1 · DATADownload/merge master 1M history or upload a proxy dataset for validation. Stored on Railway volume.
2 · RESEARCHRun protected CONTROL research and truth audit.
3 · EVOLUTIONRun historical challengers against the stored data.
4 · REVIEWInspect tournament, stress test, Shadow Pine and evidence.
Market Data + Research
Databento 1M is the master feed. EXEC stays 5M. 1M is used as a precision observation layer around the protected white line.
Checking persistent master data…
AI Research Brain
Loading AI switch…
OFF = deterministic Python research only. ON = GPT may analyze the completed evidence and propose challengers. Evolution OS remains zero-AI either way.
Upload archive.zip or a direct OHLCV CSV. Files are stored on the persistent Railway volume, not inside the deploy package.
Dataset
Market
Timeframe
Candles
Date range
Status
Action
HISTORICAL VALIDATION
Cross-dataset validation of a frozen G91 candidate. PROXY ROBUSTNESS never recalibrates the candidate and never counts as native ES locked/promotion evidence.
IDLE · 0%
No proxy validation run yet.
Legacy Multi-R Research Metric
Target-touch research metric. A target not hit inside the current research horizon counts as a non-win; this is not the same thing as closed-trade profit win rate.
No run yet.
Research Brain
Waiting for a research run.
Execution Truth Audit
Read-only audit of the existing EXEC results. It compares the current conservative 5M labels with chronological 1M evidence and separates target wins, stop losses and unresolved trades. EXEC and all white/red/green chart lines remain untouched.
Audit evidence does not rewrite research labels automatically. Any parity correction remains human-approved.
SNIPER Auto-Lab
Runs automatically with every research cycle. It tests single + layered causal EXEC-time filters and exact closed-1M RECLAIM HOLD / RECLAIM + BOS timing profiles, then keeps a locked chronological test split. The current research target is at least +10 percentage points over EXEC without changing the white/red/green chart geometry.
Prototype edge is research evidence, not a guaranteed future win rate. Production promotion remains locked.
SNIPER Evolution OS · Historical
Isolated CONTROL-vs-CHALLENGER evolution over historical data already stored in BLVK Lab. It generates and mutates challengers locally in Python, tests train → validation → locked test, stress-tests the winner, and exports a Shadow Pine challenger. No external AI/API calls are made by this operation.
CONTROL READ-ONLY · SEPARATE /app/data/evolution/ STORAGE · $0 EXTERNAL AI CREDITS · NO PRODUCTION AUTO-PROMOTION
AI Research Lab · CONTROL vs CHALLENGER
Professional evidence layer built on top of the existing Evolution engine. Every metric traces to immutable trade records. G90 parameter stress is preserved; G91 investigates context/regime attribution. No candidate can overwrite CONTROL automatically.
Select a challenger to see every score component and weight.
EVOLUTION TREE
No lineage captured yet.
TRADE INSPECTOR
Click INSPECT on a candidate.
Time
Dir
Session
Entry
Stop
Target
R
Outcome
Regime
Research Lab settings · execution, validation + promotion gates
These settings affect future research simulation and evidence gates only. CONTROL strategy mathematics remain frozen. Conservative stop-first intrabar handling cannot be disabled.
EXECUTION / COST MODEL
1M fidelity limit: latency/queue position is recorded but cannot be truthfully reconstructed below the 1-minute bar. TP+SL ambiguity remains STOP FIRST.
Fast prototype testing without touching CONTROL. Paste a supported BLVK SNIPER Pine prototype, or upload manual signal timestamps from your own visual TradingView test. Results are matched against the stored 1M history and scored with realistic costs. Arbitrary Pine is never fake-executed: unsupported custom logic returns PARITY REQUIRED.
Prototype Test Bench ready.
IDLE · 0%
Mode A · BLVK Pine Adapter
Paste a BLVK SNIPER-compatible Pine prototype. The tester extracts the supported timing/context parameters and runs the matching Python-equivalent model against the same historical data.
Mode B · Manual Signal Replay
Useful when you hide BLVK EXEC in TradingView and mark/test ideas manually. CSV minimum: timestamp,direction,stop. Optional: entry,target,order_type,setup_type,notes. If entry is absent, the next 1M open is used. Both TP and SL in one bar = STOP FIRST.
Sandbox rule: once you inspect these historical results, that history is research-exposed. A promising prototype still needs professional walk-forward, fresh locked data and forward shadow before it can challenge CONTROL.
Autonomous Research
Checks for new market data every hour, but only spends research/API work when there is new data. Default market is ES. Production promotion remains locked.
Live Market Listener
Databento OHLCV-1M streams straight into the persistent master file. Automatic SNIPER research runs in economy mode and skips OpenAI spend.
Requires Databento live entitlement. If unavailable, historical AUTO mode continues separately.
Permanent BLVK Brain
Research memory, hypotheses, rule candidates and chart evidence are stored under /app/data on your Railway volume.
Parity is deferred. AI may research now, but no finding can become a production indicator rule until parity is certified later.
Live Safety
Every received 1M bar is append-only and deduplicated by timestamp. Research reads a locked snapshot so the market socket never edits EXEC or blocks on a backtest.
PROTECTED EXEC: UNCHANGED · PRODUCTION PROMOTION: LOCKED
Shadow SNIPER
Runs the Shadow SNIPER V2 tournament around the protected EXEC white line. Many causal 1M profiles compete on the same setups; the best candidate is persisted and retested on future cycles. No broker orders are placed.
Shadow Results
No shadow cycle yet.
Shadow/replay is simulation only. It can improve research evidence, but it never changes or places a production trade.
Latest Visual Evidence
These are the same 5M + 1M setup images the OpenAI research scientist can inspect.
BLVK Doctor · Maintenance Center
Use this area to understand platform problems without touching protected trading logic. Diagnostics and safe repairs operate on UI/job/runtime state only. AI Doctor can explain likely bugs when AI Brain is ON; code-level changes are never hot-applied to CONTROL.
Run diagnostics to inspect the platform.
AI Bug Fixer
Describe what is wrong in normal language. BLVK Doctor reads current diagnostics and recent errors. It may recommend a minimal code fix, but protected strategy files are forbidden and no live code is silently rewritten.
No diagnosis yet.
Advanced platform settings
Workspace-only controls. They change how BLVK Lab is displayed/polled, never strategy mathematics.
Parity Workbench — deferred for now
When you are ready, this compares TradingView's indicator outputs against BLVK Lab. Research can continue without it; production promotion cannot.