Always receiving 2016 sample data instead of real forecasts using timegpt-1

[mikomiko1258]

Description

<title>GitHub Issue: Always receiving 2016 sample data instead of real forecasts using timegpt-1</title>

Always receiving 2016 sample data instead of real forecasts using timegpt-1

Hello Nixtla team,

I’m attempting to use the TimeGPT endpoint (timegpt-1 model) but keep receiving the 2016 “Peyton Manning” sample dataset instead of an actual forecast. The HTTP status is 200, yet the response body contains timestamps such as "2016-01-14 00:00:00" and no "mean" key, suggesting it’s the standard demo output.

What I’ve tried:

1. Different environments:
   • MQL5/EA integration and a simple Python script (using requests). Both produce the same sample dataset response.
2. Multiple data formats:
   • "df" as an array of objects: [{ "timestamp": "...", "value": ... }, ...]
   • “split” style JSON ({"columns":[],"index":[],"data":[]})
   • Varying "freq" ("D", "15T") and "h" (5, 10, 24, etc.).
3. Valid date/time and numeric fields:
   • Confirmed no missing data or invalid timestamps. I tested simple dummy data (e.g., daily for 10 days, or 15-min intervals).
4. Authorization:
   • Using "Authorization": "Bearer <my_API_key>", no error codes (e.g., 401/422) appear. The response is 200 OK, but the content is always the Peyton Manning 2016 sample.
5. Trial vs. plan constraints:
   • I’m on a trial (or free) plan. I’m unsure whether timegpt-1 is restricted in this mode.
   • If so, I’d expect some mention in the docs or an explicit error instead of silent sample data.

Questions / help needed:

1. Is there a known restriction for timegpt-1 usage on trial accounts, causing Nixtla to serve the 2016 sample?
2. Are there any hidden constraints (model or plan) that might override my real data and return a demo output?
3. If I upgrade to a paid plan (e.g., Standard or Premium), will I reliably get real forecasts ("mean" in the JSON) instead of the sample?

Below is an example of the JSON I’m sending:

{
  "freq": "D",
  "h": 10,
  "time_col": "timestamp",
  "target_col": "value",
  "df": [
    { "timestamp": "2023.01.01 00:00", "value": 100.0 },
    { "timestamp": "2023.01.02 00:00", "value": 100.5 },
    ...
  ]
}

And the response:

{
  "timestamp": ["2016-01-14 00:00:00", ...],
  "value": [...],
  ...
}

Any advice or clarity on whether timegpt-1 is fully available for trial users would be greatly appreciated. If I need to switch to another model (like azureai) or upgrade, please let me know.

Thank you for your help,
Miko

Link

No response

[jmoralez]

Is there a reason for not using this library and using requests instead?