load-curve.seed-from-profile — explanation

Why

C4/C3/C2 customers without Linky access or a CDC upload still need a plausible

half-hour shape to drive tarif négocié optimization. Rather than hand-rolling

shapes per NAF, we ground the seed in the Enedis-published **fictive load

curves library* (courbes-fictives- datasets on opendata.enedis.fr).

```

{

segment: "RES1" | "RES2" | "PRO1" | "ENT1" | "NAF",

naf: string | null,

kva_band: string | null,

annual_kwh: number,

}

```

Candidate selection — pickCurve ranks every curve in the supplied

library by a deterministic score:

- exact NAF match → +100

- exact segment match → +50

- exact kVA band match → +25

- otherwise the best partial match wins; ties broken by curve_key alphabetical order.

Rescaling — the selected curve's annual sum is normalized to 1, then

each half-hour is multiplied by annual_kwh / 8760 × 2 to obtain a kWh

per half-hour series. Hourly output is the pair-sum across the 17 520

half-hours.

The rescaling step is linear and energy-conserving by construction:

sum(output) == annual_kwh within floating-point error.

Shape is a prior, not a measurement. Thermosensitivity at the customer's

actual site may diverge — the normalization method (thermosensitive)

layers on top.

No HP/HC discrimination is performed here; the consumer is responsible for

pairing this seed with a 30-minute HP/HC mask if needed.

When the library contains no candidate for the declared NAF, the method

falls back to the closest segment-level curve and emits a warning.