How it worksResearchSafetyPricing
Docs Open app

The Research Desk

Every day the desk proposes new rules, tests each one on prices it was not built on, and paper trades the survivors. A rule reaches the catalogue only after it clears every gate.

Open app
1,195
Configurations tested
4
Timeframes searched
3
Paper trials running
3
Strategies listed

The loop

Every day the desk proposes, tests and lists new rules. Nobody signs off.

What fails leaves the loop and stays on the record. Live results feed the next round of ideas.

The research loopAgents propose rules (18 so far). The desk backtests their variants on past prices (1,195 configurations), gives the best one run on prices it never saw (3 passed), paper trades survivors for 60 days (3 running), and lists only what clears automatic gates (3 listed). Live results feed the next ideas. 15 candidates stopped on unseen prices and 6 strategies were withdrawn; all stay on the record.Retired, kept on the record6withdrawn by the gates15stopped on unseen pricesLive results feedthe next round of ideas01ProposeAgent researchers write new rulesRunningCross-sectional momentumdailyVolatility targetdailyConcentrated momentumdaily18rules proposed02BacktestVariants run on prices before 1 Mar 20261,195configurations tested03Unseen testOne run on prices it never sawsearchednever seen3 of 18passed04Paper trial60 days on live prices, no money at riskFunding exit27 DecConcentrated breakout, 4-hour27 DecConcentrated breakout, daily28 Dec3running now05GatesChecked in code. Nobody signs off.Beats holding on return by 5 pointsNot luck: both tests under 0.10Makes money on paperEngine matches the desk on every barAutomaticlisting and withdrawal06ListedOffered to your accountBreadth TrendBeing upgradedBreadth Trend 150Being upgradedBreadth Trend 100Being upgraded3strategies listed The research loopAgents propose rules (18 so far). The desk backtests their variants on past prices (1,195 configurations), gives the best one run on prices it never saw (3 passed), paper trades survivors for 60 days (3 running), and lists only what clears automatic gates (3 listed). Live results feed the next ideas. 15 candidates stopped on unseen prices and 6 strategies were withdrawn; all stay on the record.Live results feed the next round of ideas01ProposeAgent researchers write new rulesRunningCross-sectional momentumdailyVolatility targetdailyConcentrated momentumdaily18rules proposed02BacktestVariants run on prices before 1 Mar 20261,195configurations tested03Unseen testOne run on prices it never sawsearchednever seen3 of 18passed04Paper trial60 days on live prices, no money at riskFunding exit27 DecConcentrated breakout, 4-hour27 DecConcentrated breakout, daily28 Dec3running now05GatesChecked in code. Nobody signs off.Beats holding on return by 5 pointsNot luck: both tests under 0.10Makes money on paperEngine matches the desk on every barAutomaticlisting and withdrawal06ListedOffered to your accountBreadth TrendBeing upgradedBreadth Trend 150Being upgradedBreadth Trend 100Being upgraded3strategies listedRetired, kept on the record6withdrawn by the gates15stopped on unseen prices

The gates

Four gates between a new rule and your account.

  1. 01

    Unseen prices

    One look at prices it was not built on

    The search reads only prices before 1 Mar 2026. The winner is then run once on the months after, net of fees and funding, and must beat holding the same markets by 5 points of return without a deeper fall.

  2. 02

    Luck

    Its edge must not be luck

    White’s reality check and the deflated Sharpe ratio, charged for all 1,195 configurations tried so far, must both come in under 0.10.

  3. 03

    Paper

    60 days on live prices

    The rule is frozen and paper traded on new prices with no money at stake. It has to make money over the window.

  4. 04

    Engine

    The engine must trade it exactly

    Archon’s live engine has to reproduce the desk’s positions on every bar of every market before the rule can reach an account.

Catalogue

Every listed strategy has to beat holding.

On return by 5 points, with a worst fall no deeper. Listing and withdrawal are automatic. A listed strategy that slips stays on offer, marked, until a stronger one replaces it.

StrategiesBacktest · after fees · since 20 Aug 2024
StrategyReturnHoldingWorst fallHolding’s worst fall
Breadth Trend +150.0% +101.7% −23.0% −64.8%
Breadth Trend 150 +133.6% +101.7% −23.7% −64.8%
Breadth Trend 100 +120.1% +101.7% −23.2% −64.8%
Being upgraded by the research desk

The desk is hunting a version that beats holding outright. This one keeps trading until a stronger rule passes every gate and replaces it.

Breadth Trend's backtest against holding the same markets, in the Archon app

Where every candidate stands

3 on paper now. Every step on the record.

Every candidate, stage by stage3 candidates are in a 60-day paper trial after passing their unseen test; 15 more stopped on unseen prices. 3 listed strategies are being upgraded by the desk, and 6 were withdrawn by the gates.CandidateProposedUnseen testPaper trialGatesListedFunding exit4-hour candles, proposed 28 Oct 2026until 27 Dec 2026Concentrated breakout, 4-hour4-hour candles, proposed 28 Oct 2026until 27 Dec 2026Concentrated breakout, dailydaily candles, proposed 28 Oct 2026until 28 Dec 202615 other candidatesStopped on unseen prices, kept on the recordBreadth TrendBeing upgraded by the research deskBreadth Trend 150Being upgraded by the research deskBreadth Trend 100Being upgraded by the research desk6 strategiesWithdrawn by the gates, 28 Oct 2026 Every candidate, stage by stage3 candidates are in a 60-day paper trial after passing their unseen test; 15 more stopped on unseen prices. 3 listed strategies are being upgraded by the desk, and 6 were withdrawn by the gates.ProposedUnseenPaperGatesListedFunding exit4-hour candles, proposed 28 Oct 2026until 27 Dec 2026Concentrated breakout, 4-hour4-hour candles, proposed 28 Oct 2026until 27 Dec 2026Concentrated breakout, dailydaily candles, proposed 28 Oct 2026until 28 Dec 202615 other candidatesStopped on unseen prices, kept on the recordBreadth TrendBeing upgraded by the research deskBreadth Trend 150Being upgraded by the research deskBreadth Trend 100Being upgraded by the research desk6 strategiesWithdrawn by the gates, 28 Oct 2026

Desk log

28 Oct 2026

Unseen prices

Cross-sectional momentum, daily

Closed its test on unseen prices

28 Oct 2026

Unseen prices

Cross-sectional momentum, 4-hour

Closed its test on unseen prices

28 Oct 2026

Unseen prices

Volatility target, daily

Closed its test on unseen prices

28 Oct 2026

Unseen prices

Volatility target, 4-hour

Closed its test on unseen prices

28 Oct 2026

Unseen prices

Concentrated momentum, daily

Closed its test on unseen prices

28 Oct 2026

Paper trial

Concentrated breakout, daily

Started 60 days of paper trading