Separates durable trends from noise using a baseline, seasonal decomposition, and a signal test requiring persistence, magnitude beyond 2 standard deviations, 3+ independent corroborating data points, and a plausible mechanism - then characterizes the trend's stage and forecasts trajectory with a confidence range. Use when someone asks "is this a real trend or a blip", "where is this metric or market heading", "should we bet on this shift", or must defend a forecast to stakeholders. Do NOT use for scouting what content formats are trending on Instagram - use instagram-trend-scout instead; for tracking brand mentions and sentiment shifts in news and social - use media-monitor instead.
Click to play with sound.
---
name: Trend Analysis
description: Separates durable trends from noise using a baseline, seasonal decomposition, and a signal test requiring persistence, magnitude beyond 2 standard deviations, 3+ independent corroborating data points, and a plausible mechanism - then characterizes the trend's stage and forecasts trajectory with a confidence range. Use when someone asks "is this a real trend or a blip", "where is this metric or market heading", "should we bet on this shift", or must defend a forecast to stakeholders. Do NOT use for scouting what content formats are trending on Instagram - use instagram-trend-scout instead; for tracking brand mentions and sentiment shifts in news and social - use media-monitor instead.
---
# Trend Analysis
Distinguish durable trends from temporary fluctuations and project where they are headed. The hardest part is not spotting movement - it is deciding whether movement matters. The costly failure this skill prevents is the strategy built on a spike: a seasonal bump or a hype-cycle blip read as a secular shift, funded for three quarters, then quietly unwound.
## Inputs to collect
1. **The series or phenomenon** under study, and the metric that operationalizes it.
2. **History**: at least 2 full seasonal cycles of data where cycles exist (2 years for annual seasonality); with less, say the seasonality adjustment is unreliable and lower confidence.
3. **Candidate corroborating sources**: related metrics, independent datasets, external indicators.
4. **The decision at stake** and its horizon - a 6-month product bet and a 5-year infrastructure bet tolerate very different forecast uncertainty.
5. Defaults where the user has none: baseline window of 8+ periods, forecast horizon no longer than one-third of the observed history. Label assumed values as guesses.
## Operating procedure
### Step 1: Establish the baseline
Before calling anything a trend, define what "normal" looks like: compute a baseline level and expected variance over a relevant window. A data point is only notable relative to this. No baseline, no trend claim.
### Step 2: Decompose the series
… load the full skill through Skill Me