Release 0.1.1

docs/OPERATIONS.md

@@ -113,7 +113,7 @@ docker compose ps                            # bot should become "healthy" withi
 
 The image contains the code. Only `user_data/` is mounted from the host:
 config, strategies, data, databases, logs and the audit trail. To pin a
-release, set `RIDGELINE_VERSION=0.1.0` (in `.env` or the shell). The default
+release, set `RIDGELINE_VERSION=0.1.1` (in `.env` or the shell). The default
 is `latest`.
 
 Data and the strategy's pair list (see the README's Data section) can be

docs/releases/v0.1.1.md

@@ -0,0 +1,29 @@
+A research fix and an overfitting check. The app behaves the same as 0.1.0; upgrading keeps your `user_data`.
+
+**Paper money only.** The panel runs Freqtrade in dry-run mode and cannot start a live bot. No strategy is eligible for real money yet.
+
+## Install
+
+- **Windows:** download `Ridgeline-Setup-0.1.1.exe` below and run it.
+  - Installing over 0.1.0 keeps your private config, paper sessions, data and audit trail.
+  - The installer isn't code-signed, so SmartScreen may warn: choose *More info → Run anyway*.
+- **Docker:** `RIDGELINE_VERSION=0.1.1 docker compose pull && docker compose up -d`.
+- **From source:** `pip install -r requirements.txt`.
+
+## What changed
+
+- **Fixed: tuning leaked hindsight.** `research.py hyperopt` was tuning on `config.json`'s live pair list (today's top coins). It now tunes on the point-in-time pairs at the start of the in-sample period, like every backtest does.
+- **New candidate, and a negative result.** `TrendRider4hPIT` is TrendRider4h re-tuned that way (300 trials, in-sample only). It overfits:
+
+  | | In-sample | Out-of-sample | Profit factor | Walk-forward | Gates |
+  |---|---|---|---|---|---|
+  | TrendRider4h | +5.6% | **+17.3%** | **1.43** | **64%** | **8 / 9** |
+  | TrendRider4hPIT | +45.6% | +13.8% | 1.33 | 50% | 7 / 9 |
+
+  The in-sample gain didn't carry over to new data, so TrendRider4h keeps its original parameters and stays the lead candidate.
+- **Updated research:**
+  - Gates re-run for all five strategies at 1,000 and 65 USDT (≈100 AUD).
+  - `RESEARCH.md` updated.
+  - The case study (attached) has a new section 5.7 and covers five strategies.
+
+Backtests are not a promise of future results. This is software, not financial advice.

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ridgeline"
-version = "0.1.0"
+version = "0.1.1"
 description = "Freqtrade crypto bot with paper trading, point-in-time research gates, risk controls and reporting"
 requires-python = ">=3.11"
 

ridgeline/__init__.py

@@ -3,7 +3,7 @@
 from pathlib import Path
 
 # Keep in step with pyproject.toml; release tags are v<version>.
-__version__ = "0.1.0"
+__version__ = "0.1.1"
 
 ROOT = Path(__file__).resolve().parent.parent
 USERDIR = ROOT / "user_data"