving the price much | A price on a screen does not guarantee a buyer for your quantity |
| Rebalance | Adjust holdings toward chosen target proportions | Doing it too often can turn a quiet strategy into a fee machine |
| Calibration | Comparing forecast uncertainty with actual outcomes | Model agreement with itself is not evidence it is correct |
The full guide continues with 81 configuration fields, the incident record and both coding-agent briefs.
Read the full offer and license →From the guide's resources chapter
Choose a deployment path
Hardware figures below are planning allowances, not measured Kronos minimums or throughput guarantees. Your model, history length, samples, assets, dependencies and concurrent processes determine actual requirements. A bigger computer does not establish a better investment strategy.
| Path | What it is for | Planning allowance | What you must test |
|---|---|---|---|
| Your existing Mac or PC | Reading, replay and a small Kronos-mini observation experiment | Start by checking available memory; 8 GB total RAM may be constrained, 16 GB gives more working room. Allow 10–20 GB free disk for an initial Python/model workspace. | Load one model, forecast one symbol, then measure peak RAM and cycle time before increasing workload. |
| Small Linux VM for orchestration | Journals, scheduler and broker-paper integration; inference can remain on another machine | Illustrative 2 vCPU / 4 GiB RAM / 40+ GB disk allocation; not a validated all-in-one inference specification | Verify memory with your dependencies; ensure network failures between inference and execution fail safely. |
| Linux VM for a small all-in-one research setup | One model worker plus local journals and a small paper universe | Illustrative 4 vCPU / 8 GiB RAM / 80+ GB disk allocation; benchmark first | Confirm worst-case inference finishes before the next cycle without starving the independent protection worker. |
| Larger model / many assets / large sample batches | Deliberately expanded research | Size from measured CPU/RAM/VRAM demand; a GPU is optional and workload-dependent | Compare throughput and model quality separately. Measure cost per completed forecast, not only cost per machine-hour. |
These are starting budgets, not purchasing recommendations or performance claims. No hardware is required just to read this guide.
Gear we run this on
The guide's always-on path describes a small always-on Mac or Raspberry Pi, an external SSD for the workspace, and a UPS so a scheduled stop can still fire during a power blip.
See our gear on Amazon ↗As an Amazon Influencer, ForgeMesh earns from qualifying purchases. Your price does not change.
Optional hosting: DigitalOcean
Compare DigitalOcean plans — affiliate link
Open DigitalOcean pricing directly — no affiliate referral
GSD Contracts LLC / ForgeMesh may earn a commission if you purchase through the affiliate link. A VM is optional; benchmark the computer you already own first.
Want the fundamentals first? Our free self-hosting course covers the VM and networking basics.
The full guide is $19 at launch price through September 15, 2026, then $29. Read the full sales page and license, or go straight to checkout. No support of any kind is provided.
Not ready to buy? One email when the launch price ends and when the edition updates.