A transactional ledger reserves statistical risk before a test, then records it as spent. Forks delegate only remaining budget; merging transfers existing grants without creating new ones. An opt-in gate uses a bounded sequential test receipt.
Contracted infinity
Carry a validation budget through an evolving lineage.
Repeated adaptive comparisons create repeated chances to promote a false improvement. Forks, merges and rollbacks must not mint a fresh risk allowance.
The risk ledger and opt-in Trinity and Biocenosis promotion gates are connected for scoped tests.
Data provenance checks, more promotion paths and long empirical campaigns remain open.
Two branches test candidate procedures. Their allocations come from one parent grant; merging their results cannot issue another copy of that grant.
Run long adaptive campaigns with null and real improvements, plus fork, replay and merge attacks. Measure false promotions, true gains and evaluation cost.
The bound holds only if each test is valid under its actual selection, data and stopping rules. The ledger alone does not establish real-world safety.
This status describes the slice documented at the reviewed commit. These links open runtime sources, not a live demonstration.