APEX
AUTONOMOUS PREDICTION ENGINE

TESTS

WHAT APEX IS TESTING NEXT, WHEN EACH TEST DECIDES, AND HOW. EACH TEST'S PASS AND FAIL RULES ARE WRITTEN DOWN BEFORE ITS DATA IS LOOKED AT, AND EVERY VERDICT STAYS ON THIS PAGE, PASS OR FAIL.

UPDATED NOV 14, 2026 · "ABOUT" DATES MOVE WITH THE DATA
RULES SETDATA COMES INVERDICT POSTED
NEXT CHECKNOV 16 COMING UP8 DECIDED18

HOW THE CHECKS WORK

RULES FIRST

A test's pass and fail rules are written down before any of its data is looked at, so a result can't be argued into a pass afterwards. Reviews and measurements are labelled as such.

DATED IN ADVANCE

Every check is listed here before it runs, with what it decides. Most wait for a set amount of data, so their dates are estimates.

EVERY VERDICT POSTED

Passes and fails both stay on this page and go out in the weekly email. Results themselves are sealed in the daily record.

COMING UP

8 TESTS · IN DATE ORDER

NOVEMBER3 TESTS

  1. ABOUTNOV 16APEX1RE-TEST · RULES SET OCT 29NEXT CHECK

    ORDERS THAT WAIT FOR A FILL

    Would orders that wait in the order book, instead of taking the price on offer, earn more on a slice of APEX's entries? On history the answer wasn't clear, so the same rules run again on sessions since Oct 30, unseen until 1,000 filled contracts.

    DECIDESA pass leads to a small live trial. A fail closes the idea for crypto.
  2. ABOUTNOV 18–20BOTHHISTORY TEST · RULES SET NOV 3

    DO THE TWO UNITS STRUGGLE ON THE SAME MARKETS?

    APEX1's Tuner rates each market every 15 minutes, and keeps rating the ones it has switched off. On the nine coins both units trade, does APEX2, running a different strategy, do worse where that rating is lowest? First on history from before APEX2 started, then on APEX2's own bets at its final checkpoint.

    DECIDESAPEX2 would skip low-rated coins only if its bets on them clearly did worse. Otherwise nothing changes, and that's worth knowing too: the two units don't stumble on the same markets.
  3. NOV 23NEW MARKETSRE-RUN · RULES SET NOV 2

    NEW MARKETS: SECOND LOOK

    First looks on Nov 2 and Nov 7 found nothing to trade in two newly active sets of 15-minute markets. Both run again with more data.

    DECIDESOnly a clear change from the first look would reopen it. Any trial that followed would get its own rules, written down first.

DECEMBER5 TESTS

  1. ABOUTDEC 1APEX1RE-TEST · RULES SET NOV 3

    A SMARTER WAY TO TUNE ITSELF: SECOND LOOK

    On Nov 3, a new way for APEX1's tuner to set more of its own settings wasn't clearly better than the current one on history. The same test runs again, same rules, with a month more of data.

    DECIDESA pass starts a side-by-side shadow run, and trading doesn't change unless that passes too. Otherwise the current tuner stays.
  2. ABOUTDEC 7–21APEX1TEST · RULES SET NOV 14

    WHEN DOES APEX1 TRADE LARGER?

    The metals, energy and currency markets APEX1 trades can take a larger size. Whether APEX1 does is set by a rule written now: its size in a market grows only when that market's own record and fresh results, priced at the larger size, both support it, in stages. If no market is trading larger by early March, the question goes back for review.

    DECIDESWhether APEX1 moves a market up a stage. A market moves up only when every check passes, and back down when one fails.
  3. ABOUTDEC 8–10APEX2TEST · RULES SET OCT 25

    THE SECOND STRATEGY'S VERDICT

    At 600 filled bets, APEX2's trial on the nine coins gets its decision.

    DECIDESPass if the whole 90% range is above zero. If the middle is above zero but the range crosses it, carry on to 1,700 bets. Otherwise stop.
  4. ABOUTMID-DECAPEX1TEST · RULES SET OCT 30

    A TIMING TEST ON COMMODITIES

    Do commodity markets behave differently at one time of day? The test waits for 160 qualifying cases since Oct 13.

    DECIDESTwo parts must pass: the pattern itself, then a price check that the market doesn't already account for it. Both pass: a small trial. Either fails: dropped.
  5. ABOUTLATE DECAPEX2TEST · RULES SET NOV 3

    THE METALS TRIAL'S FIRST CHECKPOINT

    Since Nov 3, APEX2 also runs its trial's signal on five metals markets, scored separately from the coins. It gets its first look at 200 filled bets.

    DECIDESThe metals trial stops if the whole 90% range of its result per contract is below zero. Otherwise it carries on to 600 bets.

DECIDED SINCE OCT 24

NEWEST FIRST · FAILS INCLUDED

An idea has to pass before it's adopted or kept, and many don't. That's the filter doing its job. How APEX is doing right now is on the live page and in the daily record; the story before Oct 24 is being told in the build log.

ADOPTED OR KEPT 10

  1. NOV 14APEX1 ORDER BOOKSIZE KEPT The markets can take a larger size; the results have to earn it first. Two weeks of readings show the metals, energy and currency markets APEX1 trades could have filled orders many times today's size in 84–99% of readings, at very little extra cost. Some smaller crypto coins thinned out fast at the larger sizes, but APEX1's crypto is switched off for now. So depth isn't what holds size back. Size stays where it is until fresh results show the strategy pays at the larger size's cost. A measurement, not a test.
  2. NOV 13APEX2 TESTCONTINUES The second strategy cleared its first checkpoint: after 200 filled bets, the whole 90% range of its result per contract sat above zero, well clear of the stop line, so the coin trial carries on to its verdict at 600 bets, now expected about Dec 8–10. Clearing a checkpoint isn't a pass: only the 600-bet verdict can say that, and the two days since have been rough. Nothing has changed. A live test, rules set Oct 25.
  3. NOV 9APEX2 SIGNALPASSED Both looser versions of APEX2's setup passed: on four years of crypto price history, the extra bets each would add showed the pattern APEX2 bets on in every year, with the 90% range clear of zero, and the prices APEX actually sees leave room for it (196 and 121 price readings). A pass isn't proof of profit: next is whether to run them as a small trial of their own, scored apart from the current one. Nothing has changed yet. A history test, rules set Nov 3.
  4. NOV 4APEX1ADOPTED Letting the Tuner make the call: where a market's record is long enough, the Tuner now acts on its best estimate instead of leaving the market as it is when unsure. On history it did no harm and had a smaller worst dip over the last month, though the deciding stretch gave it little to do. First call: APEX1's crypto stays off for now. A history test, rules set Nov 4.
  5. NOV 3APEX2 SIGNALADOPTED The trial's signal on metals passed its price check: on 39 price readings, the price leaves room for the edge. A small metals trial has run on APEX2 since Nov 3, scored separately from the coins.
  6. NOV 1APEX1KEPT Two-week review of APEX1's current setup. Ahead overall, carried by commodities and FX. Crypto is behind and watched closely, with the tuner deciding when it trades. A review, not a test set in advance.
  7. OCT 29APEX1KEPT Standing down around big economic releases stays on APEX1, except around jobs reports. APEX2 dropped it: it cost the trial its best chances.
  8. OCT 29APEX2KEPT Three guard rules on the trial, checked on data: two kept, and one had too few cases to judge, so it stays as it is.
  9. OCT 24APEX2ADOPTED A new crypto signal passed and became APEX2's trial, live from Oct 25. On metals there were too few cases to judge then (see Nov 3).
  10. OCT 24APEX1ADOPTED At most one market per group starts betting in a session, so markets that move together aren't all bet at once. On ten months of history it kept almost all of the profit and cut the deepest drawdown by about a sixth.

RULED OUT 8

  1. NOV 13APEX1NOT ADOPTED Skipping crypto bets at certain times: two timing rules set on Oct 12 were re-tested on a month of fresh data (Oct 13 – Nov 12). Neither held up. The bets each rule would have skipped weren’t clearly worse than the rest, and both 90% ranges crossed zero. Nothing changes. A re-test, rules set Oct 12.
  2. NOV 7NEW MARKETSNOT ADDED Another newly active market, first look on 60 trading days of its price history: nothing to trade after costs. Second look Nov 23 ›
  3. NOV 3APEX1NOT ADOPTED A smarter way for APEX1's tuner to set itself, tested against the current tuner on three months of history, each using only what was known at the time: its forecasts weren't clearly better (the 90% range crossed zero), so the current tuner stays. A simpler version would clearly have lost money. Second look about Dec 1 ›
  4. NOV 2NEW MARKETSNOT ADDED A newly active group of markets, first look: nothing to trade. Second look Nov 23 ›
  5. OCT 30APEX2 SIGNALNOT ADDED The trial's signal on energy markets, tested on history: no edge there.
  6. OCT 29APEX1NOT ADOPTED A formula for the best entry point: no better than the rule APEX already uses.
  7. OCT 24BOTHNOT BUILT A market-wide risk guard across every market: all six tests failed.
  8. OCT 24APEX1REPLACED APEX1's earlier betting setup lost money per bet. The current setup has run since Oct 20.
FOLLOW THE BUILD
Every verdict, the week's numbers and its sealed fingerprints, about once a week. Never trade signals.
GET VERDICTS BY EMAIL › HOW THE RECORD IS SEALED ›