Predictions

The Predictions tab derives direction, risk, and possible outcomes from historical public data. A forecast is not a measurement: read every result with its window, sample, and confidence signal.

Outcome, prerequisites, and path

• Path: Channels → a channel → Predictions.
• The channel must be tracked with synced historical stats.
• Cards can have fixed windows or All Data context; do not assume the top period changes every card.
• Basic or detailed views may be locked by plan. The on-screen plan message is current.

Screen map

1. Momentum score, recent-period comparison, and detailed line chart.
2. Subscriber-loss analysis.
3. First-48-hour forecast.
4. View-velocity decay and shelf-life classes.
5. Pearson/Spearman correlation matrix.

Momentum score

Momentum is a composite product score comparing subscriber and positive view change in the last 7 days with the previous 7. The card has a fixed 14-day context and displays its current scale; on July 17, 2026, it was -100 to 200.

When previous-period change is near zero, relative change can be large. The model applies a channel-scale floor, but the result remains sensitive. Do not use the score as an absolute ranking across differently sized channels.

Detailed momentum, volatility, and consistency

• X-axis: date.
• Y-axis: 3-day moving averages derived from raw daily subscriber and view changes.
• Series/legend: Subscribers and Views; the tooltip shows date and both values.
• Volatility: standard deviation of daily changes. A higher value means a more variable series. Even if the UI adds a percent sign, do not read it as “percent of subscribers.”
• Consistency: a 0–100 product score relating volatility to average subscriber change. When the mean is near zero, small movements can strongly affect it.

The moving average reacts later than a daily reversal and does not fill missing days.

Subscriber-loss analysis

This card cannot see actual YouTube unsubscribe events. It is a 30-day proxy derived from changes in the public subscriber total.

YouTube rounding, cleanup, and data corrections can create negative days. The card cannot identify who unsubscribed or prove that a piece of content caused loss.

First-48-hour forecast

The card shows early views for recent videos alongside the channel's available 48-hour sample.

Publish time and view snapshots are required. — means insufficient 48-hour data, not zero. A few early hours can produce an extreme projection; do not treat it as a confidence band, and compare it with the actual value after 48 hours.

View velocity decay

This All Data card derives daily gains from cumulative daily views and keeps the final snapshot when several exist on the same calendar day.

View half-life is the first observed day when daily velocity falls to half of day-one velocity. Pie slices are classes and values are video counts; the tooltip shows class/count and the legend maps colors. Videos with fewer than two daily snapshots are insufficient data and excluded from the dominant class.

Half-life is sensitive to video age, sync gaps, and whether day one was partial. Do not infer SEO traffic automatically from an Evergreen class.

Correlation analysis

The All Data card compares pairs among duration, views, likes, comments, and engagement rate for eligible videos.

Rows and columns are metrics; the diagonal compares a metric with itself. Colors distinguish strong positive, strong negative, weak, and insufficient-evidence cells. The title/tooltip explains insufficient evidence.

Correlation is not causation. Topic, video age, format, and channel size can be confounders. Testing many pairs also raises the chance of a coincidental relationship; use the result as a hypothesis.

Shared states

Practice — Coğrafyanın Kodları

Verified July 17, 2026.

1. Open Coğrafyanın Kodları → Predictions and record each card's period badge.
2. Separate the raw last-7-day changes from percentage differences versus the prior 7; do not call momentum a certain future result.
3. Read volatility with consistency and add a caution if the mean is near zero.
4. Describe the loss card as a “public subscriber-count decline proxy.”
5. Distinguish a live first-48-hour projection from — or an observed value.
6. Record the dominant decay class together with videos analyzed.
7. Switch Pearson and Spearman for one metric pair; if coefficient or evidence changes, note a possible reason.

Interpretation limits

• Forecasts change as new data arrives; snapshot timing and video age matter.
• Momentum and consistency are product scores, not financial or statistical guarantees.
• The loss card does not measure individual churn or its cause.
• Correlation and classification do not replace an experiment.

What's next?