Refuting model with unrelated variables does not report significance

[nicolassanmar]

I'm trying to get a working example where refuting an estimate provides significance, and refutes it.
I am not having luck with my real dataset, and not even when generating random variables with numpy, so I must be doing something wrong.

# initialize a dataframe with 10000 rows, with random 2 independent variables, 'a' and 'b' with values from 0 to 10000
randomDataset = pd.DataFrame(np.random.randint(0,10000,size=(10000, 2)), columns=list('ab'))

# graph = a->b
graph = ... 

model = CausalModel(data = randomDataset, treatment='a', outcome='b', graph=graph)

estimand = model.identify_effect(method_name="default")
estimate = model.estimate_effect(estimand, method_name="backdoor.linear_regression")

# refute
ref1 = model.refute_estimate(estimand, estimate, method_name="random_common_cause" )
ref2 = model.refute_estimate(estimand, estimate, method_name="placebo_treatment_refuter", placebo_type="permute")
ref3 = model.refute_estimate(estimand, estimate, method_name='dummy_outcome_refuter')
ref4 = model.refute_estimate(estimand, estimate, method_name='data_subset_refuter')

All of those refutations give a p-value of around ~0.9, meaning that we could not refute the estimate.
No matter how much I tried adding different variables to the graph, and trying all kind of nonsense with real data, I could not get any refutations to work.

What am I doing wrong?