Methodology

The Cellared Ageability Index (CAI)

The Cellared Ageability Index uses up to ten scored inputs. This page is the documented methodology: what we measure, where the data comes from, and what the model cannot know.

Version
CAI v1.0
Last reviewed
Published
Author
Carson Smith, WSET L3 and CMS Intro

Key takeaways

Why drinking windows are hard

A drinking window is not a single point. It is a range with a shape: too young, approaching peak, peak, mature, past peak. The shape depends on the information available for the bottle and on the limits of what a model can know.

Cellared keeps the dimensions. The model returns a four-point range (open, peak start, peak end, close) rather than a single answer. That makes the estimate useful without pretending a bottle has one exact turning point.

The scored inputs

1

Tannin and structure

Tannin in reds and structural richness in whites are major reasons some wines improve for years while others are best young.

2

Acidity

Acidity helps a wine retain freshness and shape as it ages.

3

Grape variety

Different grapes begin with different aging profiles. When grape information is unclear, CAI leaves that factor unscored.

4

Price tier

Price is one signal about concentration and cellar intent. It is not a verdict on quality.

5

Region

Region can distinguish wines with materially different aging patterns, even when they share a grape.

6

Alcohol balance

Alcohol can support or shorten a wine’s trajectory depending on balance and style.

7

Wine type

Red, white, sparkling, rosé, dessert, and fortified wines follow different aging patterns.

8

Body and extract

Concentration and weight affect how much material a wine has to evolve.

9

Storage condition

CAI can use the storage condition recorded for a bottle. It cannot independently observe the bottle’s full storage history.

10

Bottle format

Large formats generally evolve more slowly than standard bottles. Smaller formats can move faster.

Base score, context, and personal preference

CAI starts with up to ten scored inputs from the bottle information available. It does not substitute a plausible detail when an input is missing.

Contextual adjustments and special maturity curves remain separate from the base score.

Your recorded drink-timing feedback can shift personal windows by grape variety.

A drinking window is an estimate, not a guarantee of how a specific bottle will show.

Where the data comes from

Cellared maintains reference data used to map bottle information, grapes, regions, and eligible categories. The CAI calculation begins with the information recorded for the bottle. Supporting reference data is not presented as a direct scoring factor unless it is part of the model.

Once you log a bottle as consumed and rate the experience, your recorded drink-timing feedback can shift personal windows by grape variety.

What the model does not know

Your specific storage conditions

CAI can use the storage condition recorded for a bottle. It cannot independently observe the bottle’s full storage history.

Individual bottle variation

Even from the same case, two bottles can age differently because of cork lottery and microclimate inside the rack. The model gives you a window for the wine. Your specific bottle may peak earlier or hold longer.

Your palate preferences

Your recorded drink-timing feedback can shift personal windows by grape variety.

Wines with insufficient data

When input data is missing or cannot be mapped safely, the relevant factor stays unscored. Below 3 scored factors, Cellared does not treat a new calculated window as authoritative.

How we calibrate

Methodology review

The methodology is reviewed when model behavior changes so the documented inputs and limits remain aligned with the engine.

Calibration sessions

Model changes are checked against wine examples and edge cases before release. The base score remains distinct from special maturity curves and contextual adjustments.

Validation status

CAI has not been academically peer-reviewed or independently validated. Its outputs are estimates, not measured facts about a specific bottle.

Sources and further reading

Frequently asked

Is my wine ready to drink right now?

Cellared returns a four-point window for every bottle: open, peak start, peak end, close. If today's date is past the open year and before the peak start, the wine is approachable but still developing. Peak years are the recommended drinking window. After the close year, the bottle is at risk of decline.

Does Cellared tell me when a bottle is ready to drink?

If enabled, Cellared can notify you as bottles approach, enter, and move beyond their peak window.

Can I age my wine past the projected window?

Sometimes. Storage and bottle variation can make a particular bottle reach maturity earlier or later than an estimate. CAI can use a recorded storage condition, but it cannot independently observe the bottle’s complete history.

Should I open this wine tonight?

Use Quick Pick in the app, which filters your cellar by tonight's window and your meal context. From the methodology: a wine inside its peak range and paired with a compatible food is the right open-tonight candidate. A wine still in its pre-peak phase needs more time unless the occasion calls for fruit-forward expression.

What does a wine taste like at peak?

Peak is when primary fruit, secondary development (oak, lees, malo), and tertiary notes (forest floor, dried fruit, tobacco, leather) are balanced. Before peak: primary fruit dominates and structure feels grippy. After peak: fruit fades and tertiary notes take over until they hollow.

Why does Cellared sometimes show a wide window?

When input data is missing or cannot be mapped safely, the relevant factor remains unscored. Below 3 scored factors, Cellared does not treat a new calculated window as authoritative.

How is CAI different from a generic varietal aging chart?

Generic charts start with the grape. CAI uses up to ten bottle-specific inputs, then keeps missing information visible instead of filling it with a generic default.

How often is CAI updated?

The methodology is reviewed when model behavior changes so the page continues to describe the inputs and limits the engine actually uses.

How this is versioned

The model is versioned and dated so it stays stable to reference: the current release is CAI v1.0, last reviewed 2026-07-30. The version changes when the model itself changes. Documentation corrections update the reviewed date without changing the model version.

See it on a bottle

Drinking windows on every bottle in your cellar. Try it on a wine in the guide.

Browse the Wine Guide