fintech-algorithms

fintech-algorithms

717 pure functions for market, financial and statistical calculations. Plain arrays and objects in, plain values out. Zero runtime dependencies, Node >= 22, ESM.

Docs: https://docs.thefintechbuilder.com · Authoritative agent guide: https://docs.thefintechbuilder.com/guides/ai-agents/

Non-negotiables

Four rules. Breaking any one produces output that looks right and is wrong.

1. Never invent an import path, a function name, or a parameter. Every subpath mirrors its docs URL exactly, which makes a plausible guess wrong in a way that reads as correct. Look it up — scripts/lookup.mjs or the resolution order below. If the topic does not exist, say so and stop.
2. Never guess a returned field name. Return-key casing is not consistent across the library: bollingerBands returns percent_b, macd rows return fastEma. Read the captured example output for that topic. See references/pitfalls.md.
3. State the verification tier on any numeric claim. verified (601 topics) means the arithmetic is replayed and asserted on every build against expected values the catalog computed with a Python implementation written alongside the TypeScript — cross-language parity, not an independent third-party figure. Say it that way if asked. contract (74 topics) means the signature and shape are checked but nothing asserts the numbers.
4. Analysis, not advice. These functions compute quantities. An indicator crossing is an observation about a series — not a prediction, not a signal, and never a recommendation for a specific person's money. Report what was computed, on what input, at which tier. If asked what to buy or sell, say that is a question for a licensed adviser.

The library does not fetch data

There is no HTTP client, no vendor SDK, no API key, no node:fs. If a task needs prices, the caller supplies them. This is deliberate: vendor APIs get rewritten every few years and algorithms do not.