Rion Autopilot And Strategy Discovery

Rion Autopilot and Strategy Discovery are Preview chart-chat workflows. They can move beyond conversational drafting: a recognized prompt can create files in the configured Code Lab project, compile the complete custom project, run research and historical-test stages, and save reports and watchlist records.

These workflows do not place live or simulated orders by themselves. A generated strategy can become order-capable later if a user compiles, loads, and enables it on an account-connected chart.

Release-blocking cloud-consent limitation

In the reviewed source, Strategy Discovery enables cloud ML by default. When a local scan produces candidates, the workflow can upload the full generated feature dataset as CSV plus research metadata. The background path suppresses the Research Lab window and ordinary user messages and does not open a dedicated upload-consent prompt.

This is a release blocker for unrestricted public use. Until the product adds an explicit consent gate and the signed installer is re-verified, include local only or no ml in the initial discovery prompt, verify that the result says cloud ML was skipped, and do not use Strategy Discovery with proprietary, licensed, customer, or otherwise restricted data.

Distinguish The Two Workflows

Workflow	Typical intent	Default stages in the reviewed source
Autopilot / strategy build	Turn a named pattern, setup description, transcript, notes, or trade rules into Rion-owned Code Lab source	Write source, compile, run Validator, and run Optimizer for a strategy request unless the prompt opts out
Strategy Discovery	Search for a candidate strategy or edge for an instrument	Build or select a dataset, run a local/Python prescan, run cloud ML, generate candidates, compile, validate, optimize, audit, report, and update the watchlist unless the prompt opts out

Availability still depends on the installed build, license entitlements, configured AI provider, Code Lab project, historical data, Research Lab runtime, and packaged Validator/Optimizer modules. A stage shown in the plan is not proof that it completed.

Recognized Prompt Families

Autopilot can recognize combinations such as:

• build, create, generate, write, or code with strategy, indicator, script, or Code Lab
• run autopilot, quant research, automatic build, or research mode
• named families such as moving-average crossover, breakout, opening range, mean reversion, trend following, RSI, MACD, or VWAP strategies
• setup/source requests based on pasted rules, notes, a transcript, an article summary, a screenshot description, or a trade log

Strategy Discovery recognizes phrases including:

• find me a strategy
• discover a strategy
• research a strategy
• strategy discovery
• find profitable strategy
• find an edge and code it
• research lab autopilot

Natural-language matching is broad and can change between builds. If you only want an explanation or draft, say so explicitly and do not use build/discovery wording.

Control The Stages In The Prompt

Use the opt-out in the initial request so the workflow is configured before it starts.

Phrase	Current effect
local only or no ml	Disables the Strategy Discovery cloud-ML stage. Local dataset and prescan work can still run.
no python, no prescan, or skip python	Disables the Python prescan where the workflow supports that choice.
no backtest or no validator	Skips Validator.
no optimize or no optimizer	Skips Optimizer.
code only or write only	For Autopilot build requests, skips Validator and Optimizer.

Example local-only request:

Find me a strategy for BTCUSDT on the active chart. Local only, no ML, and no optimizer.

Before proceeding, confirm the intended instrument, connection/provider, timeframe, date range, and whether the data license permits each requested local or cloud stage.

Understand The Execution Stages

1. Intake resolves the prompt, instrument, provider, timeframe, and available chart/research context.
2. Dataset and prescan can create or select Research Lab data and run local/Python candidate discovery.
3. Cloud ML, when enabled, uploads the feature dataset and requests a cloud-assisted scan.
4. Strategy specification and code generation write Rion-owned Markdown and C# artifacts.
5. Compile builds the entire Code Lab custom project, not only the generated file.
6. Validator tests generated candidates against the selected historical context.
7. Optimizer can run after a usable validation result and available optimizer prerequisites.
8. Audit and report record completed, failed, and skipped stages.
9. Watchlist stores a durable index to the generated source, metrics, diagnostics, reports, and optimizer results.

If compilation fails, Validator and Optimizer are skipped. If a result lacks a completed-stage artifact, do not infer success from conversational wording.

Generated Files And Ownership

Paths below are relative to the configured Code Lab custom-project root unless an absolute Windows path is shown.

Artifact	Current location or pattern	Important behavior
Generated indicators	Indicators\RionAuto*.cs	Rion-owned names can be rewritten by a later Rion run.
Generated strategies	Strategies\RionAuto*.cs or Strategies\RionResearch*.cs	Rion limits automatic writes to its owned naming families; review every generated order path.
Source-derived strategy specification	RionSpecs\RionSourceStrategySpec_yyyyMMdd_HHmmss.md	Can contain sensitive rules or notes supplied in the prompt.
Run report	RionReports\yyyyMMdd_HHmmss\RionFloorReport.md and RionFloorReport.json	Records stages, candidates, diagnostics, test summaries, and source/report paths.
Candidate and side-agent artifacts	Subfolders such as Candidates and SideAgents inside a run-report folder	JSON and Markdown are not guaranteed to be encrypted.
Watchlist records	Documents\HyperionX\ResearchLab\StrategyWatchlist\*.json	Separate index records that can point to generated source and reports.

Do not rename a file into the RionAuto* or RionResearch* ownership families and then expect manual edits to be preserved. Use Promote in the watchlist to create a uniquely named Promoted...Strategy.cs copy marked as user-owned, then compile Code Lab before validating or loading it.

Use Reports

Open Tools > Reports. The dashboard reads the RionReports folder under the configured Code Lab project and can expose:

• the report library and Markdown/JSON run artifacts
• stage, audit, candidate, and generated-strategy summaries
• candidate price/equity views and trade rows
• trade CSV export
• Python-versus-C# comparison artifacts when the run produced them
• direct access to the report folder

A report is evidence about that run, not proof of reproducibility or live profitability. Preserve the application version, provider/data provenance, parameters, commission and fill assumptions, and source revision separately.

Use The Rion Strategy Watchlist

Open Tools > Rion Strategy Watchlist. For a selected record, the current window can open source, dashboard, report, or optimizer output; promote the source; archive the record; or delete it.

Deletion scope is intentionally narrow:

• Delete removes only the selected watchlist JSON record.
• Clear Archived removes archived watchlist records.
• Clear All removes all watchlist records.
• These actions do not delete generated C# source, RionReports folders, Research Lab datasets, exported CSV files, or promoted strategies.

Archive is a watchlist status, not a file-retention policy. To remove a complete run, first identify every referenced path, close tools using those files, back up anything required, and remove only recognized artifacts.

Validate A Result Before Use

1. Open the generated file in Code Lab and review every parameter, data dependency, and order call.
2. Confirm the complete custom project compiles with no relevant warning or repair artifact left unresolved.
3. Open the report and verify which stages completed, failed, or were skipped.
4. Confirm the provider, instrument, timeframe, date range, and local/cloud data provenance.
5. Re-run Validator with recorded commission, leverage, fill, and merge assumptions.
6. Promote the strategy before making durable manual edits.
7. Test the promoted copy in historical validation, playback, and an explicitly selected Hyperion Sim account before considering any broker-connected use.

Related Guides

• Rion AI
• AI Data And Privacy
• Research Lab
• Code Lab
• Backtesting
• Validation
• Optimization