2026-03-05 v0.0.16 — MODEL_INFO elimination, registry SSoT, ETS refit fix, refit contract tests

eddmpython · eddmpython · commit d3d3d775f592 · 2026-03-05T13:00:19.000+09:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,42 @@ All notable changes to Vectrix will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.0.16] - 2026-03-05
+
+Code quality and architecture cleanup release — eliminates MODEL_INFO indirection layer, enforces registry as single source of truth for all model metadata, fixes ETSModel refit bug, and adds refit contract tests.
+
+### Fixed
+
+**ETSModel refit() State Corruption**
+- `engine/ets.py`: `_computeSSE()` called `_filter()` which mutated `self.level/trend/seasonal` as a side-effect during optimization
+- This caused refit() on the same data to produce 175%+ divergence from original fit()
+- Added state save/restore in `_computeSSE` using try/finally pattern
+- All 6 refit-capable models now pass same-data similarity test (<5% divergence)
+
+### Changed
+
+**Registry as Single Source of Truth (SSoT)**
+- `engine/registry.py`: Added `tier`, `seasonal`, `hourly` fields to `ModelSpec`
+- Added `selectModels()` for data-characteristic-based model selection
+- Added `getCoreModelIds()` for dynamic ensemble core pool
+- `vectrix.py`: Replaced hardcoded model lists with `selectModels()` and `getCoreModelIds()`
+- `flat_defense/diagnostic.py`: Replaced `MODEL_INFO` import with `getModelInfo()` direct call
+- `models/selector.py`: Same — `MODEL_INFO` → `getModelInfo()` direct call
+- `types.py`: Removed `_LazyModelInfo` class and `MODEL_INFO` variable entirely (48 lines deleted)
+
+**Exception Handling**
+- `vectrix.py`: Replaced 6 bare `except Exception:` with specific types (`ValueError`, `RuntimeError`, `np.linalg.LinAlgError`, etc.)
+
+**Code Deduplication**
+- `vectrix.py`: Unified sequential/parallel model evaluation result handling into `storeResult()` helper
+
+### Added
+
+**Refit Contract Tests**
+- `tests/test_refit_contract.py`: 30 tests covering 6 refit-capable models
+- Tests: returns_self, output_shape, output_finite, ci_ordering, same_data_similar_output
+- Total tests: 573 → 603
+
 ## [0.0.15] - 2026-03-05
 
 Accuracy & engine improvement release — FFT period detection bug fix eliminates spurious seasonal periods that degraded M4 Daily OWA from 0.820 to 0.996. Three new Rust DNA functions added. PeriodicDropDetector seasonal false positive fix. Visualization module (plotly). Overall M4 AVG OWA improved from 0.877 to 0.848.
diff --git a/Cargo.toml b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "vectrix"
-version = "0.0.15"
+version = "0.0.16"
 edition = "2021"
 license-file = "LICENSE"
 
diff --git a/constants.json b/constants.json
@@ -1,5 +1,5 @@
 {
-  "version": "0.0.15",
+  "version": "0.0.16",
   "rustFunctions": 29,
   "testCount": 603,
   "modelCount": "30+",
diff --git a/docs/overrides/home.html b/docs/overrides/home.html
@@ -11,7 +11,7 @@
   <section class="vx-hero">
     <div class="vx-hero-badge">
       <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"></polygon></svg>
-      v0.0.15 — Built-in Rust Engine
+      v0.0.16 — Built-in Rust Engine
     </div>
     <h1 class="vx-hero-title">
       Time Series Forecasting<br>
diff --git a/landing/src/lib/components/sections/Hero.svelte b/landing/src/lib/components/sections/Hero.svelte
@@ -15,7 +15,7 @@
 	<div class="relative z-10 max-w-4xl mx-auto">
 		<Badge class="mb-6">
 			<Zap class="w-3.5 h-3.5" />
-			v0.0.15 — Built-in Rust Engine
+			v0.0.16 — Built-in Rust Engine
 		</Badge>
 
 		<h1 class="text-5xl md:text-7xl font-extrabold tracking-tight leading-[1.1] mb-4">
diff --git a/landing/src/routes/+page.svelte b/landing/src/routes/+page.svelte
@@ -41,7 +41,7 @@
 		"description": "Zero-config time series forecasting for Python. 30+ models, one line of code, built-in Rust engine.",
 		"url": "https://eddmpython.github.io/vectrix/",
 		"downloadUrl": "https://pypi.org/project/vectrix/",
-		"softwareVersion": "0.0.15",
+		"softwareVersion": "0.0.16",
 		"author": { "@type": "Person", "name": "eddmpython", "url": "https://github.com/eddmpython" },
 		"offers": { "@type": "Offer", "price": "0", "priceCurrency": "USD" },
 		"license": "https://opensource.org/licenses/MIT",
diff --git a/landing/static/llms-full.txt b/landing/static/llms-full.txt
@@ -4,7 +4,7 @@
 
 Vectrix is a Python library for time series forecasting. It evaluates 30+ statistical models (ETS, ARIMA, Theta, CES, DOT, MSTL, TBATS, GARCH, and more), validates each on a holdout set, and returns the best one with 95% confidence intervals — in a single function call.
 
-- Version: 0.0.15
+- Version: 0.0.16
 - License: MIT
 - Python: 3.10+
 - Dependencies: NumPy, pandas, SciPy (only 3)
diff --git a/llms-full.txt b/llms-full.txt
@@ -2,7 +2,7 @@
 
 > Zero-config time series forecasting library for Python. Automatic model selection with built-in Rust engine (29 accelerated functions), adaptive intelligence, full regression suite, and business analytics.
 
-- Version: 0.0.15
+- Version: 0.0.16
 - Python: >=3.10
 - Core deps: numpy>=1.24, pandas>=2.0, scipy>=1.10
 - Install: `pip install vectrix`
diff --git a/llms.txt b/llms.txt
@@ -2,7 +2,7 @@
 
 > Zero-config time series forecasting library for Python. Automatic model selection with built-in Rust engine. NumPy/SciPy/Pandas with adaptive intelligence, regression, and business analytics.
 
-- Version: 0.0.15
+- Version: 0.0.16
 - Python: 3.10+
 - Dependencies: numpy, pandas, scipy (core only)
 - Install: `pip install vectrix`
diff --git a/pyproject.toml b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "vectrix"
-version = "0.0.15"
+version = "0.0.16"
 description = "Zero-config time series forecasting & analysis library. 30+ models with built-in Rust engine for blazing-fast performance."
 readme = "README.md"
 license = {file = "LICENSE"}
diff --git a/rust/Cargo.toml b/rust/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "vectrix-core"
-version = "0.0.15"
+version = "0.0.16"
 edition = "2021"
 
 [lib]
diff --git a/rust/pyproject.toml b/rust/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "vectrix-core"
-version = "0.0.15"
+version = "0.0.16"
 requires-python = ">=3.10"
 description = "Rust-accelerated core for Vectrix time series forecasting"
 license = { text = "MIT" }
diff --git a/src/vectrix/flat_defense/diagnostic.py b/src/vectrix/flat_defense/diagnostic.py
@@ -8,7 +8,8 @@
 
 import numpy as np
 
-from ..types import MODEL_INFO, DataCharacteristics, FlatRiskAssessment, RiskLevel
+from ..engine.registry import getModelInfo
+from ..types import DataCharacteristics, FlatRiskAssessment, RiskLevel
 
 
 class FlatRiskDiagnostic:
@@ -237,9 +238,10 @@ def _getRecommendation(
             strategy = "standard"
             models = ['auto_arima', 'auto_ets', 'theta', 'mstl']
 
+        modelInfo = getModelInfo()
         models = [
             m for m in models
-            if dataLength >= MODEL_INFO.get(m, {}).get('minData', 10)
+            if dataLength >= modelInfo.get(m, {}).get('minData', 10)
         ]
 
         if not models:
diff --git a/src/vectrix/models/selector.py b/src/vectrix/models/selector.py
@@ -6,8 +6,8 @@
 
 from typing import Any, Dict, List, Optional
 
-from ..engine.registry import getRegistry
-from ..types import MODEL_INFO, DataCharacteristics, FlatRiskAssessment, RiskLevel
+from ..engine.registry import getModelInfo, getRegistry
+from ..types import DataCharacteristics, FlatRiskAssessment, RiskLevel
 
 
 def _buildFromRegistry(field: str, fallback: Dict[str, Any]) -> Dict[str, Any]:
@@ -132,7 +132,7 @@ def _prioritizeSeasonalModels(self, models: List[str]) -> List[str]:
 
     def getModelInfo(self, modelId: str) -> Dict[str, Any]:
         """Return model information"""
-        return MODEL_INFO.get(modelId, {
+        return getModelInfo().get(modelId, {
             'name': modelId,
             'description': '',
             'flatResistance': 0.5,
diff --git a/src/vectrix/types.py b/src/vectrix/types.py
@@ -193,51 +193,3 @@ def getSummary(self) -> Dict[str, Any]:
             'flatDetected': self.flatInfo.isFlat if self.flatInfo else False,
             'warningCount': len(self.warnings)
         }
-
-
-def _getModelInfo():
-    """Get MODEL_INFO from the registry (single source of truth)."""
-    from .engine.registry import getModelInfo
-    return getModelInfo()
-
-
-class _LazyModelInfo(dict):
-    """Lazy-loading MODEL_INFO that initializes from registry on first access."""
-
-    _loaded = False
-
-    def _ensureLoaded(self):
-        if not self._loaded:
-            self.update(_getModelInfo())
-            self._loaded = True
-
-    def __getitem__(self, key):
-        self._ensureLoaded()
-        return super().__getitem__(key)
-
-    def __contains__(self, key):
-        self._ensureLoaded()
-        return super().__contains__(key)
-
-    def __iter__(self):
-        self._ensureLoaded()
-        return super().__iter__()
-
-    def keys(self):
-        self._ensureLoaded()
-        return super().keys()
-
-    def values(self):
-        self._ensureLoaded()
-        return super().values()
-
-    def items(self):
-        self._ensureLoaded()
-        return super().items()
-
-    def get(self, key, default=None):
-        self._ensureLoaded()
-        return super().get(key, default)
-
-
-MODEL_INFO = _LazyModelInfo()
diff --git a/src/vectrix/vectrix.py b/src/vectrix/vectrix.py
@@ -43,7 +43,7 @@ class Vectrix:
     Dependencies: numpy, pandas, scipy (required), numba (optional)
     """
 
-    VERSION = "0.0.14"
+    VERSION = "0.0.16"
 
     def __init__(
         self,
@@ -301,20 +301,23 @@ def evaluateSingle(modelId):
             )
             return modelId, result, fittedModel
 
+        def storeResult(mid, result, fittedModel, idx):
+            results[mid] = result
+            if fittedModel is not None:
+                self._fittedModels[mid] = fittedModel
+            self._progress(f'{result.modelName} done ({idx}/{totalModels})')
+            if self.verbose:
+                flatMark = "⚠️" if result.flatInfo and result.flatInfo.isFlat else "✓"
+                print(f"  {flatMark} {mid}: MAPE={result.mape:.2f}%")
+
         nWorkers = min(totalModels, 4) if self.nJobs != 1 else 1
 
         if nWorkers <= 1:
             for i, modelId in enumerate(modelIds):
                 try:
                     self._progress(f'Training {self._getModelName(modelId)}...')
                     mid, result, fittedModel = evaluateSingle(modelId)
-                    results[mid] = result
-                    if fittedModel is not None:
-                        self._fittedModels[mid] = fittedModel
-                    self._progress(f'{result.modelName} done ({i+1}/{totalModels})')
-                    if self.verbose:
-                        flatMark = "⚠️" if result.flatInfo and result.flatInfo.isFlat else "✓"
-                        print(f"  {flatMark} {modelId}: MAPE={result.mape:.2f}%")
+                    storeResult(mid, result, fittedModel, i + 1)
                 except (ValueError, RuntimeError, np.linalg.LinAlgError) as e:
                     if self.verbose:
                         print(f"  ✗ {modelId} error: {str(e)[:50]}")
@@ -330,13 +333,7 @@ def evaluateSingle(modelId):
                     completed += 1
                     try:
                         mid, result, fittedModel = future.result()
-                        results[mid] = result
-                        if fittedModel is not None:
-                            self._fittedModels[mid] = fittedModel
-                        self._progress(f'{result.modelName} done ({completed}/{totalModels})')
-                        if self.verbose:
-                            flatMark = "⚠️" if result.flatInfo and result.flatInfo.isFlat else "✓"
-                            print(f"  {flatMark} {modelId}: MAPE={result.mape:.2f}%")
+                        storeResult(mid, result, fittedModel, completed)
                     except (ValueError, RuntimeError, np.linalg.LinAlgError) as e:
                         if self.verbose:
                             print(f"  ✗ {modelId} error: {str(e)[:50]}")

Original file line number	Diff line number	Diff line change
`@@ -1,5 +1,5 @@`
`1`	`1`	`{`
`2`		`- "version": "0.0.15",`
	`2`	`+ "version": "0.0.16",`
`3`	`3`	`"rustFunctions": 29,`
`4`	`4`	`"testCount": 603,`
`5`	`5`	`"modelCount": "30+",`