Growth Autonomous Trading
updated loading…
01 · MARKET BOARD

Markets

Loading market data…
Sources & method
  • How it works. Prices are pulled every hour by a scheduled job and stored. This page reads that stored file — it never calls a price API itself, so it costs nothing to visit and behaves the same at any level of traffic.
  • Crypto — CoinGecko. Equities, ETFs, indices, DXY, oil & gas — Yahoo Finance; London-listed funds are quoted from their USD lines. Gold, silver — COMEX front-month futures (not spot), price and history. Copper — live spot. US M2 — the St. Louis Fed (FRED).
  • Oil and natural gas are front-month futures. There is no free spot feed for either, and the front-month contract is the price the market actually quotes.
  • US M2 is published weekly and in arrears, so its date is older than the rest of the table and it has no 24h figure. Its changes are measured from the publication date, not from today.
  • N/A means a figure genuinely cannot be computed yet — not zero. Copper has no free spot history, so its changes fill in as days accumulate. SpaceX listed in June 2026, so it has no year-start price to measure against.
  • Percentages come from each source's own history where it exists, so a missed day never distorts a number.
02 · EDITED FOR CONSEQUENCE

Daily Brief

Loading brief…
03 · FROZEN BEFORE, SCORED AFTER

Weekly Review

Loading weekly review…
04 · APPEND-ONLY REGISTER

Alpha Hunt

Loading alpha hunt…
05 · TWO EXPERIMENTS, FINAL RESULTS

Experiments

2 experiments, both ended 9 Nov 2026final, not updated
Final resultEnded 9 Nov 2026Paper money · $8,000This page is final and will not change

MGAT Alpha v3.1: could rule-based trading programs beat simply holding the same coins?

Eight automated programs traded cryptocurrencies on paper, every hour, for eleven weeks, against rules for judging that were written and signed before the first trade. This is the final account of what they did and what it shows.

Summary of results

No strategy family showed a reliable advantage over holding the same coins after fees. The programs lagged holding while prices rose and lost less while prices fell, which is what holding cash does, not what good trade selection does. The experiment was stopped on 9 November 2026 and will not continue with real money.

Trades closed139of a planned 200; 7 still open at the stop
Closed trades, retained−$227.15135 trades; +$170.52 as booked with 4 unverifiable trades
Account value$8,154.90+1.9% as booked, open positions included
Profit factor0.78dollars won per dollar lost, 135 trades
  • Formal verdict: inconclusive for all four families. The outside price check that was meant to confirm every trade confirmed none of them, so under the signed rules no family could pass or fail.
  • Picking small coins lost. The two small-coin families (BC, MED) lost money in the falling segment and had the largest samples.
  • The AI adviser could not tell winners from losers. Trades it approved and trades it rejected lost almost the same (−5.3% and −5.5%).
  • Three independent AI readers, from three different companies, read the frozen records and all recommended ending the experiment.

1The experiment

MGAT tested one question: does an automated trading program earn more than an investor who buys the same coins and holds them? Version 3.1 removed everything except the trading logic. It traded on paper: real market prices, simulated money, no slippage. That made the rules the only thing being tested.

A dedicated computer ran one decision round every hour, day and night. It held $8,000 of paper money in eight programs of $1,000 each: one each for the L1, BC and MED families and five for DEX, one per blockchain network. Each purchase used 15% of a program's available cash (minimum $20, maximum $600), at most three positions per program, with exchange fees of 0.1% to 0.3% charged on every buy and sell.

The four strategy families. Each family is a separate hypothesis with its own entry and exit rules.
FamilyWhat it boughtTake-profitStop-lossTime limit
L120 of the largest coins (Bitcoin, Ether, Solana and others) when up 2% in 24 hours and above their 20-day average+12%−8%7 days
DEXNew tokens on decentralised exchanges with rising price and enough trading; five programs, but only the Sui and BNB Chain ones ever traded+25%−12%3 days
BCCoins worth over $100 million, up 3% in 24 hours, with a guard against buying a sharp fall+20%−10%7 days
MEDCoins that had been heavily sold (14-day RSI of 35 or less) or had just closed back above their 20-day average+15%−10%14 days

Two AI advisers recorded an opinion on trades without any power to change them: a simple built-in rule from the start, and from 13 October a large language model (Sonnet).

2Test period and how it was judged

The programs traded from 25 August to 8 November 2026. The operator stopped the run on 9 November at 20:26 UTC, at 139 of the 200 planned trades; the books were frozen and fingerprinted at that moment. The rules for judging were signed before the first trade, so the results could not be read selectively afterwards:

  • The run ends at 200 closed trades, on 30 January 2027, or earlier by the operator.
  • A software repair on 7 October splits the run into segment 1 (before) and segment 2 (after). Each family is judged on segment 2 alone and the two are never pooled.
  • A family passes only with at least 30 trades, a positive average after fees, a positive average even without its two best trades, and a better yearly return than both the Nasdaq-100 and the S&P 500.
  • Every trade must be confirmed by a separate, outside price record.

Reaching 200 trades would have taken about five more weeks, and no family was close enough to the pass line for those trades to change the answer, so the run was stopped early.

3Results

The fleet as a whole

The 139 closed trades booked +$170.52. Four of them, on BNB Chain, took their prices from a single source that could not be cross-checked; one alone booked +$515.83. The signed rules exclude them. The 135 retained trades lost $227.15, an average of −1.4% per trade after fees, with 34% of trades profitable. At the stop the account stood at $8,154.90, including 7 open positions bought for $728.11. The largest fall from a peak was 5.4% as booked, 11.2% without the four unverifiable trades.

92.59597.5100102.5105107.5110repair, 7 Oct1 Sep1 Oct1 NovAs booked 102.2Adjusted 97.2Bitcoin 95.3
Account value against Bitcoin, both indexed to 100 on 25 August. Adjusted: account value minus the profit of the four unverifiable trades (ends at 97.2). As booked: exactly as recorded (102.2). Bitcoin ended at 95.3. Last hourly record 8 November. The fleet fell less than Bitcoin mainly because it held cash much of the time.

By strategy family

−$200−$150−$100−$50$0$50$100repair, 7 Oct1 Sep1 Oct1 NovDEX +$58L1 −$3BC −$130MED −$152
Cumulative profit and loss of each family, in dollars, unverifiable trades excluded. BC and MED drifted down after the October market peak, L1 gave back its earlier gains, and DEX rests on only eight trades.
Results by family and segment, after fees. "Without best 2" removes each group's two best trades, to test whether a result depends on a lucky few.
FamilyPeriodTradesWin rateAverage per tradeWithout best 2Profit / loss
L1Segment 1 · 25 Aug–7 Oct1753%+2.8%+1.5%+$55.65
L1Segment 2 · 7 Oct–8 Nov50%−8.3%−10.2%−$58.45
DEXSegment 1 · 25 Aug–7 Oct667%+9.0%+2.6%+$76.21
DEXSegment 2 · 7 Oct–8 Nov20%−5.8%—−$18.61
BCSegment 1 · 25 Aug–7 Oct4045%+1.4%+0.1%+$49.33
BCSegment 2 · 7 Oct–8 Nov3219%−4.9%−6.8%−$179.19
MEDSegment 1 · 25 Aug–7 Oct1833%−2.0%−5.4%−$49.22
MEDSegment 2 · 7 Oct–8 Nov1520%−6.4%−9.8%−$102.87
All25 Aug–8 Nov13534%−1.4%−1.9%−$227.15

In segment 2, the period the rules judge, only BC reached 30 trades, and it lost 4.9% per trade. L1 (5 trades), DEX (2) and MED (15) never reached the minimum.

Against simply holding the same coins

following the strategyholding the same coinsdifference−20%−10%0+10%+20%+30%L1 Aug–Oct (seg. 1)−7.1 ptsL1 Oct–Nov (seg. 2)+1.5 ptsDEX Aug–Oct (seg. 1)−5.0 ptsBC Aug–Oct (seg. 1)−15.4 ptsBC Oct–Nov (seg. 2)+2.5 ptsMED Aug–Oct (seg. 1)+0.2 ptsMED Oct–Nov (seg. 2)+1.8 pts
Strategy result against holding the same coins over the same dates (equal money per coin, held from the first purchase to the last sale in each segment). In the rising segment 1, L1, DEX and BC earned clearly less than holding and MED was level. In the falling segment 2 every family lost slightly less. DEX segment 2 (2 trades) is too small to compare.

The pattern is the same in every family, which points to its cause: a program that is out of the market much of the time earns less when prices rise and loses less when they fall. Some of that idle cash was built in, since three of the five DEX programs ($3,000 of the $8,000) never made a trade.

How trades ended

Stop-loss66 trades · avg −11.7%Time limit43 trades · avg +1.7%Take-profit26 trades · avg +19.8%
Exits of the 135 retained trades. Half ended at the stop-loss, losing 11.7% on average; take-profits gained 19.8% but were too few to cover them. Prices were checked once an hour, so stops often filled beyond their level.

The AI advisers

The built-in adviser approved every trade in three families and rejected every trade in the fourth, so it was a label for the family rather than a judgement. The Sonnet adviser was scored on 50 trades: the 5 it approved lost 5.3% on average and the 45 it rejected lost 5.5%. A permutation test gives p ≈ 0.97, meaning the difference is easily explained by chance. Its questions contained only information available at the time of each trade.

4Limits of the evidence

  • No outside confirmation. The separate price record stored only the latest price, so it confirmed none of the 139 trades. Every figure is the program's own record.
  • Small samples. L1 completed 22 trades and DEX 8. A small genuine advantage could hide in numbers this small.
  • One market cycle. The period held one rise, to early October, and one fall. A sideways market was never tested.
  • Paper is not real. Fees were modelled, but slippage, minimum order sizes and failed orders were not. Real execution would only lower the results.

5Conclusion

On eleven weeks of hourly decisions and 135 retained trades, no MGAT strategy beat holding the same coins. Selecting small coins lost money, the AI adviser had no predictive value, and the one useful behaviour, sitting in cash while prices fell, is a known property of trend-following rules rather than a discovery. The experiment ended on 9 November 2026 and was not moved to real money.

What remains is the method: rules for judging written before the data, every number re-checked by independent readers, and a plain account of what the data cannot show.

Data. Ledger of 146 purchases and 139 sales, frozen 9 Nov 2026 20:26 UTC with SHA-256 fingerprints checked by all three readers. Market prices: daily closes. Averages are simple means of each trade's percentage result after fees; profit factor uses dollars.

Status. Final. This page replaces the weekly reports and the status card and will not be updated.

Final resultEnded 9 Nov 2026Real money · $130This page is final and will not change

Hades: could an AI program left alone with real money beat simply holding?

One AI program held its own Solana wallet with $130 of real money, decided every hour with no human in the loop, and was told the whole market was open. Four AI set-ups from three makers took turns over eleven weeks. This is the final account.

Summary of results

No AI model found an advantage over simply holding SOL. Left free, the most capable model made no trade in eight days. Given a deadline, it made three small, planned trades and finished slightly ahead of holding, but short of the five trades the test asked for. The experiment ended on 9 November 2026.

Duration77 days24 Aug – 9 Nov 2026, hourly decisions
Final wallet$126.70$3.30 below the $130 stake, because SOL fell
Against holding+1.3%1.4606 SOL against 1.4423 SOL held
Real cost≈ $10plus about $24 of AI usage fees
  • No edge found. Over 77 days the wallet ended within about $1.60 of what plain holding would have been worth.
  • Freedom did not produce action. With every restriction lifted, the model held for eight days without a trade.
  • It never left one trading pair. Every trade was between SOL and a US-dollar token, because the one exchange route it built for itself covered only that pair.
  • Every transaction was checked on the public Solana blockchain.

1The experiment

MGAT tested fixed rules. Hades tested the opposite idea: give an AI program real money and freedom, and see whether it finds and acts on an advantage that fixed rules miss.

  • Money: $130 of real money, 1.3623 SOL, in a dedicated Solana wallet. A one-time top-up of 0.08 SOL was added later.
  • Freedom: internet access, its own wallet, and a wake every hour to decide on its own. No human approved or placed any trade.
  • One rule: the wallet must never reach zero.
  • Benchmark: simply holding the same SOL, measured in SOL and in dollars at the same moment.

2Test period and stages

The experiment ran from 24 August to 9 November 2026. The decision-maker changed in five stages, each one designed from what the previous stage showed. The last two stages were formal tests with a pass line set in advance.

The five stages. Stages 4 and 5 overlap by a day: the stage-5 deadline was given on 4 November and stage 4 was formally closed on 5 November.
StageDecision-makerDatesTradesResult
1DeepSeek24 Aug – 6 Oct12$174.13 against $174.29 for holding: level
2Gemini6 – 12 Oct2too short to score
3Sonnet, with limits12 – 27 Oct0held throughout
4Sonnet, every limit lifted (test 1)27 Oct – 5 Nov0test 1 failed: no trade in 8 days
5Sonnet, given a deadline (test 2)4 – 9 Nov3+0.63% against holding, but 3 of the 5 trades required: stopped

Test 1 asked whether the model would trade at all once nothing held it back. Test 2 asked for at least five trades it could explain, and a finish ahead of holding SOL, by 11 November. It reached three trades, so the experiment was ended on 9 November.

3Results

1 DeepSeek23 Sonnet, limits4 free5$120$130$140$150$160$170$130 stake1 Sep1 Oct1 NovWallet $126.70Holding $125.11
Wallet value against simply holding the same SOL, in dollars. Weekly readings from the published weekly reports, plus the start and the final on-chain reading of 9 November. The two lines move together because almost all of the wallet stayed in SOL: the market, not the trading, set the result. The peak of $175.92 came with SOL at $121 in late September.

Measured in dollars, the wallet rose to $175.92 in late September and fell to $126.70 by the end, almost exactly in step with SOL itself (about $95 at the start, $121 at the peak, $86.75 at the end). Holding the same SOL, including the top-up, would have ended at $125.11.

Pinned at the start1.3623 SOLStart plus the 0.08 top-up (holding)1.4423 SOLFinal wallet, SOL-equivalent1.4606 SOLaxis starts at 1.30 SOL to show the small differences
The wallet counted in SOL. The final wallet held SOL, wrapped SOL and a US-dollar token worth 1.4606 SOL together: 1.3% more than holding the starting SOL plus the top-up. Most of the 7.2% gain over the pinned start is the top-up itself.

What the AI models did

  • DeepSeek traded most, twelve times in six weeks, and ended level with holding.
  • Gemini made two trades in six days, too few to judge.
  • Sonnet, with limits, held for fifteen days. When every limit was lifted it changed its words, not its actions: eight more days, no trade.
  • Sonnet, given a deadline, acted within the hour. Each trade had a stated reason, size, target and stop, written before execution. It finished 0.63% ahead of holding on about $1 of gains, but its first stated reason was the deadline, not the market.

4Limits of the evidence

  • One small wallet. $130 cannot show how an AI program behaves with sums where costs, slippage and risk matter more.
  • One asset in practice. The program only ever traded SOL against a US-dollar token, so the rest of the market was never tested.
  • One market phase. SOL rose into late September and fell through October and November.
  • Changing set-ups. Each stage used a different model or instruction, so no single set-up ran long enough to judge on its own.

5Conclusion

Over 77 days with real money, four AI set-ups from three makers found no advantage over simply holding SOL. Given freedom, the most capable model chose not to trade; given a deadline, it traded small and carefully, which is discipline but not evidence of an edge. The real cost was about $10 plus AI usage fees. The experiment ended on 9 November 2026, the wallet was emptied and retired, and no successor is running.

Data. Every transaction read from the public Solana blockchain. Final reading 9 Nov 2026 16:47 UTC: 1.4606 SOL-equivalent at SOL $86.745. Weekly values from the eleven published weekly reports.

Status. Final. This page replaces the weekly reports and the status card and will not be updated.

06 · REAL WEIGHTS, ONLY IN PERCENT

Allocation

Loading allocation…