Add selected opening-hour walk-forward research

.claude-plugin/marketplace.json

@@ -28,8 +28,8 @@
       "name": "quant-stock-technical",
       "displayName": "Quant Stock Technical",
       "source": "./plugins/quant-stock-technical",
-      "description": "Acquire audited KIS Korean and U.S. EOD data and screen a deterministic KOSPI, KOSDAQ, NYSE, and NASDAQ technical universe.",
-      "version": "0.4.5",
+      "description": "Acquire audited KIS EOD data, screen four exchanges, and evaluate selected opening-hour outcomes.",
+      "version": "0.4.6",
       "author": {
         "name": "mrcha033"
       },

.github/workflows/validate.yml

@@ -34,6 +34,10 @@ jobs:
         run: python -B skills/quant-stock-technical/scripts/collect_open1h_research.py --self-test
       - name: Run opening-hour evaluator self-test
         run: python -B skills/quant-stock-technical/scripts/evaluate_open1h_research.py --self-test
+      - name: Run selected opening-hour outcomes self-test
+        run: python -B skills/quant-stock-technical/scripts/selected_open1h_outcomes.py --self-test
+      - name: Run selected opening-hour walk-forward self-test
+        run: python -B skills/quant-stock-technical/scripts/walk_forward_open1h.py --self-test
       - name: Run four-exchange universe builder self-test
         run: python -B skills/quant-stock-technical/scripts/build_universe_manifest.py --self-test
       - name: Run deterministic universe screen self-test

plugins/quant-stock-technical/.claude-plugin/plugin.json

@@ -2,8 +2,8 @@
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "quant-stock-technical",
   "displayName": "Quant Stock Technical",
-  "version": "0.4.5",
-  "description": "Acquire audited KIS Korean and U.S. EOD data and screen a deterministic KOSPI, KOSDAQ, NYSE, and NASDAQ technical universe.",
+  "version": "0.4.6",
+  "description": "Acquire audited KIS EOD data, screen four exchanges, and evaluate selected opening-hour outcomes.",
   "author": {
     "name": "mrcha033",
     "url": "https://github.com/mrcha033"

plugins/quant-stock-technical/.codex-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "quant-stock-technical",
-  "version": "0.4.5",
-  "description": "Acquire audited KIS Korean and U.S. EOD data and screen a deterministic four-exchange stock universe.",
+  "version": "0.4.6",
+  "description": "Acquire audited KIS EOD data, screen four exchanges, and evaluate selected opening-hour outcomes.",
   "author": {
     "name": "mrcha033",
     "url": "https://github.com/mrcha033"
@@ -16,8 +16,8 @@
   "skills": "./skills/",
   "interface": {
     "displayName": "Quant Stock Technical",
-    "shortDescription": "Four-exchange EOD acquisition and screening",
-    "longDescription": "Acquire resumable read-only KIS Korean and U.S. adjusted EOD data, audit invalid provider rows without interpolation, and build a source-hashed KOSPI, KOSDAQ, NYSE, and NASDAQ universe.",
+    "shortDescription": "Four-exchange screening and opening-hour research",
+    "longDescription": "Acquire resumable read-only KIS Korean and U.S. adjusted EOD data, build a source-hashed four-exchange universe, and evaluate frozen selected candidates against opening-hour outcomes.",
     "developerName": "mrcha033",
     "category": "Finance",
     "capabilities": [
@@ -28,7 +28,7 @@
     "defaultPrompt": [
       "Acquire resumable read-only KIS Korean and U.S. adjusted-EOD bundles.",
       "Calculate a technical report from these completed OHLCV files.",
-      "Build and screen a frozen four-exchange technical universe."
+      "Evaluate frozen selected candidates against opening-hour outcomes."
     ]
   }
 }

plugins/quant-stock-technical/skills/quant-stock-technical/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: quant-stock-technical
-description: Acquire resumable read-only KIS Korean and U.S. adjusted-EOD bundles, calculate reproducible numbers-only stock technical analysis, and screen deterministic KOSPI, KOSDAQ, NYSE, and NASDAQ target universes. Use when the user needs completed historical OHLCV and benchmarks, an as-of-date market/ticker report, short-, medium-, and long-horizon opinions, a quantitative risk counterpoint, entry/stop/take-profit levels, a 0-100 technical-strength score, or a frozen exchange-aware screen for a separate execution system. Do not use news, fundamentals, analyst views, narrative judgment, intraday execution, or live-order claims.
+description: Acquire resumable read-only KIS Korean and U.S. adjusted-EOD bundles, calculate reproducible numbers-only stock technical analysis, screen deterministic KOSPI, KOSDAQ, NYSE, and NASDAQ target universes, and evaluate frozen selected candidates against opening-hour outcomes. Use when the user needs completed historical OHLCV and benchmarks, an as-of-date market/ticker report, short-, medium-, and long-horizon opinions, a quantitative risk counterpoint, entry/stop/take-profit levels, a 0-100 technical-strength score, a frozen exchange-aware screen, or a read-only point-in-time or explicitly survivor-biased walk-forward diagnostic. Do not use news, fundamentals, analyst views, narrative judgment, intraday execution, or live-order claims.
 ---
 
 # Quant Stock Technical
@@ -104,9 +104,17 @@ python3 scripts/build_universe_manifest.py --self-test
 python3 scripts/analyze_stock_v2.py --self-test
 python3 scripts/collect_open1h_research.py --self-test
 python3 scripts/evaluate_open1h_research.py --self-test
+python3 scripts/selected_open1h_outcomes.py --self-test
+python3 scripts/walk_forward_open1h.py --self-test
 python3 scripts/screen_universe.py --self-test
 ```
 
+For a longer selected-candidate diagnostic, use
+`scripts/walk_forward_open1h.py`. It holds one source manifest membership
+static across prior sessions and therefore labels the result
+`STATIC_SOURCE_MANIFEST_SURVIVOR_BIASED`; it is research evidence, not a
+point-in-time universe backtest or promotion evidence.
+
 When `$stock-scenario-story` will run next, save the unmodified JSON output to a file. That downstream skill requires `source_skill`, `result_schema`, `method_version`, and the complete calculator payload; do not summarize or reconstruct the handoff manually.
 
 ## Output boundaries

plugins/quant-stock-technical/skills/quant-stock-technical/agents/openai.yaml

@@ -1,4 +1,4 @@
 interface:
   display_name: "Quant Stock Technical"
-  short_description: "Four-exchange EOD acquisition and screening"
-  default_prompt: "Use $quant-stock-technical to acquire resumable read-only KIS Korean and U.S. adjusted-EOD bundles, calculate reproducible technical scores, or build and screen a frozen KOSPI, KOSDAQ, NYSE, and NASDAQ universe."
+  short_description: "Four-exchange screening and opening-hour research"
+  default_prompt: "Use $quant-stock-technical to acquire read-only KIS EOD data, calculate reproducible technical scores, screen a frozen four-exchange universe, or evaluate selected opening-hour outcomes."

plugins/quant-stock-technical/skills/quant-stock-technical/scripts/analyze_stock.py

@@ -215,6 +215,14 @@ def atr_adx(bars: list[Bar], period: int = 14) -> tuple[list[Optional[float]], l
             minus_di[index] = 100.0 * minus_value / atr_value
             denominator = plus_di[index] + minus_di[index]
             dx[index] = 0.0 if denominator == 0 else 100.0 * abs(plus_di[index] - minus_di[index]) / denominator
+        else:
+            # A newly listed or previously suspended instrument can begin with
+            # a flat true-range seed.  Wilder's directional movement is zero
+            # in that case; leaving DX unavailable would prevent ADX from ever
+            # initializing, even after the instrument starts trading.
+            plus_di[index] = 0.0
+            minus_di[index] = 0.0
+            dx[index] = 0.0
 
     first_adx = 2 * period - 1
     initial_dx = [value for value in dx[period : first_adx + 1] if value is not None]
@@ -459,7 +467,10 @@ def calculate(ticker_bars: list[Bar], benchmark_bars: list[Bar], market: str, ti
     percentiles: dict[str, float] = {}
     reference_counts: dict[str, int] = {}
     for name, series in {**features, **risk_features}.items():
-        percentiles[name], reference_counts[name] = percentile_rank(series, index)
+        try:
+            percentiles[name], reference_counts[name] = percentile_rank(series, index)
+        except BlockedError as exc:
+            raise BlockedError(f"{name}: {exc}") from exc
 
     short_weights = {
         "return_5": 0.25,
@@ -600,6 +611,20 @@ def synthetic_bars(length: int, drift: float, phase: float) -> list[Bar]:
 def self_test() -> None:
     ticker_bars = synthetic_bars(1100, 0.0006, 0.0)
     benchmark_bars = synthetic_bars(1100, 0.0003, 1.0)
+    flat_then_active = [
+        Bar(
+            day=date(2020, 1, 1) + timedelta(days=index),
+            open=100.0 if index < 20 else 100.0 + index - 20,
+            high=100.0 if index < 20 else 101.0 + index - 20,
+            low=100.0 if index < 20 else 99.0 + index - 20,
+            close=100.0 if index < 20 else 100.5 + index - 20,
+            volume=0.0 if index < 20 else 1_000.0,
+        )
+        for index in range(80)
+    ]
+    _, restarted_adx, _, _ = atr_adx(flat_then_active, 14)
+    assert restarted_adx[-1] is not None
+    assert math.isfinite(restarted_adx[-1])
     with tempfile.TemporaryDirectory(prefix="qta-self-test-") as directory:
         paths = [Path(directory) / "ticker.csv", Path(directory) / "benchmark.csv"]
         for path, bars in zip(paths, (ticker_bars, benchmark_bars)):