# The Game Plan — the whole library, one system

*2026-08-11. The user's instruction: find all the methods across the full
D:\downloads library; find how the pieces connect — many will be measuring the
same thing in different ways; use Cognee for discovery and Neo4j as the record;
make it one system.*

## What the library actually is

Surveyed 2026-08-11 from the device listing. Roughly 490 files, ~2.7 GB, in the
eight named folders plus what was already ingested:

| folder | files | what it holds |
|---|---:|---|
| gann | 94 | Gann's own courses (Master Stock Market, Commodities, Tunnel) + the interpreters: Mikula, Ferrera, Gilmore, Jeanne Long, Pesavento, Jacobs, Hyerczyk |
| george bayer | 3 | **Bayer's own books at last**: Time Factors, Handbook on Trend Determination, Turning Astrology to Practical Use |
| june30th | 10 | rare originals: 1955 Master Calculator, 1953 Mathematical Formula, 1950 Tops & Bottoms, 1954 May Coffee worked chart, Cowan's Rare Writings of W.D. Gann |
| may0526 | 101 | Unpublished Forecasting Courses, How to Make Profits in Commodities, Natural Squares Calculator, Cowan 4D **full Volume I**, Truth of the Stock Tape + a large esoteric shelf (Walter Russell, Oahspe, sacred geometry) |
| michael jenkins | 11 | the complete Jenkins library: Geometry of Stock Market Profits 1-3, Secret Science, Secret Angle Method, Complete Trading & Forecasting, Chart Reading |
| sepherial | 24 | **his actual market books**: The Arcana / Stock and Share Key, Law of Values, Key to Sugar Values, On the Markets, Silver Key, Kabala of Numbers, Eclipses, Cosmic Symbolism |
| numerology books | 244 | a handful of real sources (Luo Clement 1908, Sefer Yetzirah study 1914, Pythagorean Triangle 1875, Curtiss) buried in modern pop numerology |
| others - stocks | 5 | the full Helmholtz, McClain's Myths of Invariance, two DSP textbooks |
| (also present) | ~1000 | astrology books, horsedata — out of scope until asked |

## The phases

**Phase 1 — DATA (the gate on everything).** Yahoo via yfinance on the user's
machine: S&P 500 daily from 1927, futures for the grains Gann and Cowan
actually worked, long stocks from 1962. `tools/fetch_deep_history_yahoo.py`.
Without a century of bars, no long-cycle method can even be scored; with it,
every measured number in this repo gets re-run.

**Phase 2 — INGEST in waves, by method density, never by folder order.**
- Wave 1 (staged 2026-08-11, 41 files): Bayer's three books; all ten june30th
  originals; Sepharial's ten market/number books; the eleven Jenkins books;
  Cowan's full 4D Volume I. Corpus layout: `corpus/library/{author}/{book}/`
  page files + sidecars, same hashing discipline as the Baumring corpus.
- Wave 2: Gann originals from gann/ + may0526 (Unpublished Courses, How to Make
  Profits, Natural Squares Calculator) and the serious interpreters (Mikula,
  Ferrera, Gilmore) — interpreters mined only for METHOD statements, tiered as
  secondary; they never outrank a primary page.
- Wave 3: others-stocks (full Helmholtz, McClain) + the real numerology sources
  after a hard triage — the 200+ pop titles are struck by list, with the list
  recorded so nothing is silently skipped.
- Wave 4: astrology books folder only on explicit request.

**Phase 3 — MINE.** Per book: every stated procedure becomes a rule record
(quote, computed page hash, honest derivation tier) and, where falsifiable, a
node under the existing discipline: pure function of MarketView, band > 0,
`_horizon` imported, `level_is_testable` gated, measured on the development
region before wiring, holdout untouched. Where a book restates a method already
built, it becomes a CITATION on the existing record, not a new node — the
family-wise correction prices every new rule, so duplication costs evidence.

**Phase 4 — CONNECT (the point of the whole exercise).** The user's words:
"some will connect. will be measuring the same thing in different ways."
- Cognee ingests the new corpus (`cognee_ingest_corpus.py` extended to
  corpus/library); it PROPOSES links; `cognee_propose_links.py --promote`
  writes only quote-verified ones to Neo4j. The one-way gate stays.
- The doctrine graph gets the cross-library edges with a citation on both ends:
  Jenkins' measured move against Moore/Woods/Pickell-Daniel; Bayer's Time
  Factors against Gann's seasonal dates and the Eitel solar terms; Sepharial's
  Arcana key against his bound kabalism chapters; Cowan's full 4D text against
  the charts volume already mined; the interpreters' square-of-nine
  constructions against the wheel plates read by vision.
- Same-quantity detection is the explicit goal: where two schools' nodes emit
  the same claim from the same inputs, the edge marks them one witness, so
  confluence stops double-counting — that is what "the bigger picture" means
  operationally.

**Phase 5 — MEASURE.** Full replay on the deep data with the complete method
set; family-wise correction at the enlarged N; survivors named. Then, once and
only once, the sealed year on the best candidate; then paper trading through
the ladder's gates; the website last, built from whatever is actually true.

## Standing rules, unchanged

Every claim cites a page whose hash a test recomputes. Vision reads what OCR
destroys. Refusals are recorded, not skipped. Development performance chooses
candidates and is never evidence. The holdout is spent once. No method is
retired for scoring badly on the data it was developed on.