Data

What’s here

The data on this page come from the situation reports (“sitreps”) that INSP, the DRC’s national public health institute, publishes on the outbreak. The reports count cases by health zone, the district-sized unit the country’s health system is organised into, and health zones roll up into provinces.

Every number and every figure on this page is read from a file the pipeline builds, at the moment the page renders. If a figure has no build rule behind it, or a file hasn’t been built yet, the render stops with an error, so nothing here can be typed in by hand or left over from an old build. To rebuild it all yourself, run make build, make streams, make figures and make compare.

The archive holds 155 situation reports (through 2026-10-23). 145 are read at health-zone level — every report that prints a per-zone table — for 6,998 zone-day rows; the provincial summary covers 140 reports, and three early reports are read from their own prose. Every held report contributes case data at the finest level its pages print.

'data/build/figures/latest/availability.png'

What each report states, at what geographic level (country, province or health zone), and which table layout I read it through.

A few dates are worth keeping in your head as you read that figure. The suspected-case and contact-tracing series stop at the start of June, which is also when counts of confirmed deaths begin. On 13 July the reports start splitting each day’s deaths by how they were found: community deaths, confirmed by a post-mortem swab, and deaths in treatment centres. And sitreps 084 to 086 use a layout that prints no per-zone table at all, so for those three reports I only have province totals.

The reports annotate their own early numbers, and I treat those annotations as data. Nearly every report from 004 through 033 carries a footnote saying its headline cumulative count is still being harmonised, that is, reconciled across the levels of the health system. From sitrep 026 (9 June) the footnote also says which way the error runs. The count leaves out confirmed cases who are still in the community and patients who escaped from care, so it is too low. Sitreps 011 and 012 have a footnote saying the death count itself is underestimated. Later, from sitrep 060 to 083, the reports warn that the last few days’ changes should be read with caution, because notifications arrive late and the data are still being consolidated. Each of these statements is a row in provisional_caveat.csv, whose kind column says which sort of caveat it is: harmonisation_pending (still being harmonised), harmonisation_undercount (being harmonised, and stated to be too low), deaths_underestimated, or recent_data_provisional (the warning about recent days, flagged “caveat” in the index). The index below flags them report by report. This matters if you fit the first four weeks of the outbreak, because you’d be fitting numbers the source itself called provisional, and from 9 June called too low.

Index of situation reports

One row per report. Route is the table layout it was read through (— when no zone table exists); identity asks whether the page’s own arithmetic closes; flags mark the source’s own provisionality caveat, a hand-verified exception, or a rejected route.

Table 1: Index of situation reports

report	date	route	rows	identity	residual	flags
001	2026-05-14	—	0	—
002	2026-05-17	—	0	—
004	2026-05-18	zone_early	8	ok	0	harmonisation
005	2026-05-19	zone_early	8	ok	0	harmonisation
006	2026-05-20	zone_early	10	ok	0	harmonisation
007	2026-05-21	zone_early	13	ok	0	harmonisation
008	2026-05-22	zone_early	13	ok	0	harmonisation
009	2026-05-23	zone_early	14	ok	0	harmonisation
010	2026-05-24	zone_early	18	ok	0	harmonisation
011	2026-05-25	zone_early	18	ok	0	harmonisation, deaths understated
012	2026-05-26	—	0	—		deaths understated
013	2026-05-27	zone_early	21	ok	0	harmonisation
014	2026-05-28	—	0	—
015	2026-05-29	zone_mixed	31	ok	0
016	2026-05-30	—	0	—		harmonisation
017	2026-05-31	—	0	—		harmonisation
018	2026-06-01	zone_cumulative	29	ok	0	harmonisation
019	2026-06-02	zone_cumulative	30	ok	0	harmonisation
020	2026-06-03	zone_cumulative	30	ok	0	harmonisation
021	2026-06-04	zone_cumulative	30	ok	0	harmonisation
022	2026-06-05	zone_cumulative	30	ok	0	harmonisation
023	2026-06-06	zone_cumulative	30	ok	0	harmonisation
024	2026-06-07	zone_cumulative	30	ok	0	harmonisation
025	2026-06-08	zone_cumulative	30	ok	0	harmonisation
026	2026-06-09	zone_cumulative	31	ok	0	stated undercount
027	2026-06-10	zone_cumulative	34	ok	0	stated undercount
028	2026-06-11	zone_cumulative	34	ok	0	stated undercount
030	2026-06-13	zone_cumulative	36	ok	0	stated undercount
031	2026-06-14	zone_cumulative	36	ok	0	stated undercount
032	2026-06-15	zone_cumulative	36	ok	0	stated undercount
033	2026-06-16	—	0	—		stated undercount, route rejected
034	2026-06-17	zone_cumulative	39	ok	0	exception
035	2026-06-18	zone_cumulative	39	ok	0
036	2026-06-19	zone_cumulative	39	ok	0
037	2026-06-20	zone_cumulative	39	ok	0
038	2026-06-21	zone_cumulative	39	ok	0
039	2026-06-22	zone_cumulative	39	ok	0
040	2026-06-23	zone_cumulative	39	ok	0
041	2026-06-24	zone_cumulative	39	ok	0
042	2026-06-25	zone_cumulative	39	ok	0
044	2026-06-27	zone_cumulative	40	ok	0
046	2026-06-29	zone_cumulative	40	ok	0
047	2026-06-30	zone_cumulative	41	ok	0
048	2026-07-01	zone_cumulative	41	ok	0
049	2026-07-02	zone_cumulative	41	ok	0
050	2026-07-03	zone_cumulative	41	ok	0
051	2026-07-04	zone_cumulative	41	ok	0
052	2026-07-05	zone_cumulative	41	ok	0
053	2026-07-06	zone_cumulative	42	fails	-5	exception
054	2026-07-07	zone_cumulative	42	ok	0
055	2026-07-08	zone_cumulative	42	ok	0	exception
056	2026-07-09	zone_cumulative	42	ok	0
057	2026-07-10	zone_cumulative	48	fails	5
058	2026-07-11	zone_cumulative	49	ok	0
059	2026-07-12	zone_cumulative	49	ok	0	exception
060	2026-07-13	zone_cumulative	52	ok	0	caveat
061	2026-07-14	zone_cumulative	52	ok	0	caveat
062	2026-07-15	zone_cumulative	53	ok	0	caveat
063	2026-07-16	zone_cumulative	53	ok	0	caveat, exception
064	2026-07-17	zone_cumulative	53	ok	0	caveat
065	2026-07-18	zone_cumulative	53	ok	0	caveat
066	2026-07-19	zone_cumulative	54	ok	0	caveat
067	2026-07-20	zone_cumulative	54	ok	0	caveat
068	2026-07-21	zone_cumulative	54	ok	0	caveat
069	2026-07-22	zone_cumulative	53	ok	0	caveat
070	2026-07-23	zone_cumulative	54	ok	0	caveat
071	2026-07-24	zone_cumulative	54	ok	0	caveat
072	2026-07-25	zone_cumulative	54	ok	0	caveat
073	2026-07-26	zone_cumulative	54	ok	0	caveat
074	2026-07-27	zone_cumulative	54	ok	0	caveat
077	2026-07-30	zone_cumulative	55	ok	0	caveat
078	2026-07-31	zone_cumulative	55	ok	0	caveat
079	2026-08-01	zone_cumulative	56	ok	0	caveat
080	2026-08-02	zone_cumulative	57	ok	0	caveat, route rejected
081	2026-08-03	zone_cumulative	57	ok	0	caveat, exception
082	2026-08-04	zone_cumulative	57	ok	0	caveat
083	2026-08-05	zone_cumulative	59	ok	0	caveat, exception
084	2026-08-06	—	0	—
085	2026-08-07	—	0	—
086	2026-08-08	—	0	—
087	2026-08-09	zone_daily	34	ok	0
088	2026-08-10	zone_cumulative	60	ok	0
089	2026-08-11	zone_cumulative	60	fails	1	exception
090	2026-08-12	zone_cumulative	62	ok	0
091	2026-08-13	zone_cumulative	63	ok	0
092	2026-08-14	zone_cumulative	63	ok	0
093	2026-08-15	zone_cumulative	63	ok	0
094	2026-08-16	zone_cumulative	63	ok	0
095	2026-08-17	zone_cumulative	63	ok	0
096	2026-08-18	zone_cumulative	64	ok	0
097	2026-08-19	zone_cumulative	64	ok	0
098	2026-08-20	zone_cumulative	64	ok	0
099	2026-08-21	zone_cumulative	65	ok	0
100	2026-08-22	zone_cumulative	65	ok	0
101	2026-08-23	zone_cumulative	65	ok	0
102	2026-08-24	zone_cumulative	66	ok	0
103	2026-08-25	zone_cumulative	66	ok	0
104	2026-08-26	zone_cumulative	68	ok	0
105	2026-08-27	zone_cumulative	68	ok	0
106	2026-08-28	zone_cumulative	68	ok	0
107	2026-08-29	zone_cumulative	68	ok	0
108	2026-08-30	zone_cumulative	68	ok	0
109	2026-08-31	zone_cumulative	68	ok	0
110	2026-09-01	zone_cumulative	68	ok	0
111	2026-09-02	zone_cumulative	68	ok	0
112	2026-09-03	zone_cumulative	68	ok	0
113	2026-09-04	zone_cumulative	69	ok	0
114	2026-09-05	zone_cumulative	69	ok	0
115	2026-09-06	zone_cumulative	69	ok	0
116	2026-09-07	zone_cumulative	69	ok	0
117	2026-09-08	zone_cumulative	69	ok	0
118	2026-09-09	zone_cumulative	69	ok	0
119	2026-09-10	zone_cumulative	71	ok	0
120	2026-09-11	zone_cumulative	71	ok	0
121	2026-09-12	zone_cumulative	71	ok	0
122	2026-09-13	zone_cumulative	71	ok	0
123	2026-09-14	zone_cumulative	71	ok	0
124	2026-09-15	zone_cumulative	71	ok	0
125	2026-09-16	zone_cumulative	71	ok	0
126	2026-09-17	zone_cumulative	71	ok	0
127	2026-09-18	zone_cumulative	72	ok	0
128	2026-09-19	zone_cumulative	72	ok	0
129	2026-09-20	zone_cumulative	72	ok	0
130	2026-09-21	zone_cumulative	72	ok	0
131	2026-09-22	zone_cumulative	72	ok	0
132	2026-09-23	zone_cumulative	72	ok	0
133	2026-09-24	zone_cumulative	72	ok	0
134	2026-09-25	zone_cumulative	72	ok	0
135	2026-09-26	zone_cumulative	72	ok	0
136	2026-09-27	zone_cumulative	72	ok	0
137	2026-09-28	zone_cumulative	72	ok	0
138	2026-09-29	zone_cumulative	72	ok	0
139	2026-09-30	zone_cumulative	72	ok	0
140	2026-10-01	zone_cumulative	72	ok	0
141	2026-10-02	zone_cumulative	73	ok	0
143	2026-10-04	zone_cumulative	73	ok	0
144	2026-10-05	zone_cumulative	73	ok	0
145	2026-10-06	zone_cumulative	73	ok	0
146	2026-10-07	zone_cumulative	73	ok	0
147	2026-10-08	zone_cumulative	74	ok	0
148	2026-10-09	zone_cumulative	74	ok	0
149	2026-10-10	zone_cumulative	74	ok	0
150	2026-10-11	zone_cumulative	74	ok	0
151	2026-10-12	zone_cumulative	74	ok	0
152	2026-10-13	zone_cumulative	74	ok	0
153	2026-10-14	zone_cumulative	74	ok	0
154	2026-10-15	zone_cumulative	74	ok	0
155	2026-10-16	zone_cumulative	76	ok	0
156	2026-10-17	zone_cumulative	76	ok	0
157	2026-10-18	zone_cumulative	76	ok	0
158	2026-10-19	zone_cumulative	76	ok	0
159	2026-10-20	zone_cumulative	76	ok	0
160	2026-10-21	zone_cumulative	76	ok	0
161	2026-10-22	zone_cumulative	76	ok	0
163	2026-10-23	zone_cumulative	76	ok	0

One primary source, and two readings I only check against

The primary source is the set of situation-report PDFs that INSP (Institut National de Santé Publique) posts at insp.cd. I download them and pin each one with a committed index and a manifest of SHA-256 hashes, so I can tell if a file ever changes. Then I read the numbers from the table geometry drawn in the PDF itself.

Two other groups have read the same documents. INRB-UMIE, the national laboratory, keeps its own transcription, and epiforecasts has a scan. I use both only as checks; nothing on this page is built from them. Since all three of us read the same pages, agreement between us tells you the transcription is right. It tells you nothing about whether the counts themselves are right. A disagreement, on the other hand, always means something: either one of the readings has an error, or two versions of a revised report were read.

Against INRB-UMIE, per zone-day: 5,592 of 5,609 case values agree exactly and 5,490 of 5,499 death values; 23 rows disagree on either, each recorded with both values. The denominator is the inner join of the two lineages — 1,384 of our 6,998 zone-day rows have no INRB counterpart and sit outside every agreement figure, and they are not random: the INRB series currently ends 2026-10-18, ours 2026-10-23, so the most recent reports are uncompared by construction. Against the epiforecasts scan, the national cumulative agrees on 133 of 134 reports exactly. The largest disagreement is not ours: sitrep 032 prints Rwampara at 159 twice over — in its table and its narrative — where the INRB file carries 192.

Our values plotted against INRB-UMIE’s, with the line where the two would be equal, and each disagreement over time with its sign.

'data/build/figures/latest/comparison_series.png'

The three readings as time series: our health-zone rows summed up, INSP’s own printed totals (rings), and INRB-UMIE’s transcription summed through the same mapping from zones to provinces. The bottom row shows where they differ. Ours minus INRB mixes two things, values we disagree on and rows missing from their file. Ours minus printed is plain arithmetic against the page, and it is zero except where the source disagrees with itself, each case of which is recorded in a register (see below). INRB’s national file agrees with ours exactly wherever both have a value, and it starts a few days before our zone-level series does.

Every table is checked against the page’s own arithmetic

Every table I extract has to add up the way the page says it does. A province’s health-zone rows, plus any rows the source counts but hasn’t assigned to a zone yet, must equal the province total printed on the page, and the provinces must add up to the printed national total. When they don’t, I assume the mistake is mine until reading the page shows otherwise.

Sometimes the page really doesn’t add up. When that happens the judgement goes into a register, a hand-kept file with one row per judgement. Each row is tied to the SHA-256 hash of that exact PDF, and it excuses that one gap on that one document and nothing else.

Province-reports: 719 reconcile exactly, 10 carry a hand-verified exception (the page disagrees with itself, read and recorded), 0 disagree and remain open, 30 print no total to check against.

This check has a blind spot. It only catches errors that break the arithmetic. If I read the neighbouring column of a table by mistake, say suspected cases instead of confirmed, that column adds up just as well, and the check passes. Mistakes like that get caught another way: by tests that pin the values of specific named reports, and by comparing the assembled series against the printed totals, further down.

The series

'data/build/figures/latest/province_series.png'

Cumulative confirmed cases and deaths by province. The line is the sum of the health-zone rows the reports print. Rings are the values in the printed provincial summary, shown as a check. Squares are values read from sentences in the report text, which carry their own internal check.

'data/build/figures/latest/streams_detail.png'

Every daily flow I extract, by province. The reports that carried the source’s warning that recent data were still being consolidated aren’t marked here, but every daily stream file has a `consolidating` column that says which they are.

'data/build/figures/latest/national_detail.png'

The four series the source prints only for the whole country, never by province. Two of them are stocks: each value is a reading taken at one moment, like the number of patients in isolation that day, rather than a count accumulated over a day. So they’re drawn as levels and never smoothed. The calendar matches the panel above.

The streams that downstream models fit are weekly incidence per province. To build them I take the cumulative count, carry the last known value forward over days with no report, difference it, and bin the differences into weeks counted from day 0 (2026-05-19). Two checks have to pass. The total a stream emits has to equal exactly how far the cumulative count moved, and the assembled grid is compared against every province row the reports print.

Table 2: Stream reconciliation. vs source Δ must be zero by construction; grid = printed counts report dates where the assembled cumulative equals the printed province row, and the final gap is the difference on the last report — zero for every stream.

stream	obs	emitted	vs source Δ	grid = printed	final gap
`cases_bas_uele`	11	12	0	72/72	+0
`cases_daily_bas_uele`	71	12	0	—	—
`cases_daily_haut_uele`	98	430	0	—	—
`cases_daily_ituri`	142	7,105	0	—	—
`cases_daily_national`	152	9,843	0	—	—
`cases_daily_nord_kivu`	143	2,209	0	—	—
`cases_daily_nord_ubangi`	7	1	0	—	—
`cases_daily_sud_kivu`	140	2	0	—	—
`cases_daily_sud_ubangi`	42	1	0	—	—
`cases_daily_tshopo`	98	73	0	—	—
`cases_haut_uele`	16	430	0	99/99	+0
`cases_ituri`	24	7,105	0	134/136	+0
`cases_national`	24	9,843	0	—	—
`cases_nord_kivu`	24	2,209	0	133/136	+0
`cases_nord_ubangi`	2	1	0	8/8	+0
`cases_sud_kivu`	23	2	0	136/136	+0
`cases_sud_ubangi`	7	1	0	43/43	+0
`cases_tshopo`	16	73	0	99/99	+0
`community_deaths_bas_uele`	11	3 of 4	—	12/12 daily	—
`community_deaths_daily_bas_uele`	72	3 of 4	—	0 flagged	—
`community_deaths_daily_haut_uele`	87	75 of 165	—	13 flagged	—
`community_deaths_daily_ituri`	96	1,434 of 2,380	—	21 flagged	—
`community_deaths_daily_national`	96	2,313 of 3,636	—	21 flagged	—
`community_deaths_daily_nord_kivu`	96	789 of 1,063	—	21 flagged	—
`community_deaths_daily_nord_ubangi`	8	1 of 1	—	0 flagged	—
`community_deaths_daily_sud_kivu`	80	8 of 22	—	6 flagged	—
`community_deaths_daily_sud_ubangi`	43	1 of 1	—	0 flagged	—
`community_deaths_daily_tshopo`	79	10 of 22	—	5 flagged	—
`community_deaths_haut_uele`	16	75 of 165	—	72/84 daily	—
`community_deaths_ituri`	16	1,434 of 2,380	—	36/95 daily	—
`community_deaths_national`	16	2,313 of 3,636	—	93/96 daily	—
`community_deaths_nord_kivu`	16	789 of 1,063	—	81/96 daily	—
`community_deaths_nord_ubangi`	2	1 of 1	—	1/1 daily	—
`community_deaths_sud_kivu`	14	8 of 22	—	6/6 daily	—
`community_deaths_sud_ubangi`	7	1 of 1	—	2/2 daily	—
`community_deaths_tshopo`	13	10 of 22	—	38/41 daily	—
`cumul_recovered`	147	0 → 2,646	—	stock, 2026-05-17 to 2026-10-23	—
`deaths_bas_uele`	11	3	0	72/72	+0
`deaths_daily_bas_uele`	71	3	0	—	—
`deaths_daily_haut_uele`	98	188	0	—	—
`deaths_daily_ituri`	133	3,303	0	—	—
`deaths_daily_national`	153	4,799	0	—	—
`deaths_daily_nord_kivu`	133	1,236	0	—	—
`deaths_daily_nord_ubangi`	7	0	0	—	—
`deaths_daily_sud_kivu`	134	0	0	—	—
`deaths_daily_sud_ubangi`	42	0	0	—	—
`deaths_daily_tshopo`	98	25	0	—	—
`deaths_haut_uele`	16	188	0	99/99	+0
`deaths_ituri`	22	3,303	0	132/135	+0
`deaths_national`	24	4,799	0	—	—
`deaths_nord_kivu`	22	1,236	0	129/135	+0
`deaths_nord_ubangi`	2	0	0	8/8	+0
`deaths_sud_kivu`	22	0	0	135/135	+0
`deaths_sud_ubangi`	7	0	0	43/43	+0
`deaths_tshopo`	16	25	0	98/99	+0
`lab_daily_bas_uele`	33	7 of 270 tests	—	—	—
`lab_daily_haut_uele`	69	281 of 1,545 tests	—	—	—
`lab_daily_ituri`	125	5,675 of 25,619 tests	—	—	—
`lab_daily_national`	130	7,991 of 44,014 tests	—	1–6 provinces/day	—
`lab_daily_nord_kivu`	111	1,915 of 15,648 tests	—	—	—
`lab_daily_nord_ubangi`	1	1 of 4 tests	—	—	—
`lab_daily_sud_kivu`	26	0 of 137 tests	—	—	—
`lab_daily_tshopo`	42	112 of 791 tests	—	—	—
`lab_national`	21	7,991 of 44,014 tests	—	317/333 vs new cases	—
`patients_isolated`	139	173 → 1,018	—	stock, 2026-06-01 to 2026-10-23	—
`patients_isolated_confirmed`	49	47 → 335	—	stock, 2026-06-02 to 2026-08-02	—

The registers hold the judgements I made by reading a page

Everything else in this repository is either downloaded or computed. The registers are different. Each row is a judgement I made after reading a page. Every row is keyed on the SHA-256 hash of the document it’s about, so if INSP revises a report, the hash stops matching and the row has to be judged again. And each row excuses exactly one value, so if the value we read changes, the excuse no longer covers it.

register	rows	each row asserts
`reconciliation_exceptions.tsv`	15	this page’s printed rows do not sum to its printed total, by exactly this much
`rejected_routes.tsv`	2	this route, on this document, reads the wrong table – or reads the right one with named columns displaced
`zone_province_overrides.tsv`	13	the page states in words which province this row belongs to
`zone_aliases_local.csv`	8	this printed spelling is that canonical health zone
`sitrep_number_overrides.tsv`	29	this upload’s filename misstates its report number or data date
`external_sitreps.tsv`	3	INSP does not serve this report; it is held from a secondary archive and its PDF is committed, because nothing can fetch it again

The care ledger

The reports also keep a ledger of the treatment centres, written out in words for each province in the section on patient care. It gives admissions; exits, and the four ways a patient can leave a centre; the census at the end of each day, split into confirmed and suspected patients; and the number of beds installed. extract_care_prose.py reads all of it into data/build/series/care_prose.csv, and the release publishes it as the care_* stream files, which the streams README lists. The confirmed census is worth a look if you’re fitting: it measures a state (how many confirmed patients are in care at one moment), while the case and death streams measure flows.

'data/build/figures/latest/care_ledger.png'

The care ledger, as the reports write it out in words. Panel A is why I trust it. The front-page tile and the care section’s sentence about the whole response are two readings of the same page, made by two different extractors in this repository, and wherever both print a census of confirmed patients in isolation, they agree exactly. Panel B is occupancy against capacity for each province. It’s a ratio of sums, so it can’t see one full centre inside a half-full province. Panel C is the ledger itself, split into panels by the passage that printed each value, and never summed across panels. Panel D is every count of absconders (patients who left a centre without being discharged) that the reports print, drawn as separate marks because that’s what the data are.

What isn’t here

I don’t extract the laboratory, alerts or contact-tracing tables yet. The reference pipeline reads them, and this one will. The care ledger comes from the report text only. The reports also print a patient-movement table with the same quantities per province, absconders included, and reading it would roughly triple the absconder coverage you see in panel D. I also don’t read the daily suspected inflow from the front-page dashboard after 12 July, so the national suspected series ends where the early layout stops printing it. The 23 disagreements with INRB are waiting to be settled: I’ll read each page, then decide case by case whether to report it upstream. Finally, no fit uses these streams yet. That work belongs to the modelling repository.