Facility-level care data: added sense check and indicators

[kathsherratt]

I've been doing some work scraping facility indicators from the INSP sitreps, and triangulating with the WHO AFRO sitreps and open source checks on facility names/locations: over in bvd-capacity.

This

• reads the INSP occupation tables into a province capacity series, independently of the scan and blind read behind fix(treatment): province beds, occupancy and admissions on one capped basis #969
• It also holds a facility register built from the situation reports. Overview and methods: https://epiforecasts.io/bvd-capacity/

Agreement with this repo's numbers (#969)

Against #969 at 126d07f, SitReps 018 to 080 (1 June to 2 August). Note the bvd-capacity is out of date in the most recent sitreps extracted (... ran out of tokens).

bvd-capacity	#969	cells	agree	#969 blind read only
patients_in_isolation	patients_isolated	152	152	0
admissions (24h)	admissions_24h	150	150	0
deaths_in_facility	deaths_incare_24h	147	147	147
bed_occupancy_pct	occupancy_rate_pct	124	124	9
discharges_recovered	0	120
beds_capacity	beds	44	44	0

• TLDR; near-exact agreement. SitRep 057 (10 July) is in neither fix(treatment): province beds, occupancy and admissions on one capped basis #969 read. bvd-capacity has it for Ituri, Nord-Kivu and Sud-Kivu: 543 Nord-Kivu at 135.2% occupancy.
• The corrections fix(treatment): province beds, occupancy and admissions on one capped basis #969 settles against the PDFs (SitReps 081 to 083 admissions; 092, 093 and 095 rates) are after 2 August. Run against main at 442b059, only those six cells change.

Details at:
epiforecasts.io/bvd-capacity and code at: R/33_compare_bos.R joins the two on SitRep and province and writes checks/compare_bos_province_care.csv.

Added in bvd-capacity

| Data | Coverage | Bears on |
| Province escapes (absconds), 24h | 151 cells, SitReps 018 to 080 | #969 forecasts abscond exits natnce split |
| Patients in a bed (Patients présents au lit, J-1), distinct from patients in isolation | 155 cells | #95ccupancy figure counts |
| Facility register: eacholation centre, its provinceand health zone, and its opening interval | 70 treatment centres, 10 in Nord-Kivu;SitReps to 122 (13 Septembe|
| WHO AFRO capacity figures, with a preferred figure where a document gives two | 28 June to 20 September, we it mostly restates INSP(median ratio 1.00 where both print). CC BY-NC-SA 3.0 IGO, so read by path rather than copied in |

Added in bvd-capacity

Data	Coverage	Bears on
Province escapes (absconds), 24h	151 cells, SitReps 018 to 080	#969 forecasts abscond exits nationally; this is the province split
Patients in a bed (Patients présents au lit, J-1), distinct from patients in isolation	155 cells	#958's question of what the occupancy figure counts
Facility register: each treatment, transit and isolation centre, its province and health zone, and its opening interval	70 treatment centres, 10 in Nord-Kivu; SitReps to 122 (13 September)	#958 and #920, below
WHO AFRO capacity figures, with a preferred figure where a document gives two	28 June to 20 September, weekly	A cross-check only: it mostly restates INSP (median ratio 1.00 where both print). CC BY-NC-SA 3.0 IGO, so read by path rather than copied in

Potential fit with planned work

• fix(treatment): province beds, occupancy and admissions on one capped basis #969: the comparison above could run as a check whenever the province blocks are advanced. (e.g. it reads a checkout of this repository by path)
• Evidence for the effective-beds rule (fix(treatment): province beds, occupancy and admissions on one capped basis #969), other than the occupancy rate. Per-province admissions forecast bounded by province beds #958 asks why Nord-Kivu holds more patients than its printed beds. The register dates Nord-Kivu openings, for example CTE ITAV (after 20 August) and CTE Matanda (after 5 September), so it can test whether centres opened while the printed bed count stood still. Not yet checked.
• Opening intervals by province give dates a time-varying share to step at. e.g.: Province bed occupancy: estimates page caps at printed beds, forecasts use uncapped shares, Nord-Kivu shows a spurious shortfall #920: a fixed capacity share cannot follow Nord-Kivu from 141 to 308 beds.
• Patient escapes and patients in a bed could belong in the candidate-signal register if they are not fitted (data: register of candidate SitRep signals not read by the model #799)

Drafted by @kathsherratt, co-authored-by @kathsherratt-via-llm