Sika

Sika is an in-progress, research-validated trade-planning system for XAUUSDm on an Exness Standard account. Its intended output is a discretionary trade plan: direction, entry, stop-loss, take-profit, risk/reward, expiry, and a calibrated confidence estimate. Telegram delivery is planned after the research and paper-trading gates pass.

Sika is currently a read-only market-data system. It cannot place, modify, or close trades, and it does not yet produce a trade call. Treat every current command as data collection or validation tooling, not trading advice.

Current status

The MT5-to-research data boundary is working end to end:

• the terminal is restricted to the agreed XAUUSDm Exness demo context;
• MQL5 scripts export completed M1, M15, and H1 bars without order access;
• Python validators check identity, ranges, ordering, prices, and checksums;
• the five-year backfill covers October 2021 through September 2026; and
• gap reconciliation produces explicit session exclusions instead of filling or ignoring missing data.

The dataset remains fail-closed. The next milestone is a canonical dataset builder that proves excluded sessions cannot enter research. Model development, confidence calibration, Telegram delivery, and trade execution have not begun.

The authoritative project documents are:

• Decision journal — plain-language decisions, evidence, current state, and next steps.
• Research contract — strategy, risk, evaluation, and promotion rules.
• Architecture — system boundaries and reproducibility decisions.
• MT5 operations — Wine/MT5 setup, export, validation, and failure recovery.
• Contributing — the local quality workflow and review rules.

Product contract

When implemented, a trade call must contain all of the following or resolve to NO_TRADE:

• exact instrument and long/short direction;
• a bounded entry rule rather than an untradeable daily prediction;
• stop-loss, take-profit, and explicit risk/reward;
• a session-aware expiry;
• calibrated confidence for that exact setup; and
• the assumptions needed to reproduce the call.

The initial operating window is 08:00–16:00 New York time. New M15 candidates may be created from 08:15 through 14:00, using H1 context; the final two hours are for monitoring existing plans. Execution remains manual until a separate, explicitly reviewed execution phase is approved.

Repository layout

config/                 versioned research inputs and closure calendars
docs/                   architecture, research, and operating documentation
mt5/Scripts/            read-only MQL5 exporters and probes
src/sika/               active Python package
tests/                  safety and data-contract regression tests
archive/                frozen predecessor source; excluded from active tooling
data/, logs/, models/    ignored local outputs, with tracked placeholders only

The predecessor daily-direction prototype is frozen under archive/daily_direction_v0. It is not part of the active package, dependency lock, test suite, or product claims.

Development setup

Prerequisites:

• Python 3.12.9
• uv
• MT5 only for terminal-facing workflows

Create the locked Linux development environment:

uv sync --locked

On native 64-bit Windows, install the optional MetaTrader5 bridge as well:

uv sync --locked --extra windows-mt5

Run the focused regression suite and static checks:

uv run python -m unittest discover -s tests -v
uvx --from ruff==0.15.20 ruff check .

The MQL5 sources must additionally compile in the supported MetaEditor build with zero errors and zero warnings. Compiled .ex5 files are local build artifacts and are not committed.

Data workflow

The detailed operator procedure is in the MT5 runbook. The primary validation commands are:

uv run sika-mt5-validate-export path/to/export_manifest.json

uv run sika-mt5-validate-backfill path/to/backfill \
  --output data/xauusdm-backfill-audit.json

uv run sika-mt5-reconcile-gaps path/to/backfill \
  --calendar config/market_data/xauusdm-closures-v1.json \
  --output data/xauusdm-gap-reconciliation.json

Important nonzero statuses are deliberate safety outcomes:

• exit 3: structurally valid backfill still has gaps requiring review;
• exit 4: reconciliation succeeded, but downstream research must enforce the reported quarantine and exclusion dates.

Never synthesize bars or add a closure date merely to make these commands pass.

Configuration and logs

Research assumptions live in reviewed, version-controlled files under config/. Operational logging is the only environment-configured behavior:

export SIKA_LOG_LEVEL=INFO
export SIKA_LOG_FORMAT=json
export SIKA_LOG_FILE=logs/sika.jsonl

Supported levels are DEBUG, INFO, WARNING, ERROR, and CRITICAL; supported formats are text and json. Logs rotate at 5 MiB with five backups. JSON output contains a timestamp, level, logger, message, and approved event name. Arbitrary record fields are not serialized, which reduces accidental credential leakage.

Local data, reports, logs, compiled MQL5 binaries, models, virtual environments, and .env files are ignored. Commit schemas, calendars, source, and tests—not broker credentials, account numbers, Telegram tokens, or generated datasets.

Safety notice

Trading leveraged products can result in substantial loss. Passing software tests proves only that the code follows its declared contracts; it does not prove profitability or suitability for live trading.