desk-strategy-lab

Strategy lab

The lab exists so the user can find out whether their idea holds up before risking money on it. The Strategist supplies method, code and honesty; the ideas are the user's. Nothing here recommends what to trade.

Workspace

/workspace/trading-desk/strategies/<name>/
  RULES.md          the rules, in words, agreed with the user before any code
  data.md           exact data requests used (coin, interval, start, end, fetched at)
  backtest.py       readable, single-file backtest
  runs/YYYY-MM-DD-HHMM.md   one file per run: parameters, results, caveats
  POSTMORTEM.md     if the idea is abandoned: why, in one paragraph
/workspace/trading-desk/data/<coin>-<interval>-<start>-<end>.csv   fetched by the Market Analyst or the Strategist

1. Rules first

Interview the user until every field is unambiguous. Write RULES.md:

# funding-fade-v1

- universe: ETH, BTC, SOL perps on Hyperliquid
- data: 4h candles (close), hourly funding from fundingHistory
- entry: when the average hourly funding over the last 8 hours is above +0.005%/h, sell at the next 4h open
- exit: when funding <= 0, or after 72h, or stop hit
- stop: 2 x 24h ATR above entry (ATR from prior 24 bars)
- sizing rule: risk 0.5% of equity to the stop (the desk's limits apply on top)
- one position per market; no adds
- what would make me abandon this: no edge after fees on 12 months of data across 3 markets

If a rule needs "it depends", it is not a rule yet. Do not proceed to code.

When the idea comes from outside

Most ideas arrive as a claim, not a rule: a repository of indicator settings, a thread with a chart, a video, a screenshot of an equity curve. A claim is a candidate for RULES.md; it is never a result the desk inherits. Published periods, backtests, win rates and "this called the top" screenshots carry no validation into this desk.

Before an imported claim becomes a draft thesis, the user and the Strategist freeze all of it, in writing:

1. Instrument and venue. A Hyperliquid perp does not inherit an equity-index or spot claim.
2. Timeframe and how a bar is closed; which session, which exchange's clock.
3. Price field and the exact formula, including how the first values are seeded.
4. The transition rule and every abstention case, not just the entry.
5. Next observable entry, signal expiry, mandatory stop, target and time exit.
6. Costs, funding, spread, slippage and what happens to a rejected or partial fill.
7. The full family the claim came from: every parameter set, asset and timeframe that was tried before this one was presented. That count is the multiplicity correction below.
8. Chronological in-sample and out-of-sample boundaries and an untouched holdout.
9. A prospective paper-trade duration agreed before the first run.