ARIMA_SAX_Hybrid_Forecaster/MALE5_v3.2.0_CANONICAL_DEPENDENCY_FINALIZATION.md

# MALE5 v3.2.0 CANONICAL DEPENDENCY FINALIZATION REPORT

## 1. Upstream URL
https://github.com/MegaJoctan/MALE5

## 2. Tag
`v3.2.0`

## 3. Exact commit SHA
`d6fb0695a7c2fbeaa03de379557a8b5447915384`

Dereferenced from annotated tag object `2b239123e204bdf27fdcbe28fd95d20bd5d4a000`
(tagger: Omega Joctan, message: "GBDTs release tag").

## 4. Exact archive SHA-256
`ad161d75ae0b54ee54a1fd3b0d48c315cf776eebc97077a4b1ae7a629f0c0d4d`

- Independently recomputed from a fresh authoritative download:
  `https://codeload.github.com/MegaJoctan/MALE5/zip/refs/tags/v3.2.0` (108225 bytes).
- Byte-for-byte identical (fc /b) to the prior forensic artifact `archive_v320_github.zip`.
- Prefix `ad161d75...` matches the previously reported figure. NO conflict.

## 5. Canonical local path
`D:\TradingTerminal\MetaTrader 5\MQL5\Include\MALE5`

Installed from the verified GitHub v3.2.0 archive only (MALE5-3.2.0 snapshot).
No v3.5.0 files, no patching, no include-path edits. 48 files, all byte-identical
(SHA-256) to the verified archive snapshot. `checked=48 mismatch=0`.

## 6. Local dependency identity
Canonical MALE5 v3.2.0 (commit `d6fb0695...`). Top-level structure matches v3.2.0 only:
`.github`, `Clustering`, `Decision Tree`, `Dimensionality Reduction`, `Ensemble`,
`LightGBM`, `Linear Models`, `MqPlotLib`, `Naive Bayes`, `Neighbors`, `Neural Networks`,
`SVM`, `XGBoost`, and root: `cross_validation.mqh`, `kernels.mqh`, `LICENSE`, `linalg.mqh`,
`MatrixExtend.mqh`, `metrics.mqh`, `preprocessing.mqh`, `README.md`, `README_russian.md`,
`requirements.txt`, `Tensors.mqh`, `TODO.txt`, `.gitignore`. (48 files)

## 7. Relevant source hashes (RF dependency closure, canonical install)
| File | SHA-256 |
|------|---------|
| Ensemble\Random Forest.mqh | CB0935AC6B69283FF0D2DE2B2FC7C1460D14983F23A56E0B3D2E9CE8DCCFC0DB |
| Decision Tree\tree.mqh | 9568B9F903876ACCA0CFA1340CFFE19F071FCAFEEB0C67AE4802974D6DD23B2A |
| MatrixExtend.mqh | 45BA8C7F25DAFE8BC9022BFE22DF1759E8CB063BCC024FC35BE18A705AE0EA24 |
| metrics.mqh | C989AA23C9CB1EE1DF33E848F7B27359A144763EBE24C8C15DE8657DE349D594 |
| preprocessing.mqh | CAF058E3EA4729317CFD87286C7989E6B9ACD71133F8F2A0806FCE3A2B254FC8 |
| kernels.mqh | 8D1FAD2B34EEB4E0E621B740DD6B236C31528D18F2926673C180E415EB4E8E0D |
| linalg.mqh | C835E73879CC02948F7EAAD78346E807AE6FBE6AE243F03E55FB5DE8795E56B5 |
| Tensors.mqh | FC0E70208F9C1ECC84F9F6B6F0B691BF4803084E3DC4CD17011F5DC88B0DF8D1 |
| cross_validation.mqh | 89AB4824E1C3E6EF2C00F5C53FD49EF21156F82D5542026D3513CE3737FA2D16 |
| Ensemble\AdaBoost.mqh | 4151A9292DA80F79DF97B02EC16C45BD20D51F08EAB799B3624852FC9560510A |

Include closure (as built): `Random Forest.mqh -> Decision Tree\tree.mqh, metrics.mqh
-> MatrixExtend.mqh -> preprocessing.mqh, and MqPlotLib\plots.mqh -> Graphics\Graphic.mqh`.

## 8. Compile smoke result
Compile-only smoke test `Files\Temp\MALE5_FORENSIC\smoke_finalization.mq5`
(include CRandomForestClassifier; instantiate; fit; predict on tiny synthetic data):
**0 errors, 0 warnings** (AVX512 + FMA3, build 6140).

## 9. predict_proba status
**UNSUPPORTED.** `CRandomForestClassifier` exposes only `fit`, `predict` (double),
and `predict(vector&)`/`predict(matrix&)` returning hard labels. No `predict_proba`
present in any v3.2.0 `.mqh`. Not added, not emulated. MALE5-ML protocol remains
binary-classification based unless a separate protocol is explicitly approved.

## 10. v3.5.0 exclusion confirmation
Confirmed excluded. The canonical install contains no v3.5.0-only artifacts
(`Sklearn/`, `Numpy/`, `Pandas/`, `Stats Models/`, `Utils.mqh`, `Examples/` are absent).

## 11. Any discrepancy
None. The archive SHA-256 prefix reported previously (`ad161d75...`) is consistent with
the independently computed authoritative full value.

## 12. Dependency status
**DEPENDENCY READY FOR PROTOCOL DESIGN**