Runs published but evaluations are missing

Hi there,

I have been running various experiments over the last few days and published my runs on OpenML. I have done this before and it always worked fine. Suddenly there are several problems.

1. When I click on my OpenML profile I cannot see the new runs, even the ones that are a few days old.

2. When I search for the runs on OpenML by run id I find the run, but the evaluation section is empty. ("0 Evaluation measures")

3. When I examine the runs through the python API I am able to retrieve the runs, but the evaluations dict is completely empty. ('evaluations': OrderedDict())

I have used the same code as in my previous experiments for which publishing was no problem at all.

# instantiate new classifier with best parameters 
clf = HistGradientBoostingClassifier()
pipe_best = make_pipeline(SimpleImputer(strategy='most_frequent'), clf)
pipe_best.set_params(**(sh_fit.best_params_))

# run best model on the task
run = openml.runs.run_model_on_task(pipe_best, task)

# publish the run 
run.publish()

Here are a just few of the run ids in question:
10229260
10229261
10229265
10229267

Do you have any idea what could be causing these issues?
Help would be greatly appreciated.

@amueller

Start by reproducing the behavior using the listed run IDs and the Python API calls described in the issue. Compare a recent published run with an older working run, focusing on profile visibility and the empty evaluation data. Done means the cause is identified and published runs show their evaluations both on OpenML and through the API.