reorg: 业务线重组 B1–B9（provider seam + 参数归属 + 预取 + 验收整改）

.github/agents/pipeline-doctor.agent.md

@@ -9,9 +9,9 @@ You are a data pipeline diagnostician for the OptionView project. Your job is to
 ## Architecture
 
 ```
-data_pipeline/downloader.py → raw_prices table
-data_pipeline/cleaning.py   → clean_prices table
-data_pipeline/processing.py → processed_prices table
+data_pipeline/ingest/ohlcv.py → raw_bars table
+data_pipeline/transform/cleaning.py   → clean_bars table
+data_pipeline/transform/processing.py → feature_bars table
 core/price_dynamic.py       → features DataFrame
 core/market_analyzer.py     → chart generation
 services/market/analysis/facade.py → base64 images to frontend
@@ -26,8 +26,8 @@ services/market/analysis/facade.py → base64 images to frontend
 ## Approach
 
 1. **Clarify symptom**: What's the user seeing? Empty chart, wrong data, error message?
-2. **Check DB tables** (raw_prices → clean_prices → processed_prices) for the target ticker
-3. **Look for NaN-only rows**: `SELECT count(*) FROM raw_prices WHERE ticker=? AND open IS NULL AND close IS NULL`
+2. **Check DB tables** (raw_bars → clean_bars → feature_bars) for the target ticker
+3. **Look for NaN-only rows**: `SELECT count(*) FROM raw_bars WHERE ticker=? AND open IS NULL AND close IS NULL`
 4. **Check logs**: Look for yfinance errors (429, timeout), "No new data", pipeline warnings
 5. **Trace the failure**: Which stage first produced invalid data? Follow downstream
 6. **Check connectivity**: If download is suspected, verify proxy and throttle state

.github/copilot-instructions.md

@@ -25,7 +25,7 @@ Flask-based market analysis dashboard with options strategy tools.
 
 ## Database
 
-- SQLite via `data_pipeline/db.py` — always use `get_conn()` context manager
+- SQLite via `data_pipeline/store/db.py` — always use `get_conn()` context manager
 - WAL mode enabled; `PRAGMA synchronous=NORMAL`
 - DB path from `MARKET_DB_PATH` env var, default `./market_data.sqlite`
 
@@ -59,7 +59,7 @@ gunicorn app:app -b 0.0.0.0:5000  # production
 
 Before suggesting non-trivial changes, consult these:
 
-- **⚠ [docs/plans/business_line_reorg.md](../docs/plans/business_line_reorg.md)** — **active reorg** of `data_pipeline/`, the parameter surfaces, and `routes/core.py::index`. ADRs [0011](../docs/decisions/0011-pluggable-data-provider-seam.md) / [0012](../docs/decisions/0012-parameter-ownership-and-prefetch.md) are **Accepted**. Read §0 first: work batches B1–B8 in order, one batch per PR, update the ledger in the same commit, don't re-litigate the Accepted ADRs.
+- **[docs/plans/business_line_reorg.md](../docs/plans/business_line_reorg.md)** — the 2026-09 business-line reorg (provider seam + canonical schema, ticker-only Parameters bar, readiness prefetch). Batches **B1–B9 landed**; §10 lists the deferred follow-ups. ADRs [0011](../docs/decisions/0011-pluggable-data-provider-seam.md) / [0012](../docs/decisions/0012-parameter-ownership-and-prefetch.md) are **Accepted** — don't re-litigate.
 - **[docs/constraints.md](../docs/constraints.md)** — external/historical constraints (yfinance limits, SQLite choice, single-machine assumption, intentional "magic numbers"). Read this before flagging anything as tech debt.
 - **[docs/glossary.md](../docs/glossary.md)** — domain terms (IV vs HV, Greeks, regime, anomaly flags). Read this before assuming a term means what you think it means.
 - **[docs/decisions/](../docs/decisions/)** — Architecture Decision Records. Each ADR explains the context, options considered, and accepted trade-offs for a major design choice.

.github/data/arch_baseline.json

@@ -2,5 +2,5 @@
   "layer_violations": 0,
   "cycles": 0,
   "god_files": 0,
-  "dead_code_candidates": 1
+  "dead_code_candidates": 0
 }

.github/data/failure-registry.yaml

@@ -25,8 +25,8 @@ patterns:
     resolved: false
     resolution_note: null
     related_files:
-      - data_pipeline/cleaning.py
-      - data_pipeline/downloader.py
+      - data_pipeline/transform/cleaning.py
+      - data_pipeline/ingest/ohlcv.py
       - core/price_dynamic.py
 
   empty-dataframe:
@@ -39,8 +39,8 @@ patterns:
     resolved: false
     resolution_note: null
     related_files:
-      - data_pipeline/downloader.py
-      - data_pipeline/processing.py
+      - data_pipeline/ingest/ohlcv.py
+      - data_pipeline/transform/processing.py
       - services/market/analysis/facade.py
 
   dtype-mismatch:
@@ -53,7 +53,7 @@ patterns:
     resolved: false
     resolution_note: null
     related_files:
-      - data_pipeline/db.py
+      - data_pipeline/store/db.py
       - core/options_greeks.py
       - core/market_analyzer.py
 
@@ -67,7 +67,7 @@ patterns:
     resolved: false
     resolution_note: null
     related_files:
-      - data_pipeline/downloader.py
+      - data_pipeline/ingest/ohlcv.py
       - utils/utils.py
 
   db-error:
@@ -80,7 +80,7 @@ patterns:
     resolved: false
     resolution_note: null
     related_files:
-      - data_pipeline/db.py
+      - data_pipeline/store/db.py
 
   greeks-edge-case:
     hook_regex: "greeks|black.?scholes|delta|gamma|theta|vega.*nan"

.github/data/tag_baseline.json

@@ -1,23 +1,23 @@
 {
-  "files_scanned": 136,
+  "files_scanned": 147,
   "tags_by_type": {
-    "WHY": 12,
-    "CONSTRAINT": 16,
-    "TRADEOFF": 2,
-    "INVARIANT": 3,
+    "WHY": 13,
+    "CONSTRAINT": 19,
+    "TRADEOFF": 3,
+    "INVARIANT": 9,
     "DOMAIN": 10,
     "HACK": 0,
     "WORKAROUND": 0
   },
   "uncovered_constants_count": 16,
   "uncovered_constants": [
-    "data_pipeline/data_ops/_globals.py:14: _UPDATE_COOLDOWN=60",
-    "data_pipeline/data_ops/_globals.py:16: _QUERY_CACHE_TTL=60",
-    "data_pipeline/data_ops/_range.py:12: _ENSURE_RANGE_TTL=300",
-    "data_pipeline/data_ops/_range.py:19: _SENTINEL_GAP_THRESHOLD_DAYS=365",
-    "data_pipeline/data_ops/_range.py:20: _SENTINEL_MIN_DB_SPAN_DAYS=365",
-    "services/market/charts.py:30: _CACHE_MAX_ENTRIES=64",
-    "services/options/preload.py:32: CACHE_TTL_MINUTES=15",
+    "data_pipeline/_state.py:29: _UPDATE_COOLDOWN=60",
+    "data_pipeline/_state.py:31: _QUERY_CACHE_TTL=60",
+    "data_pipeline/orchestrate/backfill.py:12: _ENSURE_RANGE_TTL=300",
+    "data_pipeline/orchestrate/backfill.py:19: _SENTINEL_GAP_THRESHOLD_DAYS=365",
+    "data_pipeline/orchestrate/backfill.py:20: _SENTINEL_MIN_DB_SPAN_DAYS=365",
+    "services/market/charts.py:29: _CACHE_MAX_ENTRIES=64",
+    "services/options/preload.py:34: CACHE_TTL_MINUTES=15",
     "services/options/simulation.py:36: MAX_STRIKES=15",
     "services/options/simulation.py:37: MAX_EXPIRIES=6",
     "services/options/simulation.py:38: MAX_IVS=5",

.github/instructions/data-pipeline.instructions.md

@@ -6,7 +6,7 @@ applyTo: "data_pipeline/**"
 # Data Pipeline Rules
 
 ## DB Access
-- Always use `get_conn()` context manager from `data_pipeline/db.py` — never raw `sqlite3.connect()`
+- Always use `get_conn()` context manager from `data_pipeline/store/db.py` — never raw `sqlite3.connect()`
 - Use `fetch_df()` for reads, `upsert_many()` for writes
 - Convert DB-sourced columns with `pd.to_numeric(col, errors='coerce')` before any math — SQLite returns `object` dtype
 

.github/prompts/diagnose.prompt.md

@@ -8,7 +8,7 @@ argument-hint: "Ticker symbol or symptom (e.g., 'NVDA empty charts', 'stale TLT
 Diagnose why a specific ticker's data is missing, stale, or showing errors in the OptionView dashboard.
 
 Steps:
-1. Check the DB for the ticker: query `raw_prices`, `clean_prices`, `processed_prices` for recent rows
+1. Check the DB for the ticker: query `raw_bars`, `clean_bars`, `feature_bars` for recent rows
 2. Look for NaN-only filler rows (root cause of empty charts)
 3. Check yfinance download logs for errors (429, timeout)
 4. Trace data flow through the 5-stage pipeline to find the failure point

.github/prompts/new-test.prompt.md

@@ -2,7 +2,7 @@
 description: "Generate a test following OptionView project patterns for a specific module or function."
 agent: "agent"
 tools: [read, search, edit]
-argument-hint: "Module or function to test (e.g., 'data_pipeline/cleaning.py clean_range')"
+argument-hint: "Module or function to test (e.g., 'data_pipeline/transform/cleaning.py clean_range')"
 ---
 
 Generate a pytest test for the specified module/function following OptionView test conventions:

.github/prompts/pipeline-status.prompt.md

@@ -10,22 +10,22 @@ Check the current health of the OptionView data pipeline:
 1. Query the SQLite database (`market_data.sqlite`) for:
    ```sql
    -- Row counts per table
-   SELECT 'raw_prices' as tbl, count(*) as rows FROM raw_prices
-   UNION SELECT 'clean_prices', count(*) FROM clean_prices
-   UNION SELECT 'processed_prices', count(*) FROM processed_prices;
+   SELECT 'raw_bars' as tbl, count(*) as rows FROM raw_bars
+   UNION SELECT 'clean_bars', count(*) FROM clean_bars
+   UNION SELECT 'feature_bars', count(*) FROM feature_bars;
 
    -- Latest data per ticker
-   SELECT ticker, MAX(date) as latest, COUNT(*) as rows FROM raw_prices GROUP BY ticker;
+   SELECT ticker, MAX(date) as latest, COUNT(*) as rows FROM raw_bars GROUP BY ticker;
 
    -- NaN-only filler rows (problematic)
-   SELECT ticker, count(*) as nan_rows FROM raw_prices
+   SELECT ticker, count(*) as nan_rows FROM raw_bars
    WHERE open IS NULL AND high IS NULL AND low IS NULL AND close IS NULL
    GROUP BY ticker HAVING nan_rows > 0;
 
    -- Data freshness (days since last update)
    SELECT ticker, MAX(date) as latest,
           julianday('now') - julianday(MAX(date)) as days_stale
-   FROM raw_prices GROUP BY ticker ORDER BY days_stale DESC;
+   FROM raw_bars GROUP BY ticker ORDER BY days_stale DESC;
    ```
 
 2. Report:

.github/skills/debug-pipeline/SKILL.md

@@ -23,26 +23,26 @@ Classify the user's report:
 | Symptom | Likely Layer |
 |---------|-------------|
 | Empty chart panels | core/ (PriceDynamic) or data_pipeline/ (NaN filler rows) |
-| "No data for TICKER" message | data_pipeline/downloader.py (download failed) |
+| "No data for TICKER" message | data_pipeline/ingest/ohlcv.py (download failed) |
 | Stale prices (dates from days ago) | data_pipeline/data_service.py (cooldown blocking refresh) |
 | 429 / timeout errors | yfinance rate-limiting or proxy issue |
-| Wrong values in analysis | data_pipeline/cleaning.py or processing.py |
+| Wrong values in analysis | data_pipeline/transform/cleaning.py or processing.py |
 
 ### Step 2: Check DB State
 
 Query the database for the target ticker using the terminal:
 ```sql
--- Check raw_prices for recent data
-SELECT ticker, date, close FROM raw_prices WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
+-- Check raw_bars for recent data
+SELECT ticker, date, close FROM raw_bars WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
 
 -- Check for NaN-only filler rows (the root cause of empty charts)
-SELECT count(*) FROM raw_prices WHERE ticker='{TICKER}' AND open IS NULL AND high IS NULL AND low IS NULL AND close IS NULL;
+SELECT count(*) FROM raw_bars WHERE ticker='{TICKER}' AND open IS NULL AND high IS NULL AND low IS NULL AND close IS NULL;
 
--- Check clean_prices status
-SELECT ticker, date, missing_any, price_jump_flag FROM clean_prices WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
+-- Check clean_bars status
+SELECT ticker, date, missing_any, price_jump_flag FROM clean_bars WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
 
--- Check processed_prices
-SELECT ticker, date, frequency FROM processed_prices WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
+-- Check feature_bars
+SELECT ticker, date, frequency FROM feature_bars WHERE ticker='{TICKER}' ORDER BY date DESC LIMIT 5;
 ```
 
 ### Step 3: Check yfinance Connectivity
@@ -66,10 +66,10 @@ print(df.tail() if not df.empty else "EMPTY - download failed")
 
 Follow the data through each stage, checking for where it breaks. See [pipeline stages reference](./references/pipeline-stages.md) for expected inputs/outputs at each stage.
 
-1. **downloader.py** → `upsert_raw_prices()` → writes to `raw_prices`
-2. **cleaning.py** → `clean_range()` → reads `raw_prices`, writes to `clean_prices`
-3. **processing.py** → `build_features()` → reads `clean_prices`, writes to `processed_prices`
-4. **core/price_dynamic.py** → `_fetch_daily_from_db()` → reads `processed_prices`
+1. **downloader.py** → `upsert_raw_prices()` → writes to `raw_bars`
+2. **cleaning.py** → `clean_range()` → reads `raw_bars`, writes to `clean_bars`
+3. **processing.py** → `build_features()` → reads `clean_bars`, writes to `feature_bars`
+4. **core/price_dynamic.py** → `_fetch_daily_from_db()` → reads `feature_bars`
 5. **core/market_analyzer.py** → uses PriceDynamic features for charts
 6. **services/market/analysis/facade.py** → calls chart methods, returns base64 images
 
@@ -78,7 +78,7 @@ Follow the data through each stage, checking for where it breaks. See [pipeline
 Common root causes:
 | Root Cause | Evidence | Fix |
 |-----------|----------|-----|
-| NaN-only filler rows from failed download | `raw_prices` has NULL in all price columns | Re-download with `yf_throttle()`, delete filler rows |
+| NaN-only filler rows from failed download | `raw_bars` has NULL in all price columns | Re-download with `yf_throttle()`, delete filler rows |
 | 60s cooldown blocking retry | Download skipped, log says "No new data" | Wait 60s or reset cooldown in `DataService._ticker_locks` |
 | Proxy unreachable | `curl: (28) Operation timed out` | Check `YF_PROXY` in `.env`, verify proxy is running |
 | yfinance 429 rate limit | `YFRateLimitError` in logs | Wait 30s, ensure `yf_throttle()` is called everywhere |

.github/skills/debug-pipeline/references/pipeline-stages.md

@@ -2,52 +2,52 @@
 
 Data flows through 5 stages. A failure at any stage can propagate downstream as empty/NaN data.
 
-## Stage 1: Download (`data_pipeline/downloader.py`)
+## Stage 1: Download (`data_pipeline/ingest/ohlcv.py`)
 
 **Function**: `upsert_raw_prices(ticker, start, end)`
 **Input**: Ticker symbol, date range
 **Output**: `PipelineResult(ok=True, rows=N)` or `PipelineResult(ok=False, error="...")`
-**Side effect**: Writes to `raw_prices` table
+**Side effect**: Writes to `raw_bars` table
 
 **What can fail**:
 - yfinance returns empty DataFrame (rate-limit, invalid ticker, network error)
 - Proxy unreachable (curl_cffi timeout)
 - Staleness check incorrectly skips download
 
-**Check**: `SELECT count(*) FROM raw_prices WHERE ticker=? AND date BETWEEN ? AND ?`
+**Check**: `SELECT count(*) FROM raw_bars WHERE ticker=? AND date BETWEEN ? AND ?`
 
-## Stage 2: Clean (`data_pipeline/cleaning.py`)
+## Stage 2: Clean (`data_pipeline/transform/cleaning.py`)
 
 **Function**: `clean_range(ticker, start, end)`
-**Input**: Reads from `raw_prices` table
-**Output**: `PipelineResult` — writes to `clean_prices` table
+**Input**: Reads from `raw_bars` table
+**Output**: `PipelineResult` — writes to `clean_bars` table
 **Side effect**: Adds anomaly flags (price_jump_flag, vol_anom_flag, ohlc_inconsistent)
 
 **What can fail**:
-- Source `raw_prices` has NaN-only filler rows → cleans "pass through" NaN
+- Source `raw_bars` has NaN-only filler rows → cleans "pass through" NaN
 - `pd.to_numeric()` coerces strings to NaN silently
 - Anomaly flag thresholds are heuristic — may miss or over-flag
 
-**Check**: `SELECT date, missing_any, price_jump_flag FROM clean_prices WHERE ticker=? ORDER BY date DESC LIMIT 10`
+**Check**: `SELECT date, missing_any, price_jump_flag FROM clean_bars WHERE ticker=? ORDER BY date DESC LIMIT 10`
 
-## Stage 3: Process (`data_pipeline/processing.py`)
+## Stage 3: Process (`data_pipeline/transform/processing.py`)
 
 **Function**: `build_features(ticker, frequency)`
-**Input**: Reads from `clean_prices` table
-**Output**: `PipelineResult` — writes to `processed_prices` table
+**Input**: Reads from `clean_bars` table
+**Output**: `PipelineResult` — writes to `feature_bars` table
 **Side effect**: Computes MA, returns, volatility features
 
 **What can fail**:
 - Insufficient clean data for rolling window calculations → features are NaN
 - Wrong frequency conversion (D→W→M) drops rows
 - `object` dtype from DB causes numpy math errors
 
-**Check**: `SELECT date, frequency, ma_20, ma_50 FROM processed_prices WHERE ticker=? AND frequency=? ORDER BY date DESC LIMIT 5`
+**Check**: `SELECT date, frequency, ma_20, ma_50 FROM feature_bars WHERE ticker=? AND frequency=? ORDER BY date DESC LIMIT 5`
 
 ## Stage 4: Core Analysis (`core/price_dynamic.py`, `core/market_analyzer.py`)
 
 **Function**: `PriceDynamic._fetch_daily_from_db()` → `MarketAnalyzer` methods
-**Input**: Reads from `processed_prices` (or `clean_prices` for some features)
+**Input**: Reads from `feature_bars` (or `clean_bars` for some features)
 **Output**: DataFrames for chart generation
 
 **What can fail**:

.github/skills/fix-review/SKILL.md

@@ -61,7 +61,7 @@ Verify the fix uses the correct error pattern for its layer:
 For each changed production file, check that a corresponding test exists:
 ```bash
 # Map production file to test file
-# data_pipeline/downloader.py → tests/test_yf_download.py or tests/test_processing.py
+# data_pipeline/ingest/ohlcv.py → tests/test_yf_download.py or tests/test_processing.py
 # core/market_analyzer.py → tests/test_market_review.py
 # services/market/validation.py → tests/test_validation.py
 ```

.github/skills/test-escalation/SKILL.md

@@ -63,7 +63,7 @@ def test_download_empty(mock_dl):
     assert result.ok  # Rows=0 is valid
     assert result.rows == 0
     # Verify no NaN filler rows were created
-    df = fetch_df("SELECT * FROM raw_prices WHERE ticker='NVDA'")
+    df = fetch_df("SELECT * FROM raw_bars WHERE ticker='NVDA'")
     assert df.empty
 ```
 
@@ -78,16 +78,16 @@ def test_full_pipeline_with_nan_data(tmp_path, monkeypatch):
 
     # Seed NaN-only filler rows (simulates failed download)
     upsert_many(
-        "raw_prices",
+        "raw_bars",
         ["ticker", "date", "open", "high", "low", "close"],
         [("NVDA", "2026-03-28", None, None, None, None)],
     )
 
     # Run cleaning — should NOT propagate NaN rows
     result = clean_range("NVDA", dt.date(2026, 3, 28), dt.date(2026, 3, 28))
 
-    # Verify: clean_prices should be empty (NaN rows filtered)
-    df = fetch_df("SELECT * FROM clean_prices WHERE ticker='NVDA'")
+    # Verify: clean_bars should be empty (NaN rows filtered)
+    df = fetch_df("SELECT * FROM clean_bars WHERE ticker='NVDA'")
     assert df.empty or df["close"].notna().all()
 ```
 

.github/skills/test-escalation/references/escalation-levels.md

@@ -32,7 +32,7 @@ Is the bug reproducible with a simple unit test?
 - **Symptom**: Empty charts, 0 historical data points
 - **Root cause**: NaN-only filler rows from failed download survive cleaning
 - **Effective level**: Level 2 (integration — needs real DB to reproduce the chain)
-- **Key assertion**: After pipeline, `processed_prices` has no NaN-only rows
+- **Key assertion**: After pipeline, `feature_bars` has no NaN-only rows
 
 ### Pattern B: yfinance Silent Failure
 - **Symptom**: Data appears stale, "No new data" in logs