Done Is Not the Same as True
A worker can correctly report that it finished an action. That still does not automatically settle what is now true in the system it tried to change.
“Completed” describes an execution. “True” describes the affected durable or external state.
A system says “completed”. The next question is different.
The natural question is: so the change happened?
Maybe. But that is a second proposition.
An executor can be completely accurate about its own work: it accepted an authorized attempt, ran a process or provider call, observed output and returned a result. Those are meaningful facts. They do not automatically settle every durable fact in the system that action was trying to change.
What the executor can say authoritatively
- Did I accept this attempt?
- What did I run?
- Did my process complete, fail, time out or get cancelled?
- What output or direct observation did I produce?
- What result claim did I return?
Baranes keeps those facts precise instead of flattening them into a generic “untrusted result”.
Why the distinction matters
Execution outcome and effect state can diverge. A process can finish successfully while a concurrent change alters the final durable state. A process can fail after creating part — or all — of the intended effect. A timeout can happen after an external system has already accepted a mutation. A result can be lost even though the effect exists.
PROCESS SUCCESS != AUTOMATIC EFFECT SUCCESSPROCESS FAILURE != AUTOMATIC NO EFFECTTIMEOUT != AUTOMATIC NO EFFECTMISSING RESULT != RETRY PERMISSION“So Baranes does not trust executors?”
No. This is not an adversarial assumption. The executor remains authoritative for its own acceptance, execution and direct observations. The boundary appears only when the proposition belongs to another semantic authority.
Practical significance
The distinction prevents operational convenience from erasing uncertainty. A missing or failed result cannot simply be relabeled as “nothing happened” if that conclusion is not established. That matters especially before a retry or another action could duplicate an external effect.
Baranes also does not require every successful execution to be checked by a universal independent verifier. The affected domain applies the evidence rule appropriate to that effect; in some cases the executor’s evidence may already be sufficient.