A canonical, machine-readable reference of every Inner Circle Trader (ICT) concept, strategy, and named model published between 2016 and 2026 — designed as a persistent, LLM-maintainable wiki rather than a one-shot document.
226 concept files across 33 directories. Each concept lives in its own self-contained markdown file with a stable name, definition, formal criteria, formula, visual, examples, common mistakes, related-concepts cross-links, and source citations. Every file ships with a machine-readable JSON block so the library can be consumed by RAG / agent systems deterministically.
This library implements Andrej Karpathy's LLM Wiki pattern: a structured, interlinked collection of markdown files maintained by an LLM agent rather than re-derived on every query. The three layers Karpathy describes:
- Raw sources — ICT mentorship videos, X threads, public commentary. Cited by stable Source IDs in
SOURCES.md. - The wiki — the 226 concept files in
concepts/. LLM-maintained. - The schema —
AGENTS.md, which tells the LLM how to ingest new sources, answer queries, and lint the wiki for this domain.
Two index files navigate the wiki:
INDEX.md— content-oriented catalog, organized by concept directory.log.md— chronological event log of activity on the wiki.
To work on this wiki with an LLM, hand it AGENTS.md — that's the operating manual.
concepts/
01-market-structure/ 12 files swing points, BOS, CHoCH, MSS
02-liquidity/ 14 files BSL, SSL, EQH/EQL, sweeps, IRL/ERL, DOL
03-order-flow/ 6 files institutional flow, APD, smart-money footprint
04-time-cycles/ 10 files 90-min cycle, macros, QT, DST, TOD pivots
05-pd-arrays/ 9 files premium/discount arrays, hierarchy, nesting
06-fair-value-gaps/ 14 files FVG, IFVG, BPR, CE, immediate/delayed, nested
07-order-blocks/ 10 files bullish/bearish OB, breakers, propulsion
08-breaker-blocks/ 6 files breaker, mitigation block, failed breaker
09-displacement/ 6 files displacement criteria, gap classification
10-killzones/ 8 files five canonical killzones in NY time
11-silver-bullet/ 7 files 3 SB windows, rules, 2025 refinement
12-power-of-three/ 6 files PO3 / AMD doctrine
13-judas-swing/ 4 files judas swing variants + failure
14-asian-range/ 6 files Asian range, sweep, projections
15-sessions/ 9 files Asia, London, NY AM/PM, London Close
16-smt-divergence/ 5 files SMT, correlated pairs, indices
17-optimal-trade-entry/ 6 files OTE 0.62 / 0.705 / 0.79
18-mitigation/ 5 files mitigation states across structure types
19-rejection-blocks/ 3 files wick rejection blocks
20-turtle-soup/ 4 files turtle soup, stop hunts
21-crt/ 4 files Candle Range Theory (community-attributed)
22-quarterly-theory/ 9 files QT fractal: yearly → 90-min
23-ipda/ 6 files IPDA + 20/40/60-day lookbacks
24-amd-cycle/ 4 files AMD cycle as time construct
25-htf-bias/ 7 files monthly/weekly/daily bias + framework
26-imbalance/ 5 files imbalance / inefficiency umbrella
27-equilibrium/ 4 files equilibrium, mean threshold
28-fibonacci-levels/ 7 files ICT fib set + SD projections
29-stop-runs/ 4 files stop run + entry variants
30-news-driven/ 5 files FOMC two-stage, NFP, CPI, blackout
31-models/ 14 files ICT 2022/2023/2024 + Unicorn + B&B + Diamond + Venom + Zircon + NDOG/NWOG
32-risk-management/ 7 files R-multiple, sizing, structural SL, partials
99-glossary/ — deferred (single-page GLOSSARY.md supersedes)
Era coverage:
- 2016: foundational mentorship (55 files)
- 2017: charter year refinements (74 files)
- 2018: block taxonomy + IFVG + IPDA (39 files)
- 2021–2024: models era (52 files combined)
- 2025–2026: Venom, Zircon (demo-stage), 2-stage FOMC, static drawdown (6 files)
This is a definitional library. It does not contain:
- Trading algorithms or backtests
- Live market data feeds
- Broker integrations
- Performance results
- Personal opinions on what "works"
Those belong in separate trading projects. The library's value is precision and consistency — describe the framework neutrally, let the reader form their own conclusions.
Every concept file follows TEMPLATE.md: 7 required top-matter fields (Category, Aliases, ICT Confidence, Year Introduced, Year Refined, Source IDs, Tags) and 10 required sections (## Definition through ## Citations), plus a machine-readable JSON block whose id matches the filename.
ICT Confidence is one of: high (ICT-original, well-documented), medium (limited public sourcing), community-attributed (e.g. CRT by Romeo), disputed (contested origin), demo-stage (e.g. Zircon Jan 2026 silent demo).
See CONTRIBUTING.md for the hard rules.
Files: kebab-case.md (e.g. bullish-order-block.md, ny-am-killzone.md). One concept per file — no exceptions. If something feels like "two concepts in one file," split it.
As a human reader: start with READING-ORDER.md — three learning tracks (Beginner / Intermediate / Advanced). Or jump from INDEX.md to whichever concept you need.
As an LLM agent: read AGENTS.md for the schema and ingest/query/lint workflow. Read INDEX.md first when answering queries; drill into specific concept files; cite by filename. Use SOURCES.md IDs when adding new content.
For RAG / programmatic consumption: every concept file has a ## Machine-Readable JSON block with id, category, aliases, criteria, timeframes, confidence, year_introduced, year_refined, related[], sources[]. Parse the JSON for deterministic queries; render the markdown for human-friendly output.
Content-complete and audit-passed. 226 files, 33 directories, all validation checks pass:
- 10/10 required sections per file
- Every JSON block parses; every
idmatches filename - Zero broken cross-links
- Zero orphan source citations (43 stable IDs in
SOURCES.md) - INDEX.md and TIMELINE.md fully aligned with disk
See CHANGELOG.md for the phase-by-phase build log and log.md for the chronological event log.
(none yet — TBD)