Funding radar
Funding on Arcus and Lighter side by side, widest gaps first. When the two venues disagree on the same market, one side is paying the other to hold the other way.
Arcus vs Lighter
Weekend gap checker
Wall Street sleeps. Your liquidation doesn't. While US markets are shut, Arcus holds every stock perp inside a price band — the reopen has no band. Here is where each stock perp trades against its Friday close, how much room is left to the band edge, and what a gap would do to a leveraged long.
Stock perps vs Friday close
Verdict Ledger
Every day the agent issues a verdict on the top launches by volume — SKIP, WATCH, or LOOKS REAL — one hour after launch, and resolves it 48 hours later. Every verdict is timestamped, linked to its chain record, and scored against the naive baseline: always SKIP. If the agent can't beat the baseline, the scoreboard says so.
Scoreboard
How the agent decides
The rules are published in full. Anyone can re-run them on the same public data and check the scoreboard. Nothing about the method is hidden — that is the point.
01Timing — T+1h, never T+10s
Launch-second dump prediction is hopeless: a trained model caught 0 of 14 real dumps, scoring below the “always guess safe” baseline. So verdicts are issued one hour after launch, when evidence exists, and resolved at T+48h.
Coverage: the top ~20 launches per day by 24h volume. Pons produces ~727 launches a day; the site says openly that 800+/day is un-coverable.
02SKIP — if any rule fires
- S1 — creator net extraction in hour 1 > 0.01 ETH (the dev is already selling).
- S2 — creator tax = 0 and fleet/bundle share of first-hour buys > 50%.
- S3 — dev has ≥3 prior launches in 30 days, ≥2 dead within 48h.
- S4 — fewer than 5 unique real buyers in hour 1 and first-hour volume < 0.05 ETH.
03LOOKS REAL — only if all hold
- L1 — creator tax > 0.
- L2 — no creator sell in hour 1.
- L3 — fleet/bundle share of first-hour buys < 25%.
- L4 — ≥20 unique real buyers in hour 1.
- L5 — clean dev history: no prior launch dead within 48h.
Anything else is WATCH. If an input can't be computed, the token gets no verdict — counted as “unverdictable”, never guessed.
04Resolution & scoring
DEAD at T+48h if there are no trades in hours 24–48, or the price is down ≥90% from the T+1h price. Otherwise ALIVE.
SKIP is correct if DEAD; LOOKS REAL is correct if ALIVE. WATCH is excluded from accuracy and reported separately. The baseline is “always SKIP” — the share of tokens that died. On this record the baseline is 22.9% (N=580): the record covers the top 20 launches per day by volume — the strongest launches, which survive far more often than the average token — and DEAD is measured from the T+1h price, after the verdict is issued. The often-quoted ~87% counts all launches from launch time; it describes a different population and is not this scoreboard's baseline.
05Funding — units normalised
Lighter quotes funding per 8 hours; Arcus quotes per hour. Both are annualised before comparing:
Arcus: rate × 24 × 365
While a stock market is shut, Arcus funding is fixed at SOFR + 0.5%/yr. Positive funding means longs pay shorts.
06Weekend bands & liquidation
Arcus bands are read per market, live — never hard-coded: bound = anchor × (1 ± IMF × multiplier), ladder 0.5x → 1x → 2x → 4x. Friday's 20:00 ET anchor carries the whole weekend.
Gap history is Incipient's own computation from Yahoo daily bars (Monday open vs prior Friday close), with N and window stated next to every number.
Trust, by construction
Falsifiable by design
Every verdict is a prediction with a public resolution. The scoring is the product — the same model as a forecasting platform, applied to launches.
Tamper-evident record
Verdicts are committed as JSON to a public git repo; the commit history is the timestamp seal. A verdict cannot be edited after the fact without the edit being visible.
No custody, no keys, no execution
v1 is read-only. The agent cannot lose user money because it never touches it — no swaps, no perps orders, no bridge. That is a deliberate, stated limitation, not a missing feature.