candidate signal: severe/critical clinical forms in isolation (`formes sévères ou critiques`)

Start with data/candidate_signals.csv and data/README.md, using the existing severe_critical_forms entries for SitReps 080 and 081 as the reference. Check the not-yet-fitted convention in data/README.md and confirm the signal is documented there with its current Ituri-only coverage and limited history.

What it measures

Ituri's "Prise en charge holistique (PECH)" prose (§6, Activités de riposte par pilier) reports a count of patients currently in isolation classified as formes sévères ou critiques (severe/critical clinical forms), alongside the isolation total, admissions and sorties for the day.

Where it appears

Prose only, inside the PECH pillar box of §6. Ituri only so far — no equivalent figure for Nord-Kivu, Haut-Uélé, Tshopo or Sud-Kivu.

Example (SitRep 081, 3 August): "471 patients en isolement (302 confirmés, 169 suspects) sur 833 lits, soit 56,5 % d'occupation, et 57 formes sévères ou critiques, pour 77 admissions et 75 sorties."

First appearance

Checked SitReps 077, 078 and 079 by grep (formes sévère|severes ou critiques, case-insensitive) — zero hits in all three. SitRep 080 (2 August) is the first vintage carrying it (Ituri 63). SitRep 081 (3 August) carries it again (Ituri 57).

History so far

Backfilled to data/candidate_signals.csv as severe_critical_forms:

• 080 (2026-08-02): 63
• 081 (2026-08-03): 57

Only 2 vintages so far — too short a run to say anything about trend yet.

Why it might inform the model

It's a clinical-severity signal distinct from what's already fitted: confirmed cases, isolation occupancy and treatment-flow counts all measure volume, not severity mix. A rising severe/critical share among isolated patients could be an earlier or complementary signal of case-fatality pressure than the death counts themselves, and might help explain time-varying mortality risk within the treatment-flow submodel.

What it would plausibly attach to

Most naturally a covariate or sub-stream of isolation_history / the treatment-flow submodel (treatment_admission_model) — e.g. the severe/critical fraction of the Ituri isolation stock. Given it's currently Ituri-only and only 2 vintages deep, it isn't fittable yet; flagging for visibility per data/README.md's not-yet-fitted convention, added as a row there in the same PR that files this issue.