Live Trading Mission Control

All pages start at $100 at the same moment and take exactly the council's trades - same coin, direction and entry. Only the exits differ, and controls 2 and 3 skip trades their filter rules out (control 2: a quiet market; control 3: outside weekdays 12-18 UTC, or no 1%+ move in the last 4 hours). So any difference between the pages is the settings alone. Controls are priced from Binance 1-minute mark prices; the council page is its real booked fills. No extra AI calls. On 25 Sep every page that had lost more than half its $100 was retired: control 1 (+40 / 300, $7.18), control 4 (1 in 10 trades at 20×, -$32.10) and control 5 (20× after a small loss, -$38.80).

Ten AI pages starting at $100 (Wisdom AI and Council + Wisdom AI on 16 Sep 21:00 UTC; the pages that use Oracle news or Noesis restarted 18 Sep 22:00 UTC after the news labels were made to prove themselves (a signal only counts as bullish or bearish if it can move crypto within hours, reports a new event, and quotes its source word for word; each signal usable for 1-3 hours from publication); Oracle AI, Axiom and Sophos use a simple rule: an hour leans bullish or bearish when at least two live news signals point that way and more point one way than the other), trading on paper with the council's own exits (+80 target, $2 stop, 4-hour limit) at $150 a trade, one trade per coin per 4-hour window. Oracle AI follows that hourly news lean; Wisdom AI follows an AI reading of market data and trading principles, one trade per coin per 4 hours; Noesis v2 (from 20 Sep 12:00 UTC) gets the same treatment as Wisdom AI v2, which is the same reader thinking out loud - it writes the case for a rise and for a fall, names the principle it used and the one it rejected, and says what would prove it wrong - and it is shown its own record first: how its past calls turned out, its failed short calls, and how each principle has done in its hands (from 17 Sep 21:00 UTC); Noesis reads both together in one AI judgement; Axiom trades only when Oracle AI and Wisdom AI agree; Sophos trades only when Wisdom AI, Oracle AI and Noesis all agree; the three Council + pages copy the council's trades but skip any that their AI points the other way on (Council + Wisdom v2 + Noesis, from 20 Sep 11:00 UTC, skips when either of those two disagrees). On 25 Sep every page that had lost more than half its $100 was retired: Wisdom AI every call ($2.67), Council + Oracle AI ($48.25), Council + Axiom ($36.56) and Council + Sophos ($37.62). These pages are for watching - the pass/fail verdicts come from tests registered in advance.

Every 15 minutes an AI invents new trading rules for BTC, ETH, SOL and BNB, a genetic breeder mutates and combines the best ones, and a random generator explores. A Rust engine built on THE CORE tests each rule on seven years of 5-minute prices in milliseconds. A rule survives only if THE CORE's core-edge check accepts it: fees included, trades never overlapping, profitable on 2024 to mid-2025 data it was not built from, better than taking the same side at random times, and clear of a bar that rises with every rule ever tried. Survivors then face mid-2025 to now exactly once, and only those are tracked live.

Every day a genetic algorithm breeds 80 different ways to weigh the council's 28 agents, over 40 generations, using every vote the council has already cast (no extra AI calls). The winner is then judged only on the next day's decisions, which it has never seen, against the council's own call and against simply always buying. It is flagged promotable only if it beats both over at least 7 unseen days with a t-score above 2. It is a shadow: it never trades, and letting a champion trade is a separate decision.

Every hour it compares BTC's price on Coinbase, where US money buys, with Binance. When that gap jumps unusually far from its past week (more than 2 standard deviations), it follows it: long BTC when US buyers are paying up, short when they are selling off, with its whole balance at 1x for 24 hours. 0.04% per round trip. Paper only.

Why watch-only: before 2024 this made +0.50% a trade (t 3.6). Since 2024 it made +0.15% a trade over 406 trades, still the same direction but too weak (t 1.3) to call proven.

This is the first idea the AI research engine found that passed every check, including realistic costs and an independent re-check. Since then the engine re-judges every live rule against a bar that rises with every idea tried, and this version now falls just short (t 3.4 against 4.2); simpler versions of the same crash bounce still clear it. It keeps trading here because only new data can settle it. Every 5 minutes it looks at BTC, ETH, SOL and BNB. When a coin has fallen sharply over 24 hours and the last hour's selling volume spikes far above normal (both limits frozen from 2019 to 2023 data), it buys $25 of that coin and sells 4 hours later. $100 start, fees included, paper only.

On data it was not built from it made +46 bps a trade (2024 to mid 2025) and +24 bps (mid 2025 to now) after fees, on every coin. It is still a candidate: sharp sell-offs are rare, so live trades will take weeks to add up, and only they can confirm it.

Every day just after the 00:00 UTC close, if BTC closed above its 200-day average it holds BTC worth twice its balance; otherwise it holds cash. It starts at $100, trades on paper at the Binance mark price, and pays 0.05% per trade plus the real funding rate on the whole position.

Why watch-only: in 2024-26 this rule made +17.9% a year against +17.5% for simply holding BTC, a difference too small to call an edge. In 2020-23 it made +53.5% a year, but at its worst it was down 92%. It is here to show what leverage does, the drops included.

BTC options usually price bigger moves than actually happen, so selling them earns a premium. Each Friday after the 08:00 UTC expiry, this page sells a one-week at-the-money straddle (a call and a put at the current price) on Deribit's live bids, 0.1 BTC each, Deribit's minimum. It only sells when the options are priced for more movement than BTC had over the past week; otherwise the week is skipped. Once a day it offsets the price-direction risk with the BTC perpetual, so what is left is the volatility bet. Fees and hedging costs are included. Paper only, no orders.

Why watch-only: on 284 weeks of real Deribit prices this rule made +0.26% a week in 2024-26 (worst week -5.2%, biggest drop -8.8%), but with a t-score of 1.33 it could still be luck, and it failed the bar set before the test (t above 2). Buying the same options lost reliably (t -2.93).

Every council approval is checked against the real Binance price 4 hours later: was the price on the side it chose? A coin flip scores 50%. This measures the council's direction skill on its own, apart from profit - the exits decide how often a trade wins, this shows whether the direction itself was right. The council's learning uses this same measure.

Every row is a real council decision on live market data. A blocked decision is the system working: the expected-value gate refuses trades whose measured expectancy cannot clear the round-trip fee, however confident the council is.

Every entry and exit is priced from the live Binance mainnet order book. Fees, lot sizes, margin, funding, stop slippage and liquidation follow Binance's own rules, and every exchange failure is counted. No real money is at risk: no API key is used and no order is sent. A new trade is refused if it would leave the account closer than the guard minimum to liquidation.