Metrics Window Summary

Overview

You send samples and a window and we return the count, the extremes, the mean, the sum and six percentiles, every figure rounded once at your declared scale. The window is half-open, so a sample exactly on the start counts and one exactly on the end does not, which is what makes a row of adjacent windows add up to their union. Samples outside the window are excluded and counted rather than refused, so an overshooting slice still gets an answer that says so. The interpolation method is echoed, so a stored summary is reproducible in a spreadsheet without reading anything else.

Percentiles are where two teams quietly disagree: one interpolates, one picks the nearest rank, one rounds before comparing, and the numbers stop matching. This publishes the method and echoes it with every answer.

When to use it

Scope

Non-goals

• Does not group rows, which data.aggregate answers
• Does not resample, convert to a rate or fill a gap
• Does not interpolate between samples — a missing sample stays missing
• Does not fetch or store a metric, and reads no clock
• Does not accept a caller-chosen percentile; the six are fixed

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"observability.summarize_metrics_window","input":{"samples":[{"t":"2026-07-29T10:00:00Z","v":10}],"window":{"start":"2026-07-29T10:00:00Z","end":"2026-07-29T10:05:00Z"}}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("observability.summarize_metrics_window", {"series":"http.server.duration","window":{"start":"2026-07-29T10:00:00Z","end":"2026-07-29T10:05:00Z"},"samples":[{"v":10,"t":"2026-07-29T10:00:00Z"},{"v":20,"t":"2026-07-29T10:01:00Z"},{"v":30,"t":"2026-07-29T10:02:00Z"},{"v":40,"t":"2026-07-29T10:05:00Z"}]});

Errors

Other limits

Limitations

• Does not group rows, which data.aggregate answers
• Does not resample, convert to a rate or fill a gap
• Does not interpolate between samples — a missing sample stays missing
• Does not fetch or store a metric, and reads no clock
• Does not accept a caller-chosen percentile; the six are fixed
• Grouping rows by a key and reducing each group, which data.aggregate answers
• Deciding whether an objective was met, which observability.calculate_sli answers

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "observability.summarize_metrics_window",
  "input": {
    "series": "http.server.duration",
    "window": {
      "start": "2026-07-29T10:00:00Z",
      "end": "2026-07-29T10:05:00Z"
    },
    "samples": [
      {
        "v": 10,
        "t": "2026-07-29T10:00:00Z"
      },
      {
        "v": 20,
        "t": "2026-07-29T10:01:00Z"
      },
      {
        "v": 30,
        "t": "2026-07-29T10:02:00Z"
      },
      {
        "v": 40,
        "t": "2026-07-29T10:05:00Z"
      }
    ]
  }
}

Examples

{
  "series": "http.server.duration",
  "window": {
    "start": "2026-07-29T10:00:00Z",
    "end": "2026-07-29T10:05:00Z"
  },
  "samples": [
    {
      "v": 10,
      "t": "2026-07-29T10:00:00Z"
    },
    {
      "v": 20,
      "t": "2026-07-29T10:01:00Z"
    },
    {
      "v": 30,
      "t": "2026-07-29T10:02:00Z"
    },
    {
      "v": 40,
      "t": "2026-07-29T10:05:00Z"
    }
  ]
}

{
  "lowResolutionPercentiles": [
    "p75",
    "p90",
    "p95",
    "p99",
    "p99_9"
  ],
  "count": 3,
  "max": 30,
  "firstSampleAt": "2026-07-29T10:00:00Z",
  "sum": 60,
  "outOfWindowCount": 1,
  "percentiles": {
    "p50": 20,
    "p75": 25,
    "p90": 28,
    "p95": 29,
    "p99": 29.8,
    "p99_9": 29.98
  },
  "sampleCoverageSeconds": 120,
  "scale": 6,
  "series": "http.server.duration",
  "interpolation": "LINEAR_ORDER_STATISTIC_TYPE_7",
  "empty": false,
  "roundingMode": "HALF_UP",
  "lastSampleAt": "2026-07-29T10:02:00Z",
  "min": 10,
  "nullCount": 0,
  "mean": 20,
  "window": {
    "start": "2026-07-29T10:00:00Z",
    "end": "2026-07-29T10:05:00Z",
    "inclusivity": "[start,end)",
    "durationSeconds": 300
  }
}

{
  "series": "http.server.duration",
  "window": {
    "start": "2026-07-29T10:00:00Z",
    "end": "2026-07-29T10:05:00Z"
  },
  "samples": [
    {
      "v": 42,
      "t": "2026-07-29T10:00:00Z"
    }
  ]
}

{
  "lowResolutionPercentiles": [
    "p50",
    "p75",
    "p90",
    "p95",
    "p99",
    "p99_9"
  ],
  "count": 1,
  "max": 42,
  "firstSampleAt": "2026-07-29T10:00:00Z",
  "sum": 42,
  "outOfWindowCount": 0,
  "percentiles": {
    "p50": 42,
    "p75": 42,
    "p90": 42,
    "p95": 42,
    "p99": 42,
    "p99_9": 42
  },
  "sampleCoverageSeconds": 0,
  "scale": 6,
  "series": "http.server.duration",
  "interpolation": "LINEAR_ORDER_STATISTIC_TYPE_7",
  "empty": false,
  "roundingMode": "HALF_UP",
  "lastSampleAt": "2026-07-29T10:00:00Z",
  "min": 42,
  "nullCount": 0,
  "mean": 42,
  "window": {
    "start": "2026-07-29T10:00:00Z",
    "end": "2026-07-29T10:05:00Z",
    "inclusivity": "[start,end)",
    "durationSeconds": 300
  }
}

Recommended use cases

• Get p95 and p99 for a five-minute slice of latency samples you already hold
• Compare two adjacent windows knowing the boundary rule cannot double-count

Avoid for

• Grouping rows by a key and reducing each group, which data.aggregate answers
• Deciding whether an objective was met, which observability.calculate_sli answers

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags