Forecast requests submitted via GitHub Actions. Click any run for full report.

Submit Forecast Run

Payload instructions

⚠️ Time series are assumed to be continuous at your time granularity! If not, backfill them.

  • Payload is JSON only.
  • For custom runs, provide values in series.
  • For M5 test mode, use series_names: ["demo_mode_m5"] and series: [].
FieldRequiredDescriptionExample
run_name_rootYesBase run name; backend prepends PST timestampmy-run
start_datetimeCustom modeFirst timestamp for the series (ISO)2026-01-01T17:15:00
granularityCustom modeTime spacing between points1h, 1d, 15m
horizonNoForecast periods to predict after history24
modelNoForecast backendnixtla, autogluon
n_seriesNoSeries count hint (used for M5 demo sampling)3
backtest_windowsNoRolling holdout window count for SMAPE backtests3
series_namesCustom modeList of names for each series (demo_mode_m5 reserved for demo)["trend","seasonal","combined"]
seriesCustom modeData values. Single list or list-of-lists (continuous, no gaps)[[...],[...],[...]]

Leaderboard (avg SMAPE)

RankModelAvg SMAPESamples
1AutoARIMA9.339636
2AutoETS9.448636

post-fix-check

nixtla · 1d · horizon 24 · history 3938 · forecast 96

run-fix-test

nixtla · 1h · horizon 24 · history 12 · forecast 48