Filling the Blank Page Between Two Entries

Opening Scene

Between two confirmed position fixes — a coastal landmark spotted at noon, a star sighting taken at midnight — the ship’s actual path across the chart isn’t a mystery, and it isn’t left blank either. The navigator draws a reasoned line between the two known points, using known speed and heading to interpolate where the ship most likely was in between. It isn’t a guess pulled from nowhere; it’s a principled estimate grounded in what’s actually known on either side of the gap.

Time-series modelling relies on the same technique whenever real data has a hole in it: gap-filling and interpolation.

In Plain English

Gap-filling is the general practice of supplying values for missing points in a time-series, so downstream calculations — averages, models, charts — don’t break on missing data or silently treat a gap as zero. Interpolation is the specific technique of estimating those missing values based on the data points before and after the gap — linearly, or using a more sophisticated curve — rather than filling with an arbitrary default. Both exist because a gap left untreated is rarely neutral: it either breaks a calculation outright or gets silently misinterpreted as a real, meaningful zero.

The Old Way

Handling missing time-series data has always demanded a deliberate choice, and the naive defaults have caused real damage when applied without thought:

• Silently filling a gap with zero was a common shortcut that treated “we have no reading” as indistinguishable from “the reading was actually zero,” a conflation that’s obviously wrong for something like a temperature sensor and dangerously misleading for something like a revenue metric.
• Simply dropping records with gaps entirely worked for some analyses but broke others that specifically needed a continuous, evenly-spaced series — many windowing and forecasting techniques assume no gaps at all, and quietly skip past a hole rather than flagging it.
• Forward-filling — repeating the last known value until a new one arrives — is a reasonable default for some data (a device’s last-known status) but actively wrong for others (a rapidly changing sensor reading), and applying the wrong strategy has produced misleadingly smooth-looking data that hides real volatility.

Choosing the right gap-filling strategy has always depended on what kind of data it is and what the gap actually, plausibly means.

What’s Changing (and Why AI Is the Reason)

1. AI-driven forecasting and anomaly detection models, covered in Articles 14 and 15, are often far more sensitive to gap-filling strategy than traditional dashboards were, since a poorly chosen interpolation method can introduce artificial patterns a model then mistakenly learns as real signal. Getting gap-filling right has shifted from a display nicety to a modelling-accuracy concern.
2. AI-assisted interpolation techniques can now go well beyond simple linear estimates, using learned patterns from similar historical periods or correlated series to fill a gap far more plausibly — covered further in Article 17’s discussion of cross-series correlation — than a straight line between two known points ever could.
3. AI systems increasingly need to know, explicitly, which values in a time-series are real observations and which are filled estimates, since treating an interpolated value with the same confidence as a genuine measurement can lead a downstream model or agent to draw conclusions the actual data never supported. Marking filled values distinctly, rather than making them indistinguishable from real ones, has become a real data-modelling requirement.

The Metaphor, Fully Extended

Ship’s Chronometer & Logbook Element	Time-Series Modelling Concept
The blank stretch of chart between two confirmed position fixes	A gap in a time-series, between two known real data points
The navigator’s reasoned line drawn between the two fixes, using known speed and heading	Interpolation, estimating missing values from surrounding known data
Assuming the ship simply stayed still during an unlogged stretch	A naive forward-fill or zero-fill, potentially misrepresenting what actually happened
A dashed line on the chart marking the interpolated stretch as estimated, not confirmed	Explicitly flagging filled values as estimates, distinct from genuine observations
A more experienced navigator using known currents and wind patterns for a better estimate	AI-assisted interpolation, using learned patterns rather than a simple straight-line estimate

Quick Recap

• Gap-filling supplies values for missing time-series data points; interpolation is the specific technique of estimating them from surrounding known data.
• Naive strategies like zero-filling or blind forward-filling can misrepresent what actually happened during a gap.
• AI forecasting and anomaly detection models are especially sensitive to gap-filling strategy, since a poor choice can introduce artificial patterns a model mistakenly learns.
• Explicitly flagging filled values as distinct from real observations is increasingly necessary as AI systems consume time-series data directly.

Where This Fits in the Series

This article closes out the series’ core hands-on techniques alongside downsampling, windowing, and late-data handling. Article 9 turns to a bigger structural decision: choosing the right kind of database for time-series data in the first place.