Your Research
For agents
- Lets you: Turn a user's view into a checkable claim, a research run and a pod draft.
- Tools:
test_run_check,build_research,build_runs,build_run,build_pod - Limit: A check uses one of the plan's daily checks unless the same spec already ran: Free 3, Plus 30, Pro and Ultra 200.
- Limit: Quick 50, Research 100, Deep 250 credits per run, charged at start; rebuilding a card costs 120; Free gets 1 for 0, then needs Plus.
- Limit: A run takes several minutes.
From sentence to pod
YouGen Z is drinking less
Lyraby when, and what proves it wrong?
Claimdrinks that replace alcohol beat brewers and distillers
Checkheld in 10 of 15 past cases
DraftCELH, MNST long; STZ, DEO short
What a draft holds
- Claim: one testable sentence.
- Holdings: each leg, its side and weight.
- Check: the grade and the past-case count.
- Ends if: what would prove it wrong ("beer sales bounce back").
- Horizon: how long the claim is meant to hold.
- Loss levels: halve and close.
- Money: the share of your fund it takes.
Claim types
| Type | Shape | Needs |
|---|---|---|
direction |
one basket beats a benchmark | benchmark |
theme |
a theme basket beats a benchmark | benchmark |
relative |
longs beat shorts | a short leg, no benchmark |
event |
a move after given dates | events |
odds |
a move after a prediction market crosses a level | market, level, cross |
regime |
a side while a condition holds | condition, side |
carry |
holding cost and funding over days | days |
rule |
a formula sets the weights | formula |
Example check
Input to test_run_check:
{"spec": {"claim": "Drinks that replace alcohol beat brewers and distillers over the next quarter",
"family": "sober-curious", "type": "relative",
"universe": {"long": ["US:CELH", "US:MNST"], "short": ["US:STZ", "US:DEO"]}, "benchmark": "",
"checks": ["past_cases", "holding_cost"], "start": "2016-06-01", "horizon_days": 90,
"capital": 3000, "leverage": 1, "market": "", "level": 0.5, "cross": "up", "condition": "",
"side": "long", "days": 0, "formula": ""}}
Output shape (numbers are an example):
{"grade": "leans_yes", "label": "Good Idea", "line": "10 of 15 past cases", "versions_tried": 1,
"as_of": "2026-11-16", "past_cases": {"wins": 10, "cases": 15, "mean": 0.031,
"period": ["2016-06-01", "2025-11-16"]}, "holding_cost_month": 0.004}
curveis added when the book holds at least 5 names on average: at most 250 points starting at 1.- A hand-picked basket grades at most Good Idea. Why: How a check grades.
Depth by plan
| Plan | Depth | Credits | Reads |
|---|---|---|---|
| Free | Quick research | 50 | web, tickers |
| Plus | Research | 100 | web, filings, data, tickers |
| Pro, Ultra | Deep research | 250 | web, papers, filings, data, tickers, plus a second reviewer |
What comes back
- A read: what the view claims, in plain words.
- The check: grade and past cases. How a check grades.
- A fund card: holdings, sizes, loss levels, holding cost.
- What ends it.
- Both land in your Lyra chat.
Rebuild the card
- Drop symbols, add picks, lean (conviction 1 to 3 per symbol).
- Structure:
singleormain_offset(a main basket with an offset). insist: build despite Not sure.- Code sets the weights, the offset, the size and the loss levels. The check stays the run's.
- Tool:
build_pod. Example input: MCP tool reference.
When Lyra holds off
- The check did not run.
- The past cases do not back it.
- The check is newer than the prices.
- Some names have too little price history.
- It would repeat a pod you already hold.
- It adds risk to your fund without enough reward.
- The money for it is too small to place every order.
Credits
- Charged at start; the charge is the run's cost ceiling.
- A refused run is refunded.
- A run that errors early is refunded, up to 3 a month.
- Out of credits: the run is refused before any spend.