Negative reconciled bootstrap samples even when nonnegative = True

[Aorus42]

What happened + What you expected to happen

Observed behavior: many reconciled bootstrap samples are negative even when nonnegative = True in the hierarchicalforecast.methods and despite base forecasts and prediction intervals being positive.

Expected behavior: strictly positive values of reconciled bootstrap samples when nonnegative = True.

No warnings or error messages were returned.

Versions / Dependencies

Python version: '3.10.18 (main, Sep 2 2025, 14:15:06) [MSC v.1944 64 bit (AMD64)]'
pandas v2.3.2
hierarchicalforecast v1.2.1

Reproduction script

summing_matrix.csv
yfitted_df.csv
yhat_df.csv

import pandas as pd
import hierarchicalforecast.core as hf_core
import hierarchicalforecast.methods as hf_methods

# Load example data
tags = {
    "Country": ["Canada"],
    "Country/Province": ["Canada/Ontario"],
    "Country/Province/City": ["Canada/Ontario/Ottawa", "Canada/Ontario/Kingston"]
}

summing_matrix = pd.read_csv("summing_matrix.csv")
yhat_df = pd.read_csv("yhat_df.csv")
yfitted_df = pd.read_csv("yfitted_df.csv")

# Reconcile
hrec = hf_core.HierarchicalReconciliation(
    reconcilers = [hf_methods.MinTrace(nonnegative = True, method = 'mint_shrink')]
)

yrec_df = hrec.reconcile(
    Y_hat_df = yhat_df,
    Y_df = yfitted_df,
    S = summing_matrix,
    tags = tags,
    level = [90],
    intervals_method = "bootstrap",
    num_samples = 1000,
    seed = 1,
    id_col = "UniqueId",
    time_col = "Date",
    target_col = "Quantity"
    )

Issue Severity

High: It blocks me from completing my task.