This article is part of the SaskPoly story. Read how it all started.
The Loop, Part 5: The Gates — Trading Rules That Grade Themselves
The sports loop's discipline, pointed at trading desks: two locked rule sets — a long-only oil gate and an event-day MSTR short gate — rendered as software that says GO or NO-GO, backtested against years of one-hour candles, honest about a three-trade sample and a market that took a week to find.
The same machine that grades sports predictions turned out to grade trading rules just as well. Two rulesets the human trader had written down — one for longing oil, one for shorting MicroStrategy on economic-release days — are now gates: software that checks the market every tick, says GO or NO-GO with the reasons attached, and keeps a ledger the same way the sports loop does. This post is both gates, including the week the oil gate confidently monitored a market that did not exist.
The gate pattern
A gate is a trader's locked rule set rendered mechanical. The philosophical content is one sentence: a rule you keep in your head is a rule you will renegotiate at the worst possible moment. The software version cannot renegotiate. It computes, verdicts, records, and shuts up. Both gates live on a free public API (Hyperliquid — no key, no account, prices straight from the browser to the page) and both have browser tools on the loop's web dashboard that refresh every sixty seconds and say the verdict out loud when it flips.
Gate one: the oil long gate
The locked ruleset, in full: long oil perps only; the one-hour trend must be up and RSI-14 at 50 or better; every trade needs a written plan paying at least 3-to-1 (risk a dollar to make three, validated at entry — the tool physically rejects a worse plan); hard breakers of three round trips per day and a two-loss lock per coin; and a kill switch — one hourly close below the 100-hour moving average and the gate is closed, full stop, no appeals.
The gate first shipped in September and immediately reported its most useful verdict of the month: CLOSED — no oil perps exist on the exchange. All 234 listed markets, and not one of them tracked crude. The tool sat there honestly monitoring nothing for days, because a gate that hallucinates a market is worse than a gate that says no.
Then the correction, which is the part worth publishing. Hyperliquid's
default board excludes builder-deployed markets; the oil perps lived on a
builder DEX inside the same exchange — 120 additional markets including
Brent crude, gold, natural gas, stock indices. The rewiring took one
parameter (dex: "xyz") and a naming lesson: WTI exposure on this venue
is the CL contract — the front-month CME crude future — because no market
named WTI exists. The gate's first real GO came on 2026-10-02: Brent RSI
58.5, price 102.26 against a rising 100-hour average at 99.12. Trend, RSI,
plan. Go. The kill switch and the 3-to-1 plan validation were doing their
jobs months before a single GO arrived, which is the correct order — safety
rails first, signal whenever.
Gate two: the MSTR event-short gate
The human's manual trade, as observed in the wild: on US macro release days — Non-Farm Payrolls, CPI — MicroStrategy's stock-populated token tends to pop and fade, so the rule is short the pre-market overbought condition, but only when the one-hour RSI reads above 70. That "only" carried the whole edge, and the backtest proved it.
The simulator walked every available one-hour candle from the token's listing through October — 14 macro event days. Shorting every event day unconditionally, the disciplined loser version of the trade: 6 wins, 8 losses, −11.61%. Macro-release days alone are a bad trade. Adding the RSI filter — the rule as actually written — triggered only three times, all Non-Farm Payrolls days, and went 2-1 for +1.76% with end-of-day exits. The filter is not decoration on the trade. The filter is the trade. An event day without the overbought condition is a coin the house shaves.
The caveats ride in the same sentence, because a 3-trade sample is a 3-trade sample: CPI dates are weekday-approximated, funding and fees are excluded, and anyone upgrading "+1.76% on three trades" into a lifestyle is doing statistics cosplay. The ledger will accumulate or embarrass the gate the same way it does everything else — over time, in public, row by row.
The finding that changed the trade itself
One research pass deepened the rule instead of just grading it. Minute-level anatomy of the October 2nd NFP release showed the release pops first — day high 170.12 at 9:33 AM New York, 2.9% above the prior close — and only then fades (156.02 by 2:40 PM). The mechanical pre-release short sat 2.8% underwater at the high before the fade paid it. Same trade, better entry: shorting the one-minute upper-wick rejection after the pop filled near 169.64 and returned +5.8% by the close, against +3.0% for the mechanical version. A rule about when survived contact with a rule about where, and the tool now carries a pop-watch panel — post-release high, distance off the high, a live wick detector — for the next event day. One day of data, stated as one day of data. The forward collector is running.
The shared spine
Stand the five parts of this series side by side and the shape is one machine wearing five hats:
- The clock keeps the belt moving and files a report every six hours.
- The ledger writes every prediction down before the event and grades it after. /articles/the-loop-part-2-the-ledger
- The rules keep a cold model honest while it calibrates, and mark their own mistakes forward. /articles/the-loop-part-3-the-rules
- The teaser engine hunts pricing errors instead of opinions. /articles/the-loop-part-4-the-teaser-engine
- The gates keep a trader's rules at arm's length from a trader's moods.
None of it publishes without a human. None of it wagers a dollar. All of it grades itself in writing. The bet is not on any game, coin, or crude contract — the bet is that written rules, graded honestly, beat talent loose in the wild.
The rule is easy. Keeping it is the product.