From idea to hypothesis
What the Research Lead does with what you give it, and how to give it more.
What you can submit
The idea box accepts plain text, chart screenshots, PDFs (scanned pages are read too), Word documents, EPUBs, HTML, spreadsheets and code files, up to 10 attachments of 25 MB each. Audio and video files are transcribed. Text is extracted and read in full: a whole book or a proven-strategy write-up can be distilled into several hypotheses in one submission. Images are looked at directly, so an annotated chart is a valid idea.
Links in the idea text are read as well, up to 10 per idea: public web pages and PDFs, GitHub repositories, and video pages, which are read from their captions or a transcript plus key frames. Each source is listed with its status, so a link that could not be read is reported rather than silently dropped.
What a hypothesis contains
A hypothesis names the instrument and venue, the bar timeframe, the signal that opens a position, the rule that closes it, the expected holding period and a risk profile with stop and position-sizing intent. It also states the economic reason the edge should exist, which the analyst later checks against the evidence.
Fee awareness from the start
Every venue in Stratmill carries its real maker and taker fees, and a round trip on a perpetual future costs about 0.1% of notional. The Research Lead is asked to propose edges whose average trade return clears that bar, and ideas that cannot are declined with an explanation rather than quietly coded.
Choosing a venue
Long-only ideas without leverage belong on spot markets; ideas that short, use leverage or trade the funding rate belong on perpetual futures; US stocks and ETFs trade on the US equities venue with session hours. The venue decides fees, leverage caps and whether shorting is possible, so it is fixed before any code is written.
Standing instructions and knowledge
If you always want a particular venue, timeframe or style, write it once as a standing instruction for the Research Lead in the AI Center. Attach knowledge files, such as your own notes on a market, and the agent reads them on every call in your workspace.
When an idea is declined
The Research Lead declines ideas with no tradable market, no testable rule, or an edge that cannot survive trading costs. The reason is recorded on the hypothesis so you can refine and resubmit. Declining early is cheaper than coding, verifying and backtesting a dead end.
The Strategy Builder
The Strategy Builder is a guided way to turn an idea into a hypothesis, step by step, with your approval before development. Open Strategy Builder from your workspace to clarify the rules, inspect feasibility and prior research, and approve or revise the draft. Availability is controlled by the workspace and deployment settings.
Questions, checks and critique
In the Strategy Builder each idea becomes a session. The idea clarifier may first ask a few questions about the market, timeframe or rules; you can answer them or skip them, and quick mode never asks. A checklist then confirms that the data, the venue and a viable fee floor exist, and similar past ideas are shown with their verdicts as advice. The Research Lead drafts the hypothesis, and the idea critic, a read-only reviewer, checks the draft for problems that can be judged before any code exists, such as an edge too small to pay its fees or a look-ahead risk. If the critic could not run, the card says the draft was not reviewed.
Approve or revise
The draft appears as a hypothesis card. Approve sends it straight to coding, where it uses credits like any other idea; Revise sends your changes back through the checks. A session stops with an explanation when the data, venue or capability is unavailable, when the idea duplicates an existing one, or when the Research Lead declines it. You can keep up to 5 sessions open, and an optional setting approves clean cards automatically. After approval the page follows the run through coding and verification to its backtest.