utkarshp845 · utkarshp845 · Jul 12, 2026 · Jul 12, 2026
diff --git a/OPERATIONS.md b/OPERATIONS.md
@@ -71,6 +71,8 @@ Small-account BTC paper rehearsal:
 python -m bot.profile_runner paper research btc
 ```
 
+Both `spy` (QQQ) and `btc` research runs now use $150 starting equity to match the real account.
+
 Outputs:
 
 - `reports/research_latest.md`
@@ -111,26 +113,36 @@ Useful optional env controls:
 
 ## Normal Bot Run
 
-Run one BTC bot cycle:
+Run one bot cycle (equity profile, the default live deployment target):
 
 ```powershell
 docker compose run --rm paper
 docker compose run --rm trade
 ```
 
+BTC variants:
+
+```powershell
+docker compose run --rm paper-btc
+docker compose run --rm trade-btc
+```
+
 Direct profile runner equivalents:
 
 ```powershell
+python -m bot.profile_runner paper trade spy
+python -m bot.profile_runner live trade spy
 python -m bot.profile_runner paper trade btc
 python -m bot.profile_runner live trade btc
 ```
 
 Profile-specific validation:
 
 ```powershell
+python -m bot.profile_runner paper validate spy
+python -m bot.profile_runner live validate spy
 python -m bot.profile_runner paper validate btc
 python -m bot.profile_runner live validate btc
-python -m bot.profile_runner live validate spy
 ```
 
 Current runtime defaults:
@@ -141,15 +153,15 @@ Current runtime defaults:
 - end-of-day flattening starts `5` minutes before the close by default; override with `FLATTEN_BEFORE_CLOSE_MINUTES`
 - `paper` and `trade` select separate Alpaca key pairs from `.env` when `ALPACA_PAPER_*` and `ALPACA_LIVE_*` variables are set
 - BTC paper writes runtime artifacts under `runtime/paper_btc`, BTC live under `runtime/live_btc`
-- The BTC paper profile mirrors the live BTC signal filters, with larger paper sizing for rehearsal
+- `config/paper_btc.env` is an exact mirror of `config/live_btc.env` (same sizing, same filters) so paper fills validate the strategy that actually runs live
 - `reports/monitor_latest.md` includes 24h/7d rejection counts, near-miss entry bars, and latest filter metrics for diagnosing quiet BTC live periods
-- `config/live_spy.env` enables fractional, long-only SPY sizing for small live accounts; replay it before relying on it live.
+- `config/live_spy.env` trades QQQ on hourly bars (fractional, long-only, ~90% notional) and is the recommended default live profile for a small account; see `docs/strategy_revamp_2026-07.md` for the replay evidence. Replay any change to it before relying on it live.
 
 ## Notes
 
 - Start Docker Desktop before using the commands above.
 - Keep real Alpaca keys only in the local untracked `.env`.
-- Alpaca's paper account balance still needs to be adjusted in the dashboard. The repo now mirrors that target by using roughly `$250` as the paper research starting equity.
+- Alpaca's paper account balance still needs to be adjusted in the dashboard. The repo now mirrors that target by using `$150` as the paper research starting equity (see `RESEARCH_STARTING_EQUITY` in `config/paper_spy.env` and `config/paper_btc.env`).
 - Review `reports/monitor_latest.md` after each trading session if you want a concise explanation of what the bot did and why.
 
 ## EC2 Deploy

diff --git a/README.md b/README.md
@@ -1,6 +1,6 @@
 # Trading Bot
 
-Trend-following trading bot for Alpaca, supporting both intraday equities (SPY) and 24/7 crypto (BTC/USD). Uses a trend-confirmed moving-average strategy with ADX, ATR, volume, momentum, and regime filters, layered risk controls, and multiple exit mechanisms.
+Trend-following trading bot for Alpaca. The default live/paper deployment is an hourly QQQ trend strategy sized for a small account; a defensive BTC/USD profile is also available. Uses a trend-confirmed moving-average strategy with ADX, ATR, volume, momentum, and regime filters, layered risk controls, and multiple exit mechanisms.
 
 It currently:
 
@@ -133,7 +133,7 @@ Run the walk-forward optimizer:
 docker compose run --rm optimize
 ```
 
-Direct BTC live tuning equivalent:
+BTC equivalent:
 
 ```powershell
 python -m bot.profile_runner live optimize btc
@@ -156,88 +156,35 @@ Main outputs:
 - `reports/optimize_latest.md`
 - `reports/optimize_latest.json`
 
-For BTC live tuning, the optimizer compares candidates against the loaded live baseline over the same bars. A candidate is marked accepted only if it clears the full replay, walk-forward, baseline-improvement, and 2x-slippage checks in the report.
+The optimizer compares candidates against the loaded live baseline over the same bars, for either market. A candidate is marked accepted only if it clears the full replay, walk-forward, baseline-improvement, and 2x-slippage checks in the report.
 
 ## Deployment
 
-The bot can be deployed to AWS EC2 using GitHub Actions.
-
-### Prerequisites
-
-1. AWS EC2 instance with Docker installed
-2. AWS ECR repository for the Docker images
-3. IAM user with ECR and EC2 access
-4. SSH key pair for EC2 access
-
-### Setup
-
-1. **Create AWS ECR Repository:**
-   ```bash
-   aws ecr create-repository --repository-name trading-bot --region your-region
-   ```
-
-2. **Configure GitHub Secrets:**
-   Add the following secrets to your GitHub repository:
-   - `AWS_ACCESS_KEY_ID`: Your AWS access key
-   - `AWS_SECRET_ACCESS_KEY`: Your AWS secret key
-   - `AWS_ACCOUNT_ID`: Your AWS account ID
-   - `AWS_REGION`: AWS region (e.g., us-east-1)
-   - `EC2_HOST`: EC2 instance public IP or DNS
-   - `EC2_USER`: SSH username (usually 'ec2-user' or 'ubuntu')
-   - `EC2_SSH_KEY`: Private SSH key for EC2 access
-
-4. **Setup EC2 Instance:**
-   - Install Docker and AWS CLI
-   - Configure AWS CLI with credentials that have ECR pull permissions
-   - Clone your repository
-   - Copy `.env` file with your Alpaca credentials to the project directory
-   - Ensure the data, logs, and reports directories exist and are writable
-   - Make sure Docker daemon is running
-   - **Set up cron job for periodic execution:**
-     ```bash
-     # Edit crontab
-     crontab -e
-
-     # Add line to run every 5 minutes during market hours (9:30 AM - 4:00 PM ET, Monday-Friday)
-     # Note: Adjust timezone as needed
-     */5 9-15 * * 1-5 docker run --rm --env-file /path/to/trading-bot/.env -v /path/to/trading-bot/data:/app/data -v /path/to/trading-bot/logs:/app/logs -v /path/to/trading-bot/reports:/app/reports trading-bot:latest
-     ```
-
-     Or create a simple run script:
-     ```bash
-     #!/bin/bash
-     docker run --rm \
-       --env-file .env \
-       -v $(pwd)/data:/app/data \
-       -v $(pwd)/logs:/app/logs \
-       -v $(pwd)/reports:/app/reports \
-       trading-bot:latest
-     ```
-
-     Use the provided `run.sh` script for this purpose.
-
-5. **Manual Deployment:**
-   ```bash
-   # On EC2 instance
-   ./deploy.sh
-   ```
-
-### Automatic Deployment
-
-Push to the `main` branch to trigger automatic deployment via GitHub Actions.
+The bot deploys to a plain EC2 instance via GitHub Actions: the workflow at
+`.github/workflows/deploy-ec2.yml` rsyncs the repo over SSH, builds the
+Docker image on the instance, and installs a cron job — no container
+registry or AWS IAM setup involved. Full setup steps, required secrets, and
+verification commands are in `docs/github_actions_ec2.md`.
+
+It triggers automatically on pushes to `master`, or manually from the
+Actions tab with a chosen profile (`live`/`paper`) and market
+(`spy`/`btc`, defaults to `spy`).
 
 ### Security Notes
 
-- Use IAM roles on EC2 instead of access keys when possible
 - Restrict EC2 security groups to only allow SSH from your IP
 - Generate a dedicated SSH key pair for deployment
-- Store sensitive credentials only in GitHub secrets and the .env file on EC2
+- Store sensitive credentials only in GitHub secrets and the `.env` file on EC2
 - Regularly rotate API keys and SSH keys
 
+## Project Structure
+
 - `bot/` - trading logic, broker integration, storage, reporting
+- `config/` - per-profile env files (`live_spy`, `paper_spy`, `live_btc`, `paper_btc`)
 - `data/` - SQLite database
 - `logs/` - runtime logs and CSV snapshots
 - `reports/` - generated reports
+- `docs/` - strategy audits, build log, and deployment guides
 
 ## Risk Controls
 
@@ -250,7 +197,7 @@ The bot has multiple independent safety layers:
 | Breakeven stop | `ENABLE_BREAKEVEN_STOP` | false |
 | Profit lock | `ENABLE_PROFIT_LOCK` | false |
 | Time-based stop | `MAX_BARS_IN_TRADE` | 12 bars |
-| Trend reversal exit | `REVERSAL_SIGNAL_STRENGTH_MIN` | 35 |
+| Regime-invalidation exit | `EXIT_ON_REGIME_INVALIDATION` | true |
 | Daily drawdown halt | `MAX_DAILY_DRAWDOWN_PCT` | 1% |
 | Dollar loss cap | `MAX_DAILY_LOSS` | disabled |
 | Consecutive loss halt | `MAX_CONSECUTIVE_LOSSES` | 3 |
@@ -266,7 +213,7 @@ Set `IS_CRYPTO=true` (or use `config/paper_btc.env` / `config/live_btc.env`) to
 - Bypasses NYSE market-hours check — trades 24/7
 - Orders use `TimeInForce.GTC` instead of `DAY`
 - Position sizing returns fractional quantities (e.g. `0.0005 BTC`)
-- Cron schedule should run every 5 minutes around the clock
+- Trades hourly bars like the equity profile; cron runs once an hour around the clock (see `docs/github_actions_ec2.md`)
 
 Important for small accounts: Alpaca crypto costs ~0.25% taker fee + spread per
 side (~0.6% per round trip). Replay evidence in `docs/strategy_revamp_2026-07.md`
@@ -279,19 +226,22 @@ profile for growth.
 
 ## Small Equity Accounts
 
-For a roughly `$150` SPY account, one whole SPY share is too large. Use `config/live_spy.env` or set:
+For a roughly `$150` account, one whole share of most ETFs is too large a
+chunk of the account to size or diversify sensibly. `config/live_spy.env`
+(despite the filename, it trades `QQQ`) already sets this up:
 
 - `ALLOW_FRACTIONAL_EQUITIES=true`
 - `ALLOW_SHORTS=false`
-- `POSITION_SIZING_MODE=atr_risk`
-- `MAX_POSITION_NOTIONAL_PCT` and `TARGET_POSITION_NOTIONAL_PCT` high enough to create a real fractional order, but low enough to cap exposure
+- `POSITION_SIZING_MODE=notional_cap` with `TARGET_POSITION_NOTIONAL_PCT=0.90` — near-full-notional single positions, since a $150 account can't usefully diversify anyway
+- `MAX_POSITION_NOTIONAL_PCT` above the target as a hard ceiling
 
 ## Notes
 
 - Keep real API keys only in your local `.env` — never commit them.
-- By default, the bot behaves as an intraday system: it flattens inherited overnight positions on the next session and exits before market close. Set `ALLOW_OVERNIGHT_HOLDING=true` and `FLATTEN_BEFORE_CLOSE_MINUTES=0` for crypto.
-- The SPY runners write to `runtime/paper` and `runtime/live`; BTC runners write to `runtime/paper_btc` and `runtime/live_btc`.
+- Without a profile, the raw `bot.main` default is an intraday equity system: it flattens inherited overnight positions on the next session and exits before market close. Both shipped profiles override this — `config/live_spy.env` and `config/live_btc.env` both set `ALLOW_OVERNIGHT_HOLDING=true` and `FLATTEN_BEFORE_CLOSE_MINUTES=0`, since the current strategy is a multi-day trend hold, not an intraday one.
+- The `spy`-market runners write to `runtime/paper` and `runtime/live`; `btc`-market runners write to `runtime/paper_btc` and `runtime/live_btc`.
 - Use the optimizer to rank parameter sets on walk-forward windows before going live.
 - `OPERATIONS.md` has day-to-day commands.
 - `docs/github_actions_ec2.md` covers the GitHub Actions EC2 deployment path.
+- `docs/BUILD_LOG.md` tracks strategy and infrastructure changes over time.
 - Run `scripts/validate.ps1` (Windows) or `scripts/validate.sh` (Unix) for a full local validation pass.
diff --git a/deploy/ec2/install_cron.sh b/deploy/ec2/install_cron.sh
@@ -5,7 +5,7 @@ set -Eeuo pipefail
 profile="${1:-live}"
 app_dir="${2:-${APP_DIR:-/opt/trading-bot/app}}"
 cron_tz="${CRON_TZ:-America/New_York}"
-schedule="${CRON_SCHEDULE:-*/5 * * * *}"
+schedule="${CRON_SCHEDULE:-5 * * * *}"
 docker_bin="${DOCKER_BIN:-$(command -v docker || true)}"
 job_marker="# trading-bot-${profile}"
 

diff --git a/docs/BUILD_LOG.md b/docs/BUILD_LOG.md
@@ -0,0 +1,124 @@
+# Build Log
+
+Running log of strategy and infrastructure changes to this bot. Newest entry
+first. Each entry should say what changed, why (with evidence where
+possible), and what to watch after deploying it.
+
+---
+
+## 2026-07-12 — Repo alignment checkover
+
+**What:** Swept the repo for leftover references to the old BTC-5m-default
+setup after the strategy revamp below, since several docs/scripts still
+assumed BTC was the primary deployed strategy and that bars were 5 minutes.
+
+**Found and fixed:**
+- Cron schedule (`deploy/ec2/install_cron.sh`) still defaulted to
+  `*/5 * * * *`, i.e. every 5 minutes, but every profile now trades hourly
+  bars (`TIMEFRAME_MINUTES=60` in all four `config/*.env` files). Left as-is,
+  a fresh EC2 deploy would have invoked the bot ~12x per hour for one hourly
+  decision — harmless (cooldown/pending-order guards prevent duplicate
+  orders) but wasteful API calls and log noise. Changed the default to
+  `5 * * * *` (once per hour, 5 minutes after each bar closes).
+- `docs/github_actions_ec2.md` explicitly said "the deployed profile is BTC,
+  24/7 scheduling is the correct default" — no longer true now that `spy`
+  (QQQ) is the default deploy market. Updated.
+- `OPERATIONS.md`: paper research equity doc said `$250`; both paper profiles
+  now use `$150` to match the real account. Also removed a line claiming the
+  BTC paper profile uses "larger sizing" than live — it's an exact mirror of
+  `live_btc.env` now, by design, so paper fills validate the same strategy
+  that runs live.
+- `README.md`:
+  - Opening description and Crypto section still framed BTC as the always-on
+    24/7 default with 5-minute cron; corrected.
+  - The "Deployment" section described AWS ECR, IAM roles, `run.sh`,
+    `deploy.sh`, and pushing to a `main` branch — none of which exist in this
+    repo. The actual deploy path is GitHub Actions → SSH/rsync → EC2 cron,
+    already documented in `docs/github_actions_ec2.md`. Replaced the stale
+    section with a pointer to that doc.
+  - Removed the `REVERSAL_SIGNAL_STRENGTH_MIN` row from the risk-controls
+    table — grepping `bot/` shows no code reads that variable (only test
+    fixtures set it). It's vestigial from an earlier strategy iteration;
+    flagged separately rather than touched here since removing it fully
+    means editing `tests/test_research_replay.py`, which is out of scope for
+    a docs pass.
+- Added "superseded" banners to `docs/strategy_audit_current.md` and
+  `docs/live_account_path_100usd.md` (both dated 2026-04-14, both centered on
+  an intraday SPY strategy that no longer exists) pointing at
+  `docs/strategy_revamp_2026-07.md`. Kept their original content intact as a
+  historical record rather than rewriting them.
+
+**Not changed:** `bot/main.py`'s hardcoded `SYMBOL` default of `SPY` and
+`TIMEFRAME_MINUTES` default of `5` — these are fallbacks for running the bot
+without a profile at all (e.g. `python -m bot.main` with a bare `.env`).
+Every shipped profile overrides both, so this doesn't affect deployed
+behavior, but it means an operator who skips the profile runner entirely
+still gets the old defaults. Worth a follow-up if that path is ever used for
+real.
+
+**Verification:** `pytest` (66 passed), `bot.validate_runtime`,
+`bot.validate_profile_env` for all four profiles — all still pass; this
+entry was docs/config/deploy-script only, no strategy logic touched.
+
+---
+
+## 2026-07-12 — Strategy revamp: QQQ hourly trend replaces BTC 5m scalp
+
+**What:** Investigated why the live BTC bot ($150 account, 5-minute bars) was
+making no valuable trades, then replaced the live deployment strategy.
+
+**Root cause (full writeup: `docs/strategy_revamp_2026-07.md`):**
+- The live filter stack required ~14 conditions to align at once; replayed
+  over 120 days of real bars it produced 2 trades, both losses.
+- Position sizing capped trades at ~$45; Alpaca crypto's ~0.6% round-trip
+  friction exceeds the gross P&L of most 5-minute trades, so even a trade
+  that fired couldn't clear its own costs.
+- BTC fell ~46% over the trailing year and Alpaca doesn't support shorting
+  crypto, so a long-only bot had no tailwind.
+- Directly tested the "more, smaller trades" hypothesis: a loosened
+  high-frequency BTC config made 184 trades/year and lost $95, ~$99 of it
+  pure fee friction. Activity was the cost, not the fix, on this venue/size.
+
+**Changes:**
+- `config/live_spy.env` + `config/paper_spy.env`: switched to `SYMBOL=QQQ`,
+  hourly bars, long-only trend-following, ~90% notional fractional sizing,
+  daily-EMA(20) regime gate, wide trailing exits, multi-day holds. Replay
+  2023-08→2026-07 at $150 start: net +$55.3 (+37%), profit factor 1.76, win
+  rate 45%, max drawdown 6.7%, positive every calendar year, stable under 3x
+  slippage stress.
+- `config/live_btc.env` + `config/paper_btc.env`: kept BTC live but made it
+  defensive — hourly bars, strict 4h-EMA(120) uptrend gate
+  (`REGIME_MIN_SLOPE_PCT=0.008`). Zero trades in the 2025-26 bear-year replay
+  (capital preserved through a 46% market decline); mildly positive
+  (+$3, PF 1.14) in the 2024-25 bull-year replay.
+- `bot/trade_controls.py::sync_replay_day`: fixed the replay harness to reset
+  `consecutive_losses` on ET day rollover, matching `bot/store.py`'s live
+  behavior. Previously the replay never reset this counter, so any backtest
+  that hit `MAX_CONSECUTIVE_LOSSES` stopped trading for the rest of the
+  replay — every prior research report in this repo undercounted trades and
+  is unreliable.
+- `bot/broker_alpaca.py::get_recent_bars`: lookback window now scales with
+  `timeframe_minutes` and asset session hours instead of a fixed 7 days.
+  The old fixed window could never warm up hourly/daily-regime indicators
+  for equities (only ~5 session-days per 7 calendar days).
+- `bot/profile.py`: profile env file values now take precedence over market
+  defaults (`SYMBOL`, `ALLOW_OVERNIGHT_HOLDING`, `FLATTEN_BEFORE_CLOSE_MINUTES`)
+  instead of being silently overwritten by them — this is what let
+  `config/live_spy.env` actually set `SYMBOL=QQQ` and hold overnight.
+- `docker-compose.yml`, `deploy/ec2/deploy_remote.sh`,
+  `.github/workflows/deploy-ec2.yml`: default deploy market switched from
+  `btc` to `spy`. BTC remains reachable via `trade-btc` / `paper-btc` /
+  `research-btc` compose services.
+- Tests: updated profile-contract assertions in `tests/test_profile.py` to
+  match the new defaults, added regression coverage for the loss-streak
+  reset and profile-env precedence.
+
+**Verification:** `pytest` (66 passed), `bot.validate_runtime`,
+`bot.validate_profile_env` for all four profiles, and a replay of the exact
+committed `config/live_spy.env` / `config/live_btc.env` files against real
+historical bars (QQQ, SPY, BTC bull year, BTC bear year) to confirm the
+numbers in `docs/strategy_revamp_2026-07.md` match what's actually deployed.
+
+**What to watch after deploy:** run `docker compose run --rm paper` for a
+week or two before trusting live fills; compare paper fills against the
+replay assumptions (slippage, fill price) before increasing size.