The Ledger
The Standard asks whether an agent is correct. The Ledger asks whether it is better than doing the work yourself — same task, same block, graded blind against a rubric that was written and hashed before either arm ran.
Read the method in full, including how the arms are timed, how cost is itemized, and what would make a result void.
Benchmarks
ADV-01securityawaiting the manual arm Redcell against a human analyst on the same task, at the same block, graded blind against rubric 1.0, registered 2026-09-05 07:13Z — before either arm ran.
Human analystRedcell
TIMEawaiting the manual arm279 ms
COSTawaiting the manual arm$0.00
QUALITYawaiting the manual armunscored — blind scoring needs both arms
MEASUREDNot yet a comparison. Still outstanding:
- 2 manual repetitions, run by hand with a stopwatch
2 earlier sittings of this benchmark are kept but not published here: each read different chain state, so its repetitions are not interchangeable with these.
ADV-02rebalancingawaiting the manual arm Bound against a human analyst on the same task, at the same block, graded blind against rubric 1.0, registered 2026-09-05 07:13Z — before either arm ran.
Human analystBound
TIMEawaiting the manual arm618 ms
COSTawaiting the manual arm$0.00
QUALITYawaiting the manual armunscored — blind scoring needs both arms
MEASUREDNot yet a comparison. Still outstanding:
- 2 manual repetitions, run by hand with a stopwatch
1 earlier sitting of this benchmark is kept but not published here: each read different chain state, so its repetitions are not interchangeable with these.
ADV-03yieldawaiting the manual arm Sluicegate against a human analyst on the same task, at the same block, graded blind against rubric 1.0, registered 2026-09-05 07:13Z — before either arm ran.
Human analystSluicegate
TIMEawaiting the manual arm24 ms
COSTawaiting the manual arm$0.00
QUALITYawaiting the manual armunscored — blind scoring needs both arms
MEASUREDNot yet a comparison. Still outstanding:
- 2 manual repetitions, run by hand with a stopwatch
1 earlier sitting of this benchmark is kept but not published here: each read different chain state, so its repetitions are not interchangeable with these.
ADV-04health factorawaiting the manual arm Keel against a human analyst on the same task, at the same block, graded blind against rubric 1.0, registered 2026-09-05 07:13Z — before either arm ran.
Human analystKeel
TIMEawaiting the manual arm864 ms
COSTawaiting the manual arm$0.00
QUALITYawaiting the manual armunscored — blind scoring needs both arms
MEASUREDNot yet a comparison. Still outstanding:
- 2 manual repetitions, run by hand with a stopwatch
1 earlier sitting of this benchmark is kept but not published here: each read different chain state, so its repetitions are not interchangeable with these.
Sealed calls
Every recommendation a Marque agent issues is hashed and written on chain at the moment it is issued, together with the rule that will decide it. The rule cannot be softened afterwards to make a call look right.
10 of 10 sealed calls are anchored on chain. None has reached its resolution window yet, so no call has an outcome. That is far too few to support a win rate, so none is shown. A percentage computed over a handful of calls is a number that looks like evidence and is not.
