Glossary

A market where a share pays $1 if an outcome happens and $0 if it does not. The price of a share is the market's probability estimate.

The two sides of a Forecast round, the same idea as YES and NO. One UP plus one DOWN always redeems for exactly $1, so their prices sum to $1. Fortell stores UP as the yes side and DOWN as the no side.

One short market with a fixed open and close, asking whether an asset's price finishes higher than it started. Jupiter Forecast runs 5-minute and 15-minute rounds continuously.

The moment the round's result becomes final and the winning side is worth $1 a share. On Jupiter Forecast the result is written into the round account on chain and never disappears from it.

The determination of which side won. Up/Down rounds resolve from the price at the round's close, recorded on the round account as the up side's settlement price in basis points: 10000 means UP paid $1, 0 means DOWN did.

The stablecoin Jupiter Forecast rounds settle in, and the unit every dollar figure on Fortell is denominated in. Fortell does not use the CASH token; that belongs to World, which is a separate protocol Fortell cannot trade.

Depositing $1 of collateral mints one UP and one DOWN share, and burning a matched pair returns the $1. The AMM calls a conditional-token program to do this set math internally. It is what makes the two prices sum to $1 and behave like probabilities.

Exchanging winning shares for collateral after settlement. Losing shares are worth nothing. On chain this is the venue's RedeemOutcome instruction, which burns a resolved outcome token for its collateral.

The separate, later step where a funded trader takes profit out of their seat. Redemption settles a position inside the seat; a claim moves money out of it. On Fortell the trader signs their own claim and nobody has to approve it.

The price read as a percentage. A 30c share implies 30%. If your own estimate is higher, the share is value; if lower, it is overpriced.

Whether prices match reality across many markets. Perfectly calibrated 70c contracts win exactly 70% of the time. Calibration is a property you measure, never one you assume.

What a position is worth on average: probability times payout, minus cost. Positive EV is the only defensible reason to enter.

Outcome minus entry price, per share, on a resolved call. Win at 40c and edge is +60c; lose and it is -40c. Averaged over many calls, edge is the input to every score Fortell computes.

The round's closing price minus your entry price. Beating the close consistently means you saw fair value before the market did. Fortell records CLV on every scored call where a closing price is available, alongside edge.

Margin a venue builds into its prices. Do not assume any venue runs at zero: the Forecast AMM's account layout includes a protocol fee account, and your fill price on Fortell also carries simulated impact.

The amount your fill price moves against you. Fortell's challenge engine charges a 0.5c base cost plus a size term, capped at 5c, added on a buy and subtracted on a sell, so a small round trip costs about a cent.

The side priced above 50c. Favorites being systematically overpriced is a real effect in some betting markets and a claim that has to be measured per venue. Fortell has published no such measurement for Jupiter Forecast.

A low-probability side. On Fortell they are largely out of reach: challenge and funded orders may only fill between 0.10 and 0.90, so the cheap tail is not tradable here at all.

The 0.10 to 0.90 range an order must fill inside. Entries near 0 or 1 are lottery tickets rather than forecasts, and mean edge scores a 1c buy that wins as +99c, which would let one near-certainty carry a whole score.

The venue Fortell trades. Its rounds are created by a market program and traded on the bisonfi AMM, program 2DNbzPochEcyCcWMbL4d9S3u9QqQEj5bbe6cSZFvKsbh, which Jupiter's own router labels BisonFi Predict.

A program that quotes both sides and takes the other side of every trade. There is no order book on Forecast: you submit a swap and it fills or it does not.

A venue a program can trade at without anyone's blessing. Measured on Forecast: a real swap needs exactly one signer, the trader at account 0, so a Fortell vault's own program address can sign for itself. That probe is why funded seats can trade here at all.

The gap between what you pay buying and receive selling. On Fortell's challenge engine the gap is the impact charge applied in both directions, so trading in and out costs roughly a cent that holding to settlement does not.

Price movement caused by your own order size. The challenge engine simulates it from the round's recent volume rather than filling you at the headline price, because a real fill would move too.

How long ago the venue's price last changed. An AMM quote is frozen between real trades while the underlying keeps moving, so a stale quote can be a decided outcome at an undecided price. A paid challenge refuses to fill on a quote older than 30 seconds; free practice keeps a looser tolerance because nothing is graded.

How much size a round can absorb without moving. Fortell does not publish a measured depth figure for Forecast. The rails cap your size long before depth does: $20 per round on Standard, $60 on Pro.

The source of truth a market resolves against. For Up/Down rounds it is the asset price at the round's close, written into the round account by the venue.

Whether a funded Fortell seat could route a real trade to a venue. It mirrors the vault program's on-chain allowlist and is the single rule the market board is filtered on: a challenge may only list markets a funded seat could trade.

A separate Solana prediction protocol (prediCt, Janus, DFlow and the CASH token). Fortell does not trade it and does not list it. Tested by execution on a mainnet fork on 2026-07-30: the vault's own signature at account 0 was accepted, and Janus then refused the call for a different missing signature that has not been identified. Janus is only ever reached through DFlow in real traffic.

The capital you can lose entirely, managed as one unit. In a challenge the bankroll is fixed for you: $500 on Standard, $1,500 on Pro, identical for every trader on that tier.

Optimal stake sizing given your edge: risk a fraction of bankroll proportional to your advantage. Most professionals use a half or a quarter of it, because overestimating your own edge is the normal case.

The randomness in outcomes even when the decisions were right. Binary markets are high variance: a genuine 60% edge still loses four times in ten.

The fall from your equity's running peak, as a fraction of that peak. It ratchets: once recorded, recovering does not erase it.

The equity level a buy may not leave you below. It is a refusal, not a grade after the fact: the order is rejected. On a funded seat the program makes the same check against the cash the seat is left holding and fails the whole transaction, so the breach never stands.

A cap on realised loss inside one UTC day: $30 on Standard, $90 on Pro. It bounds the order too, not just the day, so a buy whose worst case would breach the cap is refused rather than permitted and then punished.

The most cost basis you may have at risk in a single round: $20 on Standard, $60 on Pro. Selling part of a position releases the basis proportionally, so a partial exit gives headroom back.

The most cost basis you may have at risk across every open round at once: $100 on Standard, $300 on Pro.

20 seconds between opening and closing a position, so the scored-call count cannot be farmed by trading in and out.

No new entries in the last 30 seconds of a round. Late entries read a decided result rather than forecasting one.

Losing the bankroll before the edge can express itself. In a challenge, ruin is not gradual: a drawdown or daily breach ends the run the moment it happens.

Betting bigger and worse after losing. It shows up in the data as size rising after drawdowns, which is the opposite of what the sizing-discipline statistic rewards.

Holding both sides of the same round to cut variance. It is allowed, but a call whose opposite side covers more than half of it is dropped from scoring, so hedging cannot manufacture a safe scored call.

Fortell's 0 to 100 forecast score. It is a logistic over the lower confidence bound of your mean edge, multiplied by a sample-size shrinkage of n / (n + 20), so a tiny sample cannot reach a high score no matter how good it looks.

One resolved (round, side) position on one ledger, with entry price, size and outcome recorded. It only counts if it cost at least $0.50, its hedge ratio is at most 0.5, and its cost-weighted entry was at least 30 seconds before the round closed.

avg(edge) minus 1.645 standard errors, the 90% worst case for your true mean edge. It is what the fscore is computed from, so evidence, not just performance, moves your score.

The same quantity as the lcb, in plain language: how bad your real edge could be and still be consistent with your record. A positive worst-case edge over a real sample is the strongest single piece of evidence of skill.

The n / (n + 20) factor that pulls a small sample's score toward neutral. Ten calls keep about a third of their apparent signal; two hundred keep 91% of it.

Whether your edge survives size, measured as size-weighted edge against plain average edge. If your bigger positions do worse, your capacity is small.

The share of your active days with positive average edge. Skill repeats; luck clusters.

The correlation between your stake and your realised edge. Positive means you bet bigger when you turn out to be more right, which is the entire point of varying size.

The number of scored calls behind a statistic. Ten wins prove less than 60% over two hundred calls, and every Fortell score is built so that is arithmetically true rather than merely advised.

The paid 14-day evaluation: $75 for Standard, $150 for Pro. You trade a dedicated ledger seeded with the tier's capital at live venue prices, and the four pass criteria are fixed at purchase.

The process that decides pass or fail. It runs off chain every 300 seconds and writes the result to a database; no program validates a grade, because the inputs are a 14-day fill history and venue prices. /api/challenge/calls exists so you can replay your own grade call by call against the same SQL.

A vault account holding $500 or $1,500 of pool capital, on which a passed trader is named as a trade-only delegate. Custody stays with the vault's own program address, so the trader can trade the capital and can never withdraw it.

The human trader named on a funded seat. Lowercase in the program source, and always a person: there is no autonomous trading bot in the product.

The instruction a funded trader calls to take profit out of their seat. It pays 70% to the trader, 10% to Fortell and 20% to backers in one transaction, and returns the seat's equity to its tier size.

No open position on the seat. Deposits, settlement and claims are all allowed only when flat, because a seat's equity equals its collateral balance only when nothing is open. Rounds are 5 to 15 minutes, so the wait is short.

Fortell's own cash, reserved per seat at the seat size times its maximum drawdown, so $125 behind a $500 seat and $300 behind a $1,500 one. A seat's loss is taken out of it before backer capital moves, and a settlement that the buffer cannot cover is rejected rather than charged to backers.

Refundable first-loss capital a funded trader may optionally post on their own seat. It is separate from Fortell's pool-level buffer and is not required to be funded.

Trading with yourself or in circles to fake a record. The scoring pipeline nets hedged pairs out, floors dust positions, and punishes thin samples through the confidence bound.

Splitting activity across wallets to reroll luck. Confidence-bound scoring makes it self-defeating: each wallet has fewer calls, so each scores lower, and every challenge is paid for separately.