Research programme · crypto perpetual futures
Trade signals that survive contact with the market.
Mercury Ledger builds a robust trade-signal generator. On that foundation comes the second stage: a trading system that learns from its own results and stays profitable. Everything else we keep, the records, the receipts, the reviews, exists to make those two things true.
One goal, two stages.
The order matters. A learning system built on weak signals learns to be confidently wrong, so the generator comes first and has to earn its place.
-
Stage 1in research
A robust trade-signal generator
Robust has a precise meaning here. The signal is tested only on data it has never seen. Trading costs are paid before any edge is counted. No information from the future reaches a decision. It is judged on consistency across years, not on one lucky quarter. And every time it declines to trade, that refusal is written down next to every time it fires.
-
Stage 2designed, waits for stage 1
A profitable self-learning trading system
The system studies its own trades, proposes a better version of itself, and is promoted only when later, untouched data agrees. Hindsight is not evidence. People keep the kill switch, decide what goes live, and set how much is at risk.
Doctrine
Steady over spectacular.
We are not building a money printer. A modest return that repeats, after costs, with shallow drawdowns, is worth more than a spectacular backtest that depends on one trade.
So the selection rule is unusual: a result whose profit comes mostly from a single month is rejected even when its total is higher. Think of a geothermal plant, not a lottery ticket.
How a signal earns its place.
- 1
Write the claim first
The hypothesis, the statistics and the pass rule are frozen before anyone sees a result. A claim written after the fact is not a claim.
- 2
Test on unseen data
Rolling windows, realistic fees and slippage, and control tests designed to catch leaks from the future. A signal that only works at zero cost is not a signal.
- 3
Argue against it
Every conclusion gets a review whose job is to break it, and a chart before any verdict. Failed candidates are recorded, not quietly dropped.
- 4
Watch it forward
A live clock runs without capital until the forward record holds up. Only then does the question of real money come up, and it is a small one at first.
Why “Ledger” in the name
The ledger is the audit trail, not the product. Every observation, forecast and verdict is written once and never edited, so a signal’s record can be checked later by someone who was not in the room. That discipline is the price of trust. The goal is the signal, and the system that learns from it.
- Market capture
- 24/7
- Automated checks
- 3,900+
- Live positions
- 0by design
- Stages, one goal
- 2
Work record · read from version control
Built in the open, every day.
These figures are counted from the programme’s own commit history, across its repositories, once a day and at every publication of this page, never typed in. They measure effort, not results. Results keep their own, much slower record.
Last change · revision dc5b68f1
- Days building
- 348
- Days with commits
- 290
- Lines of code and documentation
- 599,904
- Source and document files
- 2,460
Talk to us.
Questions, collaboration, or a sceptical read of the method are all welcome.
info@mercuryledger.netMercury Ledger is a research programme. Nothing on this site is an investment product, investment advice, or an offer to buy or sell anything. Past or illustrative results do not predict future returns.