Internal: persist all collected annotations for the run.
Phoenix's experiment_evaluations endpoint is keyed by
(experiment_run_id, name) so two annotations with the same name on
the same run race each other. We collapse duplicates by name (last
wins) up front, which makes the final state deterministic; the
remaining writes target distinct names, so they post in parallel.
Internal: persist all collected annotations for the run.
Phoenix's
experiment_evaluationsendpoint is keyed by(experiment_run_id, name)so two annotations with the same name on the same run race each other. We collapse duplicates by name (last wins) up front, which makes the final state deterministic; the remaining writes target distinct names, so they post in parallel.