aether-quant 0.6.0__tar.gz → 0.8.0__tar.gz

{aether_quant-0.6.0 → aether_quant-0.8.0}/Dockerfile.retraining_worker

@@ -7,11 +7,15 @@ COPY experience/ ./experience/
 COPY performance/ ./performance/
 COPY regime/ ./regime/
 COPY experts/ ./experts/
+COPY features/ ./features/
 COPY topology/ ./topology/
 COPY moe/ ./moe/
 COPY inference/ ./inference/
+COPY risk/ ./risk/
 COPY retraining/ ./retraining/
 COPY train.py .
 COPY train_topology.py .
 COPY train_gating.py .
+COPY train_multitask.py .
+COPY train_sequence.py .
 CMD ["python", "-m", "retraining.worker"]

{aether_quant-0.6.0 → aether_quant-0.8.0}/PKG-INFO

@@ -1,5 +1,5 @@
 Metadata-Version: 2.4
 Name: aether-quant
-Version: 0.6.0
+Version: 0.8.0
 Summary: Aether Quant convenience CLI
 Requires-Python: >=3.10

{aether_quant-0.6.0 → aether_quant-0.8.0}/README.md

@@ -10,7 +10,7 @@
 
 <p align="center">
   <img src="https://img.shields.io/badge/python-3.10%2B-FF8C00?style=flat-square&labelColor=1A1A1A&logo=python&logoColor=white" alt="Python 3.10+">
-  <!-- AQ:TEST_BADGE_START --><img src="https://img.shields.io/badge/tests-611%2F611%20passing-brightgreen?style=flat-square&labelColor=1A1A1A" alt="611 of 611 tests passing"><!-- AQ:TEST_BADGE_END -->
+  <!-- AQ:TEST_BADGE_START --><img src="https://img.shields.io/badge/tests-934%2F945%20passing-red?style=flat-square&labelColor=1A1A1A" alt="934 of 945 tests passing"><!-- AQ:TEST_BADGE_END -->
   <img src="https://img.shields.io/pypi/v/aether-quant?style=flat-square&labelColor=1A1A1A&color=FF8C00" alt="PyPI version">
   <img src="https://img.shields.io/badge/docker-ghcr.io%2Faether--quant-2496ED?style=flat-square&labelColor=1A1A1A&logo=docker&logoColor=white" alt="Docker image on GHCR">
 </p>
@@ -32,14 +32,16 @@ Mixture-of-Experts ensemble (bullish/bearish/sideways/volatility specialists)
 routed by a learned gating network, a market-regime detector, a 3D market
 topology layer that combines a deterministic correlation embedding with a
 learned probabilistic overlay, a liquidity/market-impact engine that adjusts
-position sizing to real trading conditions, and a controlled retraining loop
-that lets the model itself evolve as markets do — all wired together and
-validated end-to-end inside QuantConnect's Lean engine. The thesis this
-project exists to test is simple to state and hard to prove: **markets are
-non-stationary, so a trading model should be too.** Every subsystem here
-exists to make the model adapt — to regime shifts, to changing correlation
-structure, to liquidity conditions — rather than to fit one historical
-window and hope it generalizes.
+position sizing to real trading conditions, a cross-sectional ranking signal
+that sizes each position by its predicted relative strength against the rest
+of the trading universe, and a controlled retraining loop that lets the
+model itself evolve as markets do — all wired together and validated
+end-to-end inside QuantConnect's Lean engine. The thesis this project exists
+to test is simple to state and hard to prove: **markets are non-stationary,
+so a trading model should be too.** Every subsystem here exists to make the
+model adapt — to regime shifts, to changing correlation structure, to
+liquidity conditions — rather than to fit one historical window and hope it
+generalizes.
 
 ## Table of Contents
 
@@ -158,11 +160,14 @@ that a controlled retraining loop reads from to evolve the model over time.
 
 ```mermaid
 flowchart LR
-    A["Lean data folder<br/>stocks, ETFs, crypto"] --> B["Feature pipeline<br/>train.py"]
+    A["Lean data folder<br/>stocks, ETFs, bonds, crypto"] --> B["Feature pipeline<br/>train.py<br/>62-dim: price/volume + indicators +<br/>regime + liquidity + topology + peer returns + macro"]
     B --> C["Regime detection<br/>trend, volatility, drawdown, correlation"]
     B --> D["3D topology modeling<br/>market structure and clusters"]
-    C --> E["Gating network<br/>the manager"]
+    B -.-> U["Sequence encoder (Phase 2)<br/>causal-TCN, 30-bar window<br/>informational only"]
+    B --> V["Multitask heads<br/>baseline + 4 experts<br/>direction + magnitude + volatility"]
+    C --> E["Gating network<br/>the manager<br/>blends probability + magnitude/volatility"]
     D --> E
+    V --> E
     E --> F["Expert modules"]
     F --> G["Bullish expert"]
     F --> H["Bearish expert"]
@@ -180,6 +185,7 @@ flowchart LR
     M --> O["Observation / simulation record"]
     N --> P["Redis event stream<br/>temporary low-latency buffer"]
     O --> P
+    U -.-> P
     P --> Q["Experience worker<br/>async batch persistence"]
     Q --> R["PostgreSQL experience database<br/>single source of truth"]
     R --> S["Performance triggers<br/>100 observations, drawdown, Sharpe, regime shift"]
@@ -187,6 +193,11 @@ flowchart LR
     T --> E
 ```
 
+Dashed edges mark the Phase 2 sequence encoder's **informational-only**
+path — it's computed every bar and reaches the experience log, but never
+the gating network, market analyzer, or position sizing (see
+`inference/README.md`'s Phase 2 section).
+
 #### Tech Stack
 
 ```mermaid
@@ -203,6 +214,7 @@ flowchart TB
     D --> D2["scikit-learn"]
     D --> D3["NumPy / Pandas"]
     D --> D4["MoE experts and gating network"]
+    D --> D5["Multitask heads (direction/magnitude/<br/>volatility) + causal-TCN sequence encoder"]
     E["Monitoring and UI"] --> E1["React/Vite webui — Tracing dashboard (port 3002 dev / 8001 Docker)"]
     E --> E2["FastAPI JSON API (port 8000)"]
     E --> E3["Telegram alerts (V2-19)<br/>notifications/telegram_worker.py"]
@@ -215,12 +227,111 @@ and more), the module map, and an honest analysis of what would need to
 change for this to become a genuinely low-latency/HFT system — see
 **[`development/v2_architecture.md`](development/v2_architecture.md)**.
 
+**The baseline model can now predict more than direction.** An optional
+multi-task model (`train.py::AetherNetMultiTask`, trained by
+`train_multitask.py`) predicts next-day direction, return magnitude and
+volatility jointly from one shared trunk — the direction-only baseline and
+experts are unchanged and still ship independently. All 4 experts also
+have their own optional multitask heads (`train.py::_train_expert_multitask()`),
+so `moe/gating.py` blends per-expert magnitude/volatility with a
+baseline-scale anchor the same weighted-average way it already blends
+direction. `main.py` threads the resulting `predicted_return_magnitude`/
+`predicted_volatility` through the market analyzer (informational only,
+never changes routing) and, opt-in via
+`phase_v2.dynamic_risk.use_predicted_volatility`, into position sizing —
+replacing the backward-looking `rolling_volatility_20d` average with an
+actual forward-looking forecast for that one calculation.
+
+**Regime, liquidity and topology are now genuine model input features,
+not just downstream consumers of its output.** Model input dimensionality
+grew from 30 to 48: regime one-hot/confidence/trend/risk score, an
+asset-intrinsic liquidity spread/dollar-volume estimate, and cross-asset
+topology correlation/risk state are all computed offline
+(`train.py::add_regime_features()`/`add_liquidity_features()`/
+`build_topology_features_by_date()`) and at runtime
+(`main.py::_build_model_input()`, reordered so regime is computed
+*before* the model runs) with verified train/runtime parity.
+
+**Multi-horizon direction, cross-sectional ranking, correlated-peer
+returns and technical indicators (model input 48 → 59).** A root-cause
+investigation found every model — baseline, all 4 experts, the multitask
+heads, the sequence encoder — sitting at backtest MCC 0.02-0.07 (noise) on
+next-day binary direction, and traced (and fixed) real data bugs
+confounding those numbers: an unadjusted-split bug in `train.py`'s own
+Lean-zip reader (Lean's live/backtest engine was never affected — see
+`development/Problems.md` #24) and a data-feed volume-unit discontinuity
+that blew up the sequence model's RMSE 31x (#23). Beyond the fixes,
+`AetherNetMultiTaskHorizons`/`AetherNetSequenceMultiTaskHorizons` (new
+sibling classes to the originals — experts/baseline/gating stay
+1d-direction-only by design) add 4 heads: `direction_5d`/`direction_20d`
+(longer-horizon direction) and `rank_5d`/`rank_20d` (per-date
+cross-sectional percentile rank of forward return — the most learnable
+target this investigation found, maps directly onto a long/short
+portfolio, evaluated via rank-IC, `train.py::compute_rank_ic()`, not MCC).
+11 new scaled input features: 4 correlated-peer lagged-return features
+(`topology/market_topology.py`'s `top_peers`/`top_peer_returns`, a genuine
+new information channel from correlation data the model previously only
+saw as a compressed scalar) and 7 technical indicators (`features/`
+package — RSI, ATR%, Bollinger %B, volume z-score, MACD histogram,
+distance-from-52-week-high, cross-sectional momentum rank), all computed
+identically offline and at runtime by construction (shared pure
+functions, not hand-matched duplicated formulas). All new signals are
+visible on `/neural-network` and in `*_training_metrics.json`; see
+`development/Changelog.md`'s "Frontier-model edge investigation" entry
+for the full build and the real rank-IC results.
+
+**`rank_20d` now sizes positions (opt-in).** A follow-up pass wired the
+one Phase-4 signal with a statistically significant full-series result —
+`rank_20d`, sequence model mean IC `0.073`/t-stat `4.40` — into an actual
+trading decision: `risk/position_sizing.py::rank_sizing_multiplier()`
+adds a fourth, bounded, direction-preserving factor to the sizing chain,
+scaling an already-approved trade UP toward `max_rank_multiplier` (default
+`1.25`) when the model predicts this asset near the top of the universe's
+20-day forward-return ranking, and DOWN toward `min_rank_multiplier`
+(default `0.75`) near the bottom — never flipping direction, same
+convention as the existing `topology_sizing_multiplier()`. **Off by
+default** (`phase_v2.dynamic_risk.rank_sizing_enabled: false`): the
+signal's non-overlapping-date subsample (28 independent 20-day windows)
+was not yet independently significant on its own (t-stat `1.20`), so it
+ships wired end-to-end but not promoted to on-by-default until validated
+on more out-of-sample data. `direction_5d`/`direction_20d`/`rank_5d`
+remain informational-only. See `risk/README.md`'s "Cross-sectional
+rank_20d → position sizing" section.
+
+**A causal-TCN sequence encoder (Phase 2) now exists alongside the
+flat-MLP trunk**, replacing the "zero temporal structure" limitation the
+original root-cause investigation flagged — `train.py::AetherNetSequenceMultiTask`,
+trained by `train_sequence.py`, over a rolling 30-bar window of
+already-computed model inputs. Informational only this pass
+(`main.py::_run_sequence_model()` — not yet wired into any trading
+decision); see `inference/README.md`'s Phase 2 section for the new
+`_conv1d_causal`/`_multihead_attention`/`_softmax`/`_layernorm_axis`
+interpreter primitives (each independently verified against real PyTorch
+modules).
+
+See `inference/README.md`, `moe/README.md`, `risk/README.md`,
+`regime/README.md`, `liquidity/README.md` and `topology/README.md` for
+the full contracts, and `development/Changelog.md`'s "Multi-task
+prediction" and "Phase 1 remainder + Phase 2" entries for the complete
+writeup.
+
 ## Universe Size
 
-The trading universe currently spans **20 assets** — 15 equities and 5
-crypto pairs — defined in `config.json`'s `phase1.universe.assets` and
-shared across training, validation, and backtesting
-(`phase1.universe.common_window`: `2014-12-01` to `2021-03-31`).
+The trading universe currently spans **30 assets** — 15 stocks/broad-market
+ETFs, 10 fixed-income (bond) ETFs, and 5 crypto pairs — defined in
+`config.json`'s `phase1.universe.assets` and shared across training,
+validation, and backtesting (`phase1.universe.common_window`: `2014-12-01`
+to `2021-03-31`). The bond ETF sleeve (Phase 1 of the 5/10 -> 9/10 roadmap,
+see [`development/Changelog.md`](development/Changelog.md)) was added
+specifically as a new, genuinely different information channel — not more
+of the same equity cross-section — and deliberately spans the duration
+curve (short/intermediate/long/aggregate) and credit spectrum
+(Treasury/investment-grade/high-yield/municipal/emerging-market) so the
+yield-curve-slope and credit-spread macro proxies computed from it
+(`features/macro_features.py`) are meaningful. Bond ETFs are registered
+with `security_type: "equity"` (they trade through Lean's ordinary equity
+subscription path, like every other ETF already in the universe, e.g.
+SPY/QQQ/IWM/EEM — not a new Lean security type).
 
 | Ticker | Type | Role |
 |---|---|---|
@@ -239,6 +350,16 @@ shared across training, validation, and backtesting
 | USO | Equity | Trading |
 | WM | Equity | Trading |
 | AAA | Equity | Observation-only (thin history) |
+| SHY | Equity (Fixed Income ETF) | Trading — short-duration Treasury (1-3y) |
+| IEF | Equity (Fixed Income ETF) | Trading — intermediate-duration Treasury (7-10y) |
+| TLT | Equity (Fixed Income ETF) | Trading — long-duration Treasury (20y+) |
+| AGG | Equity (Fixed Income ETF) | Trading — broad aggregate bond benchmark |
+| LQD | Equity (Fixed Income ETF) | Trading — investment-grade corporate |
+| HYG | Equity (Fixed Income ETF) | Trading — high-yield corporate |
+| TIP | Equity (Fixed Income ETF) | Trading — inflation-protected (TIPS) |
+| MBB | Equity (Fixed Income ETF) | Trading — mortgage-backed |
+| EMB | Equity (Fixed Income ETF) | Trading — emerging-market sovereign debt |
+| MUB | Equity (Fixed Income ETF) | Trading — municipal |
 | BTCUSD | Crypto | Trading |
 | ETHUSD | Crypto | Observation-only (thin history) |
 | LTCUSD | Crypto | Trading |
@@ -276,6 +397,19 @@ flowchart TD
         AAA["AAA"]
     end
 
+    subgraph FixedIncome["Fixed Income ETFs (10)"]
+        SHY["SHY"]
+        IEF["IEF"]
+        TLT["TLT"]
+        AGG["AGG"]
+        LQD["LQD"]
+        HYG["HYG"]
+        TIP["TIP"]
+        MBB["MBB"]
+        EMB["EMB"]
+        MUB["MUB"]
+    end
+
     subgraph Crypto["Crypto (5)"]
         BTCUSD["BTCUSD"]
         ETHUSD["ETHUSD"]
@@ -299,6 +433,16 @@ flowchart TD
     USO --- DNN
     WM --- DNN
     AAA --- DNN
+    SHY --- DNN
+    IEF --- DNN
+    TLT --- DNN
+    AGG --- DNN
+    LQD --- DNN
+    HYG --- DNN
+    TIP --- DNN
+    MBB --- DNN
+    EMB --- DNN
+    MUB --- DNN
     BTCUSD --- DNN
     ETHUSD --- DNN
     LTCUSD --- DNN
@@ -310,7 +454,7 @@ flowchart TD
     classDef observation fill:#3A3A3A,stroke:#FF8C00,color:#FF8C00,stroke-width:1px,stroke-dasharray: 4 2;
 
     class DNN hub;
-    class AAPL,SPY,QQQ,IWM,EEM,BAC,IBM,AIG,BNO,FB,GOOG,GOOGL,USO,WM,BTCUSD,LTCUSD trading;
+    class AAPL,SPY,QQQ,IWM,EEM,BAC,IBM,AIG,BNO,FB,GOOG,GOOGL,USO,WM,SHY,IEF,TLT,AGG,LQD,HYG,TIP,MBB,EMB,MUB,BTCUSD,LTCUSD trading;
     class AAA,ETHUSD,XRPUSD,ADAUSD observation;
 ```
 
@@ -341,12 +485,15 @@ aether-quant/
 ├── ml/                          # Model weights, datasets, versioned retraining candidates
 ├── storage/                     # Reserved for future persistent artifact storage
 ├── requirements/                # All requirements*.txt variants
-├── tests/                       # Full pytest suite (507 tests)
+├── tests/                       # Full pytest suite (828 tests)
 ├── backtests/                   # Lean backtest run outputs (gitignored)
 ├── Aether-quant-Obsidian-Vault/ # Auto-generated code-graph / architecture vault
 ├── main.py                      # Lean algorithm: inference, signal engine, risk controls
 ├── train.py                     # Training pipeline: dataset build, model training, validation
 ├── train_topology.py            # Offline trainer for the learned topology overlay
+├── train_gating.py              # Offline trainer for the learned gating blend
+├── train_multitask.py           # Offline trainer for the joint direction+magnitude+volatility model
+├── train_sequence.py            # Offline trainer for the Phase 2 causal-TCN sequence encoder
 ├── generate_backtest_report.py  # Regenerates this README's Backtest Results section
 ├── aq_cli.py                    # `aq` convenience CLI
 ├── config.json                  # Runtime configuration (phase1 / phase_v2 blocks)
@@ -380,7 +527,7 @@ and how it's wired in — this table is the index.
 | `retraining/` | Controlled retraining — plan/train/validate/backtest/commit/promote/rollback | [README](retraining/README.md) |
 | `risk/` | Dynamic position sizing, leverage caps, drawdown-aware sizing | [README](risk/README.md) |
 | `storage/` | Reserved placeholder for future persistent artifact storage | [README](storage/README.md) |
-| `tests/` | Pytest suite conventions (507 tests) | [README](tests/README.md) |
+| `tests/` | Pytest suite conventions (828 tests) | [README](tests/README.md) |
 | `topology/` | 3D market topology — deterministic SMACOF embedding + learned overlay | [README](topology/README.md) |
 | `visualization/` | Shared runtime-state JSON/CSV exports | [README](visualization/README.md) |
 | `webui/` | React/Vite dashboard (Overview, Risk, Topology, Neural Network, Tracing) | [README](webui/README.md) |
@@ -404,13 +551,13 @@ and how it's wired in — this table is the index.
 | Metric | Value |
 |---|---|
 | Backtest window | 2018-04-01 to 2021-04-02 |
-| Sharpe Ratio | -0.71 |
-| Net Profit | -8.069% |
-| Compounding Annual Return | -2.761% |
-| Drawdown | 12.500% |
-| Total Orders | 27 |
-| Win Rate | 17% |
-| Last updated | 2026-07-06 17:20 UTC (auto-generated by `aq backtest`) |
+| Sharpe Ratio | -0.758 |
+| Net Profit | -4.847% |
+| Compounding Annual Return | -1.640% |
+| Drawdown | 12.300% |
+| Total Orders | 36 |
+| Win Rate | 25% |
+| Last updated | 2026-07-08 13:10 UTC (auto-generated by `aq backtest`) |
 <!-- AQ:BACKTEST_END -->
 
 <details>
@@ -419,33 +566,33 @@ and how it's wired in — this table is the index.
 <!-- AQ:BACKTEST_FULL_STATS_START -->
 | Metric | Value |
 |---|---|
-| Total Orders | 27 |
-| Average Win | 0.18% |
-| Average Loss | -1.70% |
-| Compounding Annual Return | -2.761% |
-| Drawdown | 12.500% |
-| Expectancy | -0.816 |
+| Total Orders | 36 |
+| Average Win | 0.38% |
+| Average Loss | -0.67% |
+| Compounding Annual Return | -1.640% |
+| Drawdown | 12.300% |
+| Expectancy | -0.609 |
 | Start Equity | 100000.00 |
-| End Equity | 91930.60 |
-| Net Profit | -8.069% |
-| Sharpe Ratio | -0.71 |
-| Sortino Ratio | -0.471 |
-| Probabilistic Sharpe Ratio | 0.171% |
-| Loss Rate | 83% |
-| Win Rate | 17% |
-| Profit-Loss Ratio | 0.11 |
-| Alpha | -0.058 |
-| Beta | 0.195 |
-| Annual Standard Deviation | 0.051 |
-| Annual Variance | 0.003 |
-| Information Ratio | -0.972 |
-| Tracking Error | 0.155 |
-| Treynor Ratio | -0.184 |
-| Total Fees | $12.00 |
-| Estimated Strategy Capacity | $210000000.00 |
-| Lowest Capacity Asset | IBM R735QTJ8XC9X |
-| Portfolio Turnover | 0.07% |
-| Drawdown Recovery | 503 |
+| End Equity | 95153.40 |
+| Net Profit | -4.847% |
+| Sharpe Ratio | -0.758 |
+| Sortino Ratio | -0.363 |
+| Probabilistic Sharpe Ratio | 0.006% |
+| Loss Rate | 75% |
+| Win Rate | 25% |
+| Profit-Loss Ratio | 0.56 |
+| Alpha | -0.034 |
+| Beta | 0.047 |
+| Annual Standard Deviation | 0.038 |
+| Annual Variance | 0.001 |
+| Information Ratio | -0.786 |
+| Tracking Error | 0.182 |
+| Treynor Ratio | -0.603 |
+| Total Fees | $25.59 |
+| Estimated Strategy Capacity | $2300000.00 |
+| Lowest Capacity Asset | BNO UN3IMQ2JU1YD |
+| Portfolio Turnover | 0.14% |
+| Drawdown Recovery | 32 |
 <!-- AQ:BACKTEST_FULL_STATS_END -->
 
 </details>
@@ -488,7 +635,7 @@ genuinely halted trading exactly as designed.
 
 ## Test Suite
 
-507 tests, one file per source module, run via:
+828 tests, one file per source module, run via:
 
 ```powershell
 aq test
@@ -515,42 +662,133 @@ instead, without waiting on a PyPI release:
 pip install -e .
 ```
 
-Either way, `aq --help` gives the full command list:
+Either way, `aq --help` gives the full command list. Every command except
+`aq trade-lock` and `aq fetch` is a thin `subprocess` wrapper around a
+command already documented elsewhere in this README:
 
+#### `aq train`
+```text
+aq train [--dataset-only|--init-only|--experts-only|--gating-only|--multitask-only|--sequence-only|--walk-forward] [--step-days N] [--mode rolling|expanding]
+```
+Runs `train.py`: builds the dataset and trains the baseline + expert
+models. `--gating-only` trains just the learned gating blend
+(`train_gating.py`) and installs it straight into active `ml/`,
+mirroring what `--experts-only` already does for the expert models — see
+`moe/README.md`. `--multitask-only` does the same for the joint
+direction+magnitude+volatility model (`train_multitask.py`) — see
+`inference/README.md`/`risk/README.md`. `--sequence-only` does the same
+for the Phase 2 causal-TCN sequence encoder (`train_sequence.py`) — see
+`inference/README.md`. `--walk-forward` (Phase 4 of the 5/10 -> 9/10
+roadmap) wraps `python train.py --walk-forward`, running the dataset-build
++ training pipeline once per rolling/expanding window instead of once on
+the fixed `phase1.windows` — diagnostic only, never touches active `ml/`
+(each window writes to `ml/versions/<run-id>/window_<i>/`, same as
+`--candidate`); `--step-days`/`--mode` override
+`phase_v2.retraining.walk_forward`'s `step_days`/`mode` defaults — see
+`retraining/README.md`'s "Walk-forward retraining is a separate, scheduled
+mechanism" section.
+
+#### `aq test`
 ```text
-aq train [--dataset-only|--init-only|--experts-only|--gating-only]
 aq test
+```
+Runs the pytest suite and refreshes this README's test badge.
+
+#### `aq backtest`
+```text
 aq backtest
+```
+Runs `lean backtest .` and refreshes this README's [Backtest Results](#backtest-results) section.
+
+#### `aq report`
+```text
 aq report <backtest-folder> <result-id>
+```
+Generates Lean's own HTML backtest report (trade blotter, standard Lean
+charts) at `backtests/<backtest-folder>/report.html`.
+
+#### `aq api`
+```text
 aq api
+```
+Starts the FastAPI monitoring server on `:8001`.
+
+#### `aq webui`
+```text
 aq webui
+```
+Starts the webui dev server (`npm run dev`).
+
+#### `aq docker`
+```text
 aq docker up [--lean|--all]
 aq docker build
+```
+`up` starts local infrastructure (default: Redis + PostgreSQL only).
+`build` rebuilds the `aether-quant` app image.
+
+#### `aq config`
+```text
 aq config [get <dotted.key>|set <dotted.key> <value>|keys [<dotted.prefix>]]
+```
+Reads or edits `config.json` directly, no manual file editing needed.
+Bare `aq config` pretty-prints the whole file; `aq config keys
+[<dotted.prefix>]` lists every leaf key path (handy for finding the right
+key in a deeply nested file); `aq config get <dotted.key>` prints one
+value (scalar, or a whole nested section as JSON); `aq config set
+<dotted.key> <value>` writes it — the value is parsed as JSON first (so
+`true`/`123`/`0.5`/`["a","b"]` become their real types automatically),
+falling back to a plain string otherwise. Every `set` backs up the
+previous file to `config.json.bak` first and prints old → new so a
+mistake is immediately visible; changing a value's type (e.g. bool →
+string) prints a warning but still writes it, since this command
+intentionally gives full access to every key, not just a safe subset.
+
+#### `aq lean`
+```text
 aq lean [get <dotted.key>|set <dotted.key> <value>|keys [<dotted.prefix>]]
+```
+The exact same `get`/`set`/`keys` tool as `aq config`, just pointed at
+`lean.json` (the QuantConnect Lean CLI's own config file — broker
+credentials, environments, data providers) instead. `aq lean set
+ib-trading-mode live`, `aq lean keys environments.live-paper`, etc.
+
+#### `aq retrain`
+```text
 aq retrain <plan|train|validate|backtest|commit|promote|rollback|status> [...]
+```
+Dispatches to `python -m retraining.orchestrator <stage> ...` for a
+single manual pipeline stage.
+
+#### `aq trade-lock`
+```text
 aq trade-lock --on|--off|--auto|--status
+```
+Manually overrides `main.py`'s sticky total-drawdown trade lock (see
+`development/v2_architecture.md`'s Manual Trade-Lock Override Contract).
+`--off` deliberately clears an otherwise-permanent lock; `--auto` returns
+to fully automatic behavior.
+
+#### `aq fetch`
+```text
 aq fetch <crypto|stock> --ticker <TICKER> --start <YYYY-MM-DD> --end <YYYY-MM-DD> [--apply]
+```
+Fetches historical OHLCV from Yahoo Finance for a ticker that isn't in
+`config.json` yet, formats it into Lean's zip/CSV convention, and writes
+it to the right spot under `data/` (`data/crypto/coinbase/daily/<ticker>_trade.zip`
+or `data/equity/usa/daily/<ticker>.zip`). On `--apply`, it also appends a
+new asset block to `config.json`'s `phase1.universe.assets[]` — no manual
+editing needed. Dry run by default (no `--apply`): reports what would
+happen, writes nothing. Never runs `train.py` itself — once applied, run
+`python train.py --dataset-only` (then `python train.py` when ready)
+yourself to actually train on the new ticker. `crypto`/`stock` today; a
+`derivative` asset class is planned for V3.
+
+#### `aq status`
+```text
 aq status
 ```
-
-Every command except `aq trade-lock` and `aq fetch` is a thin `subprocess`
-wrapper around a command already documented elsewhere in this README:
-
-- **`aq train`** — runs `train.py`: builds the dataset and trains the baseline + expert models. `--gating-only` trains just the learned gating blend (`train_gating.py`) and installs it straight into active `ml/`, mirroring what `--experts-only` already does for the expert models — see `moe/README.md`.
-- **`aq test`** — runs the pytest suite and refreshes this README's test badge.
-- **`aq backtest`** — runs `lean backtest .` and refreshes this README's [Backtest Results](#backtest-results) section.
-- **`aq report <backtest-folder> <result-id>`** — generates Lean's own HTML backtest report (trade blotter, standard Lean charts) at `backtests/<backtest-folder>/report.html`.
-- **`aq api`** — starts the FastAPI monitoring server on `:8001`.
-- **`aq webui`** — starts the webui dev server (`npm run dev`).
-- **`aq docker up [--lean|--all]`** — starts local infrastructure (default: Redis + PostgreSQL only).
-- **`aq docker build`** — rebuilds the `aether-quant` app image.
-- **`aq config`** — reads or edits `config.json` directly, no manual file editing needed. Bare `aq config` pretty-prints the whole file; `aq config keys [<dotted.prefix>]` lists every leaf key path (handy for finding the right key in a deeply nested file); `aq config get <dotted.key>` prints one value (scalar, or a whole nested section as JSON); `aq config set <dotted.key> <value>` writes it — the value is parsed as JSON first (so `true`/`123`/`0.5`/`["a","b"]` become their real types automatically), falling back to a plain string otherwise. Every `set` backs up the previous file to `config.json.bak` first and prints old → new so a mistake is immediately visible; changing a value's type (e.g. bool → string) prints a warning but still writes it, since this command intentionally gives full access to every key, not just a safe subset.
-- **`aq lean`** — the exact same `get`/`set`/`keys` tool as `aq config`, just pointed at `lean.json` (the QuantConnect Lean CLI's own config file — broker credentials, environments, data providers) instead. `aq lean set ib-trading-mode live`, `aq lean keys environments.live-paper`, etc.
-- **`aq retrain <stage>`** — dispatches to `python -m retraining.orchestrator <stage> ...` for a single manual pipeline stage.
-- **`aq trade-lock --on|--off|--auto|--status`** — manually overrides `main.py`'s sticky total-drawdown trade lock (see `development/v2_architecture.md`'s Manual Trade-Lock Override Contract). `--off` deliberately clears an otherwise-permanent lock; `--auto` returns to fully automatic behavior.
-- **`aq fetch <crypto|stock> --ticker <TICKER> --start <YYYY-MM-DD> --end <YYYY-MM-DD> [--apply]`** — fetches historical OHLCV from Yahoo Finance for a ticker that isn't in `config.json` yet, formats it into Lean's zip/CSV convention, and writes it to the right spot under `data/` (`data/crypto/coinbase/daily/<ticker>_trade.zip` or `data/equity/usa/daily/<ticker>.zip`). On `--apply`, it also appends a new asset block to `config.json`'s `phase1.universe.assets[]` — no manual editing needed. Dry run by default (no `--apply`): reports what would happen, writes nothing. Never runs `train.py` itself — once applied, run `python train.py --dataset-only` (then `python train.py` when ready) yourself to actually train on the new ticker. `crypto`/`stock` today; a `derivative` asset class is planned for V3.
-- **`aq status`** — shows `git status`.
+Shows `git status`.
 
 ## Release Process
 

{aether_quant-0.6.0 → aether_quant-0.8.0}/aether_quant.egg-info/PKG-INFO

@@ -1,5 +1,5 @@
 Metadata-Version: 2.4
 Name: aether-quant
-Version: 0.6.0
+Version: 0.8.0
 Summary: Aether Quant convenience CLI
 Requires-Python: >=3.10