Totetape places no wagers. It runs frozen paper strategies against the archive since 2026-08-09 and publishes what happened, including the parts that did not work.
S10 follows late steam while every active strategy either fades or independently prices the market. It is designed to lose after takeout, excluded from ranks and trends, and kept in the population on purpose.
It is a two-sided integrity alarm. Once enough bets settle, an interval that excludes its expected 0.80 return in either direction requires an explanation. With 335 settled, the published interval is the alarm state.
Strategies rank on the lower bound of a 95% interval around flat-stake ROI, never on the point estimate. A high observed return with weak evidence cannot outrank a record whose lower bound is stronger.
Every decision uses one flat unit. Favorite–longshot calibration and the 15/1 decision-odds ceiling are hard engine rails. Win, place, and show strategies rank only inside their own cohorts.
The pooled figure includes every strategy ever run, including retired rows. Retired strategies remain browsable with their record, reason, and lineage, so weak experiments do not disappear from the denominator.
The promise is published before it costs anything. A population of hypotheses paying full takeout is expected to lose; that it does is evidence that nothing has been quietly pruned.
Each UTC day publishes a SHA-256 commitment for the canonical paper-decision log. The file contains the decision count, that day’s hash, and a chain hash derived from the previous day.
/public-data/commitments/YYYY-MM-DD.jsonday_sha256. It must equal the value printed in the file.sha256(previous chain_sha256 + day_sha256). It must equal the published chain_sha256; walk the chain back to the first file.