Availability at level 1b is updated with average of availability (by item_code and month) for levels 1b and 2. Three strategies of weighting are adopted to test calibration -

1. level2 – fully replace Level 1b availability with Level 2 values.

New RF :ResourceFile_Consumables_availability_small_level2.csv

2. national_1b_to_2_ratio – apply a single national-level 1b:2 facility ratio for weighting.

New RF :ResourceFile_Consumables_availability_small_national_1b_to_2_ratio.csv

3. district_1b_to_2_ratio – apply district-specific 1b:2 ratios for weighting.

New RF : ResourceFile_Consumables_availability_small_district_1b_to_2_ratio.csv

ResourceFile_Consumables_availability_small contains the original level 1b and 2 availability.

Hi @tbhallett. As discussed earlier this week, I've added three new versions of the consumable availability RF on this branch to see how each of these performs in comparison with data.
The average availability by level in the 4 cases is as follows -

now running on Azure: long_run_all_diseases-2025-11-03T100729Z

Errors with the workflow have been resolved - this PR's calibration workflow is now running!

@sakshimohan -- I've looked at the calibration runs for this, and nothing disastrous has happened -- so I think it's OK to proceed.

@sakshimohan -- I've looked at the calibration runs for this, and nothing disastrous has happened -- so I think it's OK to proceed.

Many thanks, @tbhallett! I assume any of the four weighting options are OK? In this case, I'll use district_1b_to_2_ratio which weights availability by the ratio of 1b to 2 facilities within each district, because this feels most sound conceptually.

@sakshimohan -- I've looked at the calibration runs for this, and nothing disastrous has happened -- so I think it's OK to proceed.

Many thanks, @tbhallett! I assume any of the four weighting options are OK? In this case, I'll use district_1b_to_2_ratio which weights availability by the ratio of 1b to 2 facilities within each district, because this feels most sound conceptually.

Ok, that's fine to proceed then.

I also made this graph to help (me) remember the difference between that scenario and what's in master:

@sakshimohan -- I've looked at the calibration runs for this, and nothing disastrous has happened -- so I think it's OK to proceed.

Many thanks, @tbhallett! I assume any of the four weighting options are OK? In this case, I'll use district_1b_to_2_ratio which weights availability by the ratio of 1b to 2 facilities within each district, because this feels most sound conceptually.

Ok, that's fine to proceed then.

I also made this graph to help (me) remember the difference between that scenario and what's in master:

This is really helpful. Just to note the reason that the availability is so vastly different for Salima, Ntchisi, Mwanza, Machinga, and Balaka (Facility IDs 2, 54, 70 102, 114) is that no facilities at level 1b were recorded in these districts as per HHFA. This means that the availability at level 1b is equated to that at level 2.
If we want to avoid this extreme case, I can just use the national_1b_to_2_ratio for the weighted average.

@sakshimohan -- I've looked at the calibration runs for this, and nothing disastrous has happened -- so I think it's OK to proceed.

Many thanks, @tbhallett! I assume any of the four weighting options are OK? In this case, I'll use district_1b_to_2_ratio which weights availability by the ratio of 1b to 2 facilities within each district, because this feels most sound conceptually.

Ok, that's fine to proceed then.
I also made this graph to help (me) remember the difference between that scenario and what's in master:

This is really helpful. Just to note the reason that the availability is so vastly different for Salima, Ntchisi, Mwanza, Machinga, and Balaka (Facility IDs 2, 54, 70 102, 114) is that no facilities at level 1b were recorded in these districts as per HHFA. This means that the availability at level 1b is equated to that at level 2. If we want to avoid this extreme case, I can just use the national_1b_to_2_ratio for the weighted average.

I see. I'd leave it to your best judgement on that then.