RSTAR

How it settles

A market's final average comes from the official daily prints of its month, posted on-chain by a keeper, checked for 24 hours, then folded into a running sum that never changes again.

The CME averaging rule

Worked example: September 2026

DaysRate
September 1 to 16 (16 days)3.63%
September 17 to 30 (14 days)3.88%
average = (16 x 3.63 + 14 x 3.88) / 30 = 3.74667% -> 3.747%

The engine's history test replays all 2,724 official prints from 2016-01-04 to 2026-11-05 through the same state machine the program runs. All 129 monthly settlements match an independent reference computation exactly.

Units on chain

QuantityUnitExample
Rate (print, collar, average)0.01 bp, so 1% = 10,0003.63% = 36300
Date in a postyyyymmdd, the NY Fed effectiveDate20260917
Date in stateUTC day number since 1970-01-01September 1, 2026 = 20697
USDC, LONG, SHORTbase units, 6 decimals1000000 = one pair
LONG payoutUSDC base units per 1,000,000 LONG base units500000 = $0.50

The life of one print

Time (ET, publication day)Event
about 9:00 a.m.the New York Fed publishes the rate for the previous business day
about 2:30 p.m.the last moment it can be revised, only for an error above 1 bp
19:30 UTC (2:30 or 3:30 p.m.) onwardthe keeper may post the print; it waits 24 hours
24 hours after the postthe print folds into the month's running sum and never changes again
  1. Post. post_print(date, rate, final) is signed by the keeper or the authority. The program refuses a post earlier than 19:30 UTC on the day after the print's date, which is after the NY Fed's 2:30 p.m. ET revision deadline in both EST and EDT. Each market queues at most 8 pending prints.
  2. Window. Each pending print waits 24 hours (DISPUTE_WINDOW_SECS = 86400). Inside the window the keeper or the authority can dispute_print it, which stops it and every later print from folding. The authority can correct_print (new rate, lifts the dispute) or void_prints (drops it and every later entry so a misdated print can be posted again).
  3. Fold. Once its window has passed, a print folds: days since the previous print take the previous rate, the print's own day takes its rate. Every post_print folds whatever is ready first, and fold_prints lets anyone fold without posting.
  4. Final print. A print dated the last day of the month is final automatically; a final print also fills the remaining days of the month and closes it.
  5. Settle. settle_market is permissionless. After the final print has folded it fixes settled_avg and long_payout_ppm.

The collar and the payout

At listing the market fixes collar_lo and collar_hi as the target-range midpoint plus and minus 2.5 percentage points, floored at zero.

LONG pays  clamp((avg - collar_lo) / (collar_hi - collar_lo), 0, 1) dollars
SHORT pays 1 - LONG

The largest four-month drift of the monthly average in the 10.8 years of data is 210 bp (2022), inside the 250 bp half width.

If nobody posts

If a month is still open 45 days after it ends (keeper and authority both gone), anyone can call settle_market. The program carries the last folded rate to the end of the month; with no print at all, the listing midpoint stands in and both sides pay $0.50. The market's Fallback flag records that this happened.

Revisions after a print folds

A folded print is permanent. If the NY Fed ever revised a rate after the keeper's post had folded, the market keeps the first published value, which is the CBOT rule for the last day of a ZQ month (SER-8019). The finality guard makes this case very unlikely: a revision can only happen before the keeper is allowed to post.