BacktestGate

Offline trust gate for TradingView Pine Script backtests.

BacktestGate reads a Pine strategy, TradingView CSV exports, and the actual run assumptions. It then produces a deterministic PASS, WARN, or BLOCK report with line-level evidence and small, explainable robustness checks.

Status: alpha (pre-release). It is not a strategy generator, profitability score, certification, or investment recommendation.

What the demo shows

strategy.pine + trades.csv + performance.csv + manifest.json
                              │
                              ▼
          code integrity / execution realism
          evidence sufficiency / robustness
                              │
                              ▼
                    PASS / WARN / BLOCK

The tool is read-only and offline. It does not execute Pine, connect to TradingView, call an LLM, upload source code, or place orders.

Requirements

• Python 3.9+
• No third-party runtime dependencies

Run the fixtures

git clone https://github.com/Jiangki/backtestgate.git
cd backtestgate
python3 -m pip install -e .

# Clean sample: exit 0
python3 -m backtestgate audit fixtures/pass

# Fragile assumptions: WARN, exit 0
python3 -m backtestgate audit fixtures/warn

# Confirmed future leak: BLOCK, exit 1
python3 -m backtestgate audit fixtures/block

# Generate an offline visual report
python3 -m backtestgate audit fixtures/block --output block-report.html

Open block-report.html in any browser. No server is required.

Expected fixture results

Fixture	Gate	Exit	Representative evidence
fixtures/pass	PASS	0	confirmed request.security value, non-zero costs, 40 trades / 390 days
fixtures/warn	WARN	0	tick/fill recalculation, zero costs, Heikin Ashi, small/unstable sample
fixtures/block	BLOCK	1	PINE001 at strategy.pine:13: lookahead_on without a confirmed expression offset

For stricter CI behavior:

python3 -m backtestgate audit fixtures/warn --fail-on warn
# exit 1

Audit your own export

Initialize a local audit directory:

python3 -m backtestgate init my-audit

The command refuses to overwrite a non-empty directory. It creates a manifest with a random validation_id and a local checklist:

my-audit/
├── README.txt
└── manifest.json

Then:

1. Save the exact Pine source used for the run as my-audit/strategy.pine.
2. In TradingView Strategy Report, export the List of trades tab as my-audit/trades.csv.
3. Export the Performance tab as my-audit/performance.csv.
4. Replace every REPLACE_ME and assumption in manifest.json with the actual Properties-panel values.
5. Run:

python3 -m backtestgate audit my-audit --output my-report.html

The MVP accepts English TradingView headers. It identifies columns such as Type, Date and time, Qty/Position size, and Net P&L <currency>. Only Exit rows are treated as closed trades.

manifest.json

{
  "validation_id": "d152cf18-8c83-4dfe-8612-0b8104f07bd2",
  "symbol": "BINANCE:BTCUSDT",
  "timeframe": "1h",
  "chart_type": "standard",
  "commission_value": 0.1,
  "slippage_ticks": 2,
  "bar_magnifier": true,
  "limit_fill_assumption_ticks": 1,
  "calc_on_every_tick": false,
  "calc_on_order_fills": false,
  "stress_cost_per_trade": 2.0
}

stress_cost_per_trade is an additional round-trip cost scenario applied to each closed trade. It does not claim to infer the correct broker cost.

Why is the manifest required? Pine defaults can be overridden in TradingView's Properties panel. Source code alone therefore cannot prove which assumptions produced the exported result.

Keep validation_id unchanged when fixing and rerunning an audit. It is a random identifier, not an account, and allows a voluntary follow-up receipt to demonstrate repeat use.

Gate semantics

Gate	Meaning	Default exit
PASS	No issue found by the rules covered in this version	0
WARN	Review execution, evidence, or robustness before relying on the run	0
BLOCK	A high-confidence correctness issue was found	1

Invalid/missing input exits 2. PASS deliberately means PASS CHECKS, not “safe” and never “will make money.”

Rules in v0.2

Code integrity

ID	Gate	Check
PINE000	BLOCK	Missing strategy() declaration
PINE001	BLOCK	request.security(... lookahead_on) without a confirmed-history offset in the requested expression
PINE002	WARN	calc_on_every_tick=true
PINE003	WARN	calc_on_order_fills=true
PINE004	WARN	timenow, varip, or realtime-only state appears
PINE005	WARN	Missing/old Pine version

PINE001 understands the official safe shape request.security(..., close[1], lookahead = barmerge.lookahead_on) and does not block it.

Execution realism

ID	Gate	Check
EXEC001	WARN	Zero commission
EXEC002	WARN	Zero slippage
EXEC003	WARN	Stop/limit orders without Bar Magnifier
EXEC004	WARN	Non-standard chart type
EXEC005	WARN	Limit orders with zero fill-verification ticks

Evidence and result robustness

ID	Gate	Check
DATA001	WARN	Trade P&L does not reconcile with Performance Summary
EVID001	WARN	Fewer than 30 closed trades
EVID002	WARN	Less than 180 calendar days
ROBUST001	WARN	Five best trades contribute at least half of net profit
ROBUST002	WARN	Removing the best trade removes all profit
ROBUST003	WARN	Additional-cost stress makes net profit non-positive
ROBUST004	WARN	One chronological half is non-profitable

The 30-trade and 180-day values are warning floors, not universal statistical sufficiency claims.

Output formats

python3 -m backtestgate audit fixtures/warn --format json
python3 -m backtestgate audit fixtures/warn --format markdown
python3 -m backtestgate audit fixtures/warn --format html

python3 -m backtestgate audit fixtures/warn --output report.json
python3 -m backtestgate audit fixtures/warn --output report.md
python3 -m backtestgate audit fixtures/warn --output report.html

The output file format is inferred from its suffix.

Privacy-safe validation receipt

Full reports intentionally include code evidence and financial metrics for local diagnosis. Do not attach them to a public issue unless you intend to disclose that data.

For public feedback, generate a minimized receipt instead:

python3 -m backtestgate audit my-audit \
  --share-output validation-receipt.json

The receipt contains the gate, dimension statuses, rule IDs, broad sample-size buckets, tool version, and random validation_id. It omits Pine source, symbols, timeframes, paths, P&L, prices, and individual trades. The CLI never uploads it; inspect the JSON before sharing.

See VALIDATION.md for how to report feedback safely.

Important limits

• BacktestGate does not execute Pine and cannot prove full script semantics.
• The Pine scanner is deliberately conservative; it is not a complete parser.
• Realized drawdown is reconstructed from closed trades, not intratrade equity.
• A single exported run cannot establish Probability of Backtest Overfitting or a Deflated Sharpe Ratio.
• Static checks cannot rule out every repainting or future-data pattern.
• Report findings require human review.

Not yet included:

• localized/non-English TradingView export headers
• pyramiding and partial-exit reconciliation
• multiple-run PBO/Deflated Sharpe analysis
• bootstrap/Monte Carlo reports
• comparison between strategy versions
• TradingView alert or broker-fill drift monitoring
• automatic mapping of every locale/export layout

Development

python3 -m unittest discover -s tests -v

See CONTRIBUTING.md before changing rule severity or the share-receipt schema.

License

MIT — see LICENSE.