Predictive Patterns
Statistical analysis of change frequency and trends to predict when the next update will happen. Available on the Pro plan.
Pro feature: Predictive Patterns are available on the Pro plan. Free users can preview the full layout with sample data to see what's available.
What Are Predictive Patterns?
Every website has a rhythm. Some update daily like clockwork. Others change unpredictably in bursts. Predictive Patterns analyzes your monitor's change history and tells you when the next change is most likely — so you're never caught off guard.
The system gets smarter with more data. With just a few changes, you'll see basic timing info. As more changes accumulate, it unlocks advanced analysis: trend detection, seasonality, frequency shift detection, and predictions with confidence bands.
How to Access
- Go to any monitor detail page and scroll to the "Predictive Patterns" card.
- Or click the "Patterns" button on any monitor card in the dashboard.
- You need at least 2 detected changes.
- Pro plan required. Free users can see a preview.
What You'll See
Frequency Pattern + Velocity
How often this site changes and whether that pace is speeding up, slowing down, or holding steady. An arrow icon tells you the trend at a glance:
- ↑ — Changes are becoming more frequent
- ↓ — Changes are slowing down
- — — Stable pace
The text below explains the pattern in plain language. The "?" tooltip explains the statistical test behind it.
Peak Activity
The days and times when changes happen most often, shown as badges: "Tuesday (7 changes)", "Around 0:00 UTC (7 changes)." Plan your reviews around these peaks so you catch changes when they're fresh.
Quiet Periods
When changes rarely happen. If you see "Thursday, Friday, Saturday (0 changes)", those days are typically quiet. Useful for scheduling maintenance or knowing when you can step away from monitoring.
Prediction
When the next change is most likely. Includes a confidence badge:
- STRONG — High confidence, based on extensive data
- MODERATE — Good confidence, reasonable data
- PRELIMINARY — Early estimate, limited data
Example: "Next change most likely in ~24 hours. Based on statistical analysis with 50% confidence between 18h and 32h." The wider the range, the less predictable the site is.
Anomalies
Unusual timing patterns. If a site suddenly shifts from weekly updates to daily, this section flags it. It also detects unexpected quiet periods — when a site that normally changes frequently goes silent.
Change Activity (Calendar Heatmap)
A grid showing 12 weeks of change history, styled like a GitHub contribution chart. Each square is one day — darker green means more changes that day. You'll instantly spot seasonal patterns, quiet weeks, and burst periods.
Focus Area Trends (Sparklines)
For monitors tracking specific values — prices, stock levels, percentages — this shows how those values change over time. Each tracked value gets:
- ATH — Current value is at its all-time high
- ATL — Current value is at its all-time low
- Near high/low — Within 5% of the record
- ↑/↓/— — Direction based on the recent trend
- Sparkline — A mini chart of the value's history
- Low/High — The full range of observed values
How Predictions Get Better with More Data
The system progressively unlocks more analysis as it collects data:
| Changes | What You Get |
|---|---|
| 2+ | Basic timing: average and median gap between changes |
| 3+ | Regularity score: is this site deterministic, regular, variable, or bursty? |
| 5+ | Trend detection: are changes speeding up or slowing down? |
| 9+ | Frequency shift detection: did the update schedule change? |
| 10+ | Day/hour patterns: which days and hours see the most changes |
| 15+ | Full predictions with confidence bands + overdue detection |
| 17+ | Prediction accuracy tracking: how reliable are the predictions? |
| 20+ | Day-of-week significance testing |
| 30+ | Periodicity detection: does this site update on a regular cycle? |
Overdue Detection
When a monitor has enough history (15+ changes), CheckSite knows when a change is "expected." If the site hasn't updated when it normally does, you'll see colored badges on your dashboard:
- Yellow "Late" — Slightly behind schedule (1.5× the normal interval)
- Orange "Overdue" — Missed the expected window (2× the normal interval)
- Red "Very Late" — Significantly past due (3× the normal interval)
This is especially useful for monitoring competitors' changelog pages. If they usually update weekly and haven't in two weeks, something may have changed in their release process.
Prediction Accuracy
CheckSite tracks how accurate its predictions are by testing each one against what actually happened:
- Coverage — What percentage of actual changes fell within the predicted window. The target is around 50% for the core prediction range.
- Accuracy tier: Excellent (40–60% coverage), Good (30–70%), Fair (20–80%).
More regular websites produce more accurate predictions. Highly irregular sites will honestly show wider prediction bands.
Tips
- More data = better predictions. Let monitors run for at least 2 weeks before expecting reliable forecasts.
- Focus area sparklines are most useful for price-tracking monitors — you'll see trends and records at a glance.
- The calendar heatmap reveals seasonal patterns you might miss by looking at individual changes.
- Set up smart triggers based on peak activity windows to get alerts at the right time.
- If a monitor shows "Very Late", investigate. The site may have changed URLs or structure.
Requirements
- Pro plan
- At least 2 detected changes for basic patterns
- 15+ changes for full predictions and overdue detection
- All statistics are computed from your data. The AI only interprets pre-computed numbers — it never sees your raw page content.