Your own walking pace (time model)

# Without a key, leave out the Authorization line (500 credits a day per address).
curl --fail-with-body 'https://api.trailsplits.com/v1/time-model/fit' \
  -H "Authorization: Bearer $TRAILSPLITS_API_KEY" \
  -H 'Content-Type: application/json' \
  --data '{"activities":[{"gpx":"<gpx>… your own timed walk …</gpx>"}]}'

{
  "type": "object",
  "properties": {
    "schema": {
      "type": "string",
      "enum": [
        "time-model/1"
      ]
    },
    "engine": {
      "type": "object"
    },
    "base_model": {
      "type": "string",
      "enum": [
        "trailsplits-hike/2"
      ]
    },
    "base_version": {
      "type": "string"
    },
    "pace_profile": {
      "type": "object",
      "properties": {
        "base_model": {
          "type": "string"
        },
        "base_version": {
          "type": "string"
        },
        "factors": {
          "type": "object",
          "properties": {
            "level": {
              "type": "number"
            },
            "up": {
              "type": "number"
            },
            "down": {
              "type": "number"
            }
          },
          "required": [
            "level",
            "up",
            "down"
          ]
        },
        "log_sd": {
          "type": "object",
          "properties": {
            "level": {
              "type": "number"
            },
            "up": {
              "type": "number"
            },
            "down": {
              "type": "number"
            }
          },
          "required": [
            "level",
            "up",
            "down"
          ]
        },
        "day_log_sd": {
          "type": "number"
        },
        "activities": {
          "type": "integer"
        }
      },
      "required": [
        "base_version",
        "factors",
        "log_sd",
        "day_log_sd"
      ]
    },
    "by_class": {
      "type": "object"
    },
    "activities": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "index": {
            "type": "integer"
          },
          "used": {
            "type": "boolean"
          },
          "reason": {
            "type": "string",
            "nullable": true
          },
          "length_m": {
            "type": "number"
          },
          "moving_minutes": {
            "type": "number"
          },
          "stopped_minutes": {
            "type": "number"
          },
          "model_minutes": {
            "type": "number"
          }
        },
        "required": [
          "index",
          "used"
        ]
      }
    },
    "method": {
      "type": "object"
    },
    "note": {
      "type": "string"
    },
    "unknowns": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "code": {
            "type": "string"
          },
          "message": {
            "type": "string"
          }
        },
        "required": [
          "code",
          "message"
        ]
      }
    }
  },
  "required": [
    "schema",
    "base_version",
    "pace_profile",
    "by_class",
    "activities",
    "unknowns"
  ]
}

Send your recent timed walks. You get your own pace against the TrailSplits hike time model: how much faster or slower you are than the model on level ground, uphill and downhill, each with its uncertainty. Pass it back as pace_profile to route analysis, the route brief, route weather or go windows, and they give your own time beside the model's.

What it is, and what it is not.

• Nothing is kept. The activities are read, fitted and dropped. They are not stored, logged or kept in a receipt. The answer carries no coordinates and no times, only totals per activity and the fitted factors.
• Relative to one model version. A profile names base_version: the model and its constants (trailsplits-hike/2@0.96/0.18 today). When TrailSplits refits the population model, every older profile is refused with 422 pace_profile_stale. Fit again; old factors are never reused silently.
• Moving time. Stops are left out: an interval where you neither cover 0.25 m/s nor climb or descend 0.05 m/s, and any interval over 5 minutes. Breaks are not in the profile, and not in the times it gives.
• Held back with little data. Each factor starts at the population model (1). One activity moves it only part of the way, and its uncertainty says how far. Five or more activities with real climbing and descending are better.
• The model's own sections. Your time is compared with the model section by section: about 100 m each, terrain heights smoothed ±25 m. Each section is level (under 5 % grade), up or down. Heights are the terrain model's at your points, the model's own basis, never your watch's.

Request

{ "activities": [
    { "gpx": "<gpx …><trk><trkseg><trkpt lat=\"45.82\" lon=\"6.97\"><time>2026-09-01T08:00:00Z</time></trkpt>…" },
    { "points": [[6.97, 45.82, "2026-09-03T07:30:00Z"], [6.971, 45.821, 1788420660], …] }
  ] }

Response

200, time-model/1:

Using the profile:

• Route analysis: pace_profile with profile: hike adds time.personal { minutes, interval_80, base_version, factors, activities }. time.minutes stays the population model's.
• Route brief: the same in its analysis, and in summary as personal_minutes and personal_interval_80. The weather part is timed with it.
• Route weather and go windows: pace_profile times the walk in place of pace_factor (not both). pace.personal names the factors.

Errors

Every refusal also carries docs (this page's code on /api/errors), and field, limit and got, hint, retry_after or request_id where they apply, as the error standard describes (api-errors-v1, since 6 Oct 2026).

JSON, { "error": { "code", "message" } }.

How it works

1. Terrain heights are read at the activity's points: every point, or every k-th past 60,000 in all.
2. Moving seconds per interval are spread over the model's sections by distance. Each activity then has actual and model minutes for level, up and down.
3. Per class: the posterior mean of log(actual / model). Each activity weighs in by its minutes in the class (full weight from 30 minutes, none under 5), with noise sd 0.12, against a normal prior of sd 0.3 around 0 (the population model). The factor is exp(mean), and log_sd is the posterior sd.
4. A route's personal time is each class's factor times the model's minutes in that class. Its 80 % interval combines each factor's uncertainty, weighted by its share of the time, with the day-to-day spread.

Rough ground: the fit uses the calibration's median share of rough ground for every activity, because activities are not matched to the map. A walker who always hikes on rough ground therefore has slightly high factors, which route analysis then applies on top of its own rough-ground share. This is a known limit of v1.

Versioning

New fields may appear in time-model/1. A change of what a factor means is time-model/2. A refit of the base model changes base_version, not the schema.