Status
The pipeline, watching itself.
Every load publishes a picture of its own ingestion. This page reads the most recent picture and compares it with what is expected — which is not guesswork: it is the measurement published on the research page.
When this page was built, ingestion was within the expected envelope.
- Data throughSeptember 26, 2026 at 11:30 PM
The ONS portal delivers the previous day in full, half-hour by half-hour.
- Lag at the edge27.1 h
Expected: between ~13 and ~43 hours, depending on the time of day.
- Loads in the last 7 days13
Expected: 14, two a day. The series started on 27 Aug 2026 — the number climbs until then.
- Last snapshotSeptember 28, 2026 at 2:37 AM
A scheduled job runs and dies, twice a day: no server stays up.
- Latest forecastSeptember 28, 2026 at 7:00 AM
Within expected — the model runs every ~12h. See forecast.
- Model in productionlightgbm-v1-residuo+correcao
Change the variant, change the name — and the methodology gets a note.
Recent health, day by day
Two series that already have enough history to draw: whether new ONS data arrived each day, and whether that day's forecast beat the baseline. Each cell is a day, in pages of 14 — the first is always the most recent, earlier ones sit behind the numbers. The legend below explains the three colors.
- Sep 14: new ONS data
- Sep 15: new ONS data
- Sep 16: new ONS data
- Sep 17: new ONS data
- Sep 18: loads ran, no new ONS data
- Sep 19: loads ran, no new ONS data
- Sep 20: loads ran, no new ONS data
- Sep 21: loads ran, no new ONS data
- Sep 22: new ONS data
- Sep 23: new ONS data
- Sep 24: new ONS data
- Sep 25: new ONS data
- Sep 26: loads ran, no new ONS data
- Sep 27: new ONS data
- Aug 31: new ONS data
- Sep 1: new ONS data
- Sep 2: new ONS data
- Sep 3: new ONS data
- Sep 4: new ONS data
- Sep 5: new ONS data
- Sep 6: new ONS data
- Sep 7: new ONS data
- Sep 8: loads ran, no new ONS data
- Sep 9: new ONS data
- Sep 10: new ONS data
- Sep 11: new ONS data
- Sep 12: new ONS data
- Sep 13: new ONS data
- Aug 27: new ONS data
- Aug 28: new ONS data
- Aug 29: new ONS data
- Aug 30: new ONS data
- Sep 13: above baseline
- Sep 14: below baseline
- Sep 15: below baseline
- Sep 16: below baseline
- Sep 17: above baseline
- Sep 18: no comparison
- Sep 19: no comparison
- Sep 20: no comparison
- Sep 21: below baseline
- Sep 22: above baseline
- Sep 23: above baseline
- Sep 24: below baseline
- Sep 25: below baseline
- Sep 26: above baseline
- Aug 30: below baseline
- Aug 31: below baseline
- Sep 1: below baseline
- Sep 2: below baseline
- Sep 3: above baseline
- Sep 4: below baseline
- Sep 5: below baseline
- Sep 6: above baseline
- Sep 7: above baseline
- Sep 8: above baseline
- Sep 9: below baseline
- Sep 10: below baseline
- Sep 11: below baseline
- Sep 12: above baseline
- Aug 28: below baseline
- Aug 29: above baseline
- Ok
- Attention
- Missing
The baseline is the error that repeating the previous day would give (mae_30d_regua_mw); the full comparison, with numbers on both sides, is on the accuracy page.
Field by field
The picture is public at the same address this page reads: nameplate.com.br/saude.json. The field contract was locked on 27 Aug 2026, on both sides of the pipeline:
snapshot_at- When the pipeline took this picture.
dado_ate- The newest instant present in the ONS data (the last
din_instante). atraso_horas- The distance between the picture and the newest data point — the lag at the edge.
cargas_7d- How many loads arrived in the last seven days, counting this one.
These numbers are baked when the site is generated; the always-current values live in saude.json itself. Each load also leaves a history line: the public version, in cargas.csv (snapshot_at, dado_ate, atraso_horas, one line per load), is what draws the day-by-day ingestion above; the full one, with the MWh per month, measured the size of ONS revisions in the 22 Sep note. The data page explains the rest of the kitchen.