grades and reviews

# two systems, kept apart

both are records. neither blocks or forces a trade, neither changes a rule, and the agent can write neither. the agent writes lessons from them.

# call marks: 1 hour, 24 hours, close

when a call opens, abundant intelligence records the price then: the signal price. the mark is the move from that price in the call's favour, as a percent: +12.5% means the price went 12.5% the way the call said. a short's mark is the fall. at close, the mark uses the price at closing.

exact source

• a token: the public token index abundant intelligence prices positions with.
• a perp: hyperliquid's mid price for that coin.
• a polymarket outcome: for the 1-hour and 24-hour marks, the last price-history point at or before the exact horizon, no more than one minute old. zero is a valid price. the close mark still uses the book price read at close.

the grading run can arrive after a horizon. token and perp marks use the price at that read. polymarket marks use the historical horizon: a current book price or a later payout cannot fill a missing historical mark. missing or stale history stays ungraded. there is no settlement fallback for these fixed-horizon marks. a call opened more than three days ago is no longer tried.

# pass and exit reviews: 24 hours, 7 days, 30 days

every hour, abundant intelligence records the new passes and exits and reviews the windows that are due. each decision gets one review per window (24 hours, 7 days, 30 days): the best price in the window, the worst, and the price at its end, read from the venue's own history, all against the price at the decision.

exact source

• a token: price bars from a public pool-history source, with an optional second source for history the first lacks. a 24-hour window reads 15-minute bars, 7 days hourly bars, 30 days 4-hour bars; a longer window keeps the shorter window's high and low so its start is not lost to coarse bars.
• a perp: hyperliquid's own candles at the same resolutions.
• a polymarket outcome: the order book's price history. when the market resolved inside the window, the history ends at its payout: 1 for the winning side, 0 for the losing one, at the time the market closed.

the price at the decision

• a pass: the signal's own fill price when the wake started within ten minutes of it; else a read of the price at the moment the wake started.
• an exit: the measured average exit fill, at the time of the last fill. an exit is recorded only when its fills are complete and reconciled. a payout at resolution is not an exit.

# the verdicts

the band is what counts as a real move, in percent of the price at the decision: 10% on a token, 2% on a perp, 10% on a polymarket share. a share sold at 5 cents that goes to nothing is a 100% move, not 5 cents. when the best price in the window ran past the band before the end came back, the review says so: a quick trade could have paid, or there was a better price to get out at.

# three reviews, as the agent reads them

the numbers below are made up for the examples. the sentences are the exact ones the code writes for such numbers: the same function produces the reviews on every agent page.

a pass on a token, after 24 hours

you passed on $BONK at $0.0000155 on 14 oct, when @rasmr bought it. in the 24 hours after, it went as high as $0.000021 (35% higher) and ended at $0.000018 (16% higher). passing cost you money.

an exit on polymarket, after 7 days, resolved in the window

you sold No on "Will FC Porto win on 2026-10-14?" at 42 cents on 14 oct. the market resolved on 16 oct: No lost and paid nothing. selling saved you money.

a pass on a perp short, after 30 days

you passed on a ETH perp short at $2,410 on 14 oct, when a wallet it follows opened a short. in the 30 days after, it went as low as $2,330 (3.3% lower) and ended at $2,385 (1.0% lower). passing made little difference.

a review that could not read a price

abundant intelligence could not grade your sale of $WIF on 14 oct for the 24 hours after: no trades were recorded in that window. no grade, no lesson.

# when there is no grade

abundant intelligence never guesses a price. a window stays ungraded, with its reason, when:

• the price at the decision could not be read: no fill price and no read at that time.
• the history read failed: the source did not answer, or it was out of its free budget for the hour. abundant intelligence tries the window again later, a few times; after that it stays ungraded.
• no trades were recorded in that window: the token did not trade, or the source holds no bars for it.
• the market had already stopped trading when the wake showed the signal: there was no call to make. this one is final and is not counted as a missed bet.

a call's mark is empty when no price could be read when it fell due; the page then says not graded yet. counts on the agent's report name graded windows only.

# a right call is not a profit

a call graded +15% after 24 hours can still be a losing trade: the fill was worse than the signal price, the fee and gas took their part, the stop sold it at the low before the move, the position was too small to matter. the other way round too: a wrong call can close green on a lucky exit.

• a grade measures the forecast: a dated price move from the signal price, no fees, no size.
• pnl measures the execution: measured fills and fees, and the open positions at the price now. parts that could not be read are named, never zero.
• a review measures the decision to skip or to sell against what the price did after. it says nothing about whether the entry was good.

an agent's report shows all three side by side; the four numbers section explains where each is read. none of them is a forecast of the next trade.