Termination Conditions — References

Termination as a formal problem, and the one setting where agent completion is checkable without argument.

Knowing a distributed computation has finished

• Termination detection for diffusing computations (Dijkstra and Scholten, 1980) — the classic algorithm for establishing that a distributed computation has genuinely finished, which is what a quiescence check approximates. Cited by name.
• Derivation of a termination detection algorithm for distributed computations (Dijkstra, Feijen and van Gasteren, 1983) — the token-based formulation, and a precise statement of why an observer cannot conclude termination from local quiet alone. Cited by name.

Limits of deciding whether something will finish

• On Computable Numbers, with an Application to the Entscheidungsproblem (Alan Turing,
  1936. — the undecidability of halting, and therefore the reason an external ceiling is a necessary component rather than an admission of defeat. Cited by name.

Completion that can actually be checked

• SWE-bench: Can Language Models Resolve Real-World GitHub Issues? (Jimenez et al., 2023), arXiv:2310.06770 — a benchmark whose success criterion is a test suite passing rather than a judgement, and the clearest demonstration of what a mechanically checkable postcondition buys.

No verified reference is recorded for terminal-reason taxonomies in agent systems. The vocabularies in use are internal engineering conventions rather than published standards.