Cycles IQ
Build your own cycle universe. In minutes, not years.
An AI skill stack for cycle scanners, dashboards and chart-grade analytics. You describe the screen you want in a paragraph of English. The AI builds it on a cycle engine that took twenty-five years to get right.
The decade-long detour
From decades of research to an afternoon of building.
The cycle research does not get faster. It still takes decades. But once it exists and is encoded properly, the application layer collapses to an afternoon.
Detrending research · DSP filter design · Bartels significance
Cycle engine v1 · v2 · v3 · indicator library
UI design · charting integration · production app
1–2 hours
from a paragraph of English to a working screen on your own machine
Two mental models
An analyst in your files, and a briefing document for it.
Not a chatbot. An analyst.
A consumer AI gives you paragraphs. Useful, but it stops at the answer. An AI coding tool such as Claude Code lives inside your files: it reads code, writes code, runs code and decides what to do next. Real files, real apps, real data on your laptop. You stay in English; it produces software.
A briefing document.
You give one to your AI the way you would brief a new analyst on their first day: playbook, references, known pitfalls, example code. When the AI meets a problem that matches the skill, it reaches for the playbook automatically. No re-deriving. No guessing at parameter names. It already knows.
23 reference documents across the four skills of the Cycles stack.
The toolkit
The four-skill stack.
Each skill is independent and composable. You install only what you need.
20 REST endpoints. Cycle scanning, dominant-cycle detection, DSP filters (detrend, CRSI, Savitzky-Golay, sinc), KDE clustering, Cycle Consensus.
TradingView-grade browser charting. Custom datafeeds, custom indicators, lazy loading, verified end to end against the current library release.
The integration layer. Composite cycle reconstruction, forward projection, backward extension over lazy-loaded bars, CRSI parameter sync.
Design-system cards and starter templates for HTML, React and Next.js. Swap one CSS file to retarget for any brand.
The honest question
“Can't I just ask the AI to build this, without your skills?”
Yes. You can. Here is what happens.
The from-scratch trap
One in 1.3 billion.
Cycle analysis is seven decisions, not one. Each step has roughly twenty defensible methods. Seven steps times twenty variants is 207, about 1.3 billion possible wirings of the pipeline. The AI picks one. It will not tell you which, or why.
- 1Pre-processdetrend, endpoint flatten× 20 variants
- 2Scanspectrum / scan engine× 20 variants
- 3Rankqualify the peaks× 20 variants
- 4Extractpick the dominant× 20 variants
- 5Projectlength, phase, forward× 20 variants
- 6In-samplewindow selection× 20 variants
- 7MethodGoertzel · FFT · MESA · …× 20 variants
20 × 20 × 20 × 20 × 20 × 20 × 20
=
1.28 billion
possible wirings of the pipeline. The AI picks one, without telling you which, or why.
The skill is the encoded answer to which one survives walk-forward testing.
Decades, compressed
Same model. One has the playbook. One does not.
What lives inside the four skills: the math, the methodology, the pitfalls. None of it gets re-derived in a chat window.
- Bartels significance (1932)
- Savitzky-Golay (1964)
- Cyclic RSI
- Cycle Consensus
- 13 documented pitfalls
- TradingView-class charting
- Verified against the library release
- Datafeed contract
- 5 reference docs
- Composite reconstruction
- Forward projection
- Lazy-load extension
- 5 reference docs
- Editorial brand system
- 22 design-system cards
- 3 starter templates
Two builds, two prompts
Minutes, not years.
Both were built live at the Cycles Conference 2026 in New York from the prompts below. Real elapsed time for the first: about twelve minutes.
Build me a Cycle Consensus screening dashboard using the cycle-tools-api and webapp-design skills. Watchlist: SPX, NDX, DJI, RUT, GLD, SLV, BTC, ETH, TLT, DXY, CL, NG. Use the /api/CycleConsensus/score endpoint. Classify at ±60. Rank bearish on the left, bullish on the right. Run it locally on port 5175.
The AI reads the skill files, scaffolds a Vite and React app, writes the API layer and the table, installs dependencies, opens the browser: twelve symbols populate, the rank order forms. Then the live tweak: “Now add my 8-symbol currency watchlist and re-rank.”
Build me a TradingView-grade chart of SPX using FintaChart and the cycle-charting skill. Show price as candles. Overlay the dominant cycle as a sine wave on price. Project it 60 bars forward. Add CRSI in a lower pane with dynamic bands.
Custom datafeed with lazy loading, composite cycle reconstruction with a 60-bar forward projection, CRSI lower pane with dynamic bands. Chart renders, the sine projects to the right. Live tweak: “Switch the symbol to EURUSD and re-detect the dominant cycle.”
Already in production
Two public systems built entirely on the stack.
A five-layer macro regime composite
Thirty-one public economic series, pulled in parallel from FRED and central banks, condensed into one 0–100 regime score across five layers: component cycles, structural cycle, liquidity, momentum, regime. Open source under the MIT licence.
The reference cycle chart
The polished production version of the second build: candles, composite cycle sine projected forward over lazy-loaded bars, CRSI in a lower pane. The canonical example shipped with the cycle-charting skill.
How to get it
Cycles IQ is the door.
There is no public release of the skill stack. Members of the Cycles IQ program receive the private skill repository, full API access, the MCP server for Claude, ChatGPT and Cursor, the advanced cycle analyzer application with extended tools and screens, pre-built applications and dashboards as starting points for their own screens, and an all-access pass to Lars's tools with feature updates, maintenance and support. Plus an individual onboarding session with Lars. It starts with a short conversation.
- Get in touch. One e-mail with who you are and what you want to build.
- We talk. A 30-minute call, one to one, no marketing drip.
- Provision. Private repository invite, API key, MCP and analyzer access, documented install for your AI coding tool.
- Onboard. A 2-hour session on your requirements and quick-start setup, with a written recap. The first dashboard you ship gets reviewed if you want.
For forty years, this kind of analysis was the privilege of people with the right experience, the right network and the right tools. Tomorrow morning, you have the tools.Lars von Thienen, Cycles Conference 2026, New York