Regression and threshold

What it is

A shelf-life study's detail screen combines study tracking (completed, pending, and overdue points) with three analyses: the sensory decay curve (the panel's real mean per time point, attribute by attribute), the sensory evolution curves (a fit computed from the data), and the per-attribute regression, which uses a linear model to estimate when each attribute will cross its acceptability threshold.

How to get there

• URL: https://app.areacacao.com/sensory-analysis/studies/shelf-life/[id]
• How to reach it: in Shelf-life studies, click the study card.

Detail sections

Header

Study title, status badge, and, on a single line, the dates and references: Start:, Estimated end:, and, only if you associated them, Recipe: and Batch: (these last two show as monospaced code when present).

Statistics

Five indicators summarize the study's progress:

• Points: completed out of the total.
• Pending: points not yet evaluated.
• Overdue: points whose scheduled date has passed without evaluation.
• Days elapsed: since the start.
• Evaluations: total accumulated.

Storage conditions

A reminder of Temperature, Humidity, Light, and Packaging, along with the cadence Every N days · Duration N days.

Target attributes

The list of monitored attributes, each with its deterioration direction (Worsens going down or Worsens going up) and its threshold, shown as minimum N or maximum N.

Time points

A table with all the study's points: number, scheduled date, status, panelists, and evaluations for each one. It's a tracking view.

Below it, the Study checkpoints section is where you record: each checkpoint has its own Record tasting button, and above them you'll see how many checkpoints have data and how many you still need before the curve is reliable. See Record a checkpoint tasting.

Sensory decay curve

For each attribute, it plots the panel's real mean at each time point across the study, with its acceptance threshold line. It's the direct visual reading of how each attribute deteriorates (or holds up), based on the recorded evaluations.

While the series loads, you'll see "Loading time series…"; if the series can't be retrieved, "Couldn't load the study's time series." appears.

Sensory evolution curves

Below the decay curve, AreaCacao draws a second chart, the Sensory evolution curves: a fit computed from the panel's data that smooths each attribute's trend over time. Unlike the decay curve (which plots the observed mean point by point), this chart needs more data to be drawn.

Recommended sensory shelf life

At the top of the regression section, AreaCacao boils the whole study down to a single figure: the recommended sensory shelf life, with its calendar date.

Here's how it's calculated:

1. Of all the monitored attributes, AreaCacao takes the first one to cross its threshold — the one that sets the limit. That attribute is flagged with the Limiting attribute badge on its card. If the conservative criterion would point to a different limiting attribute, the card says so.
2. The recommendation uses the conservative end of the 95% confidence interval, not the point estimate. So if the central estimate says 25 days and the lower band says 20, the recommendation is 20 days. The point estimate is shown alongside for reference.
3. Rounding always goes down. A safety margin that gives days away is not a safety margin.

If no attribute crosses its threshold within the measured range, the card reports shelf life undetermined and suggests extending the study: you cannot recommend a shelf life the data doesn't support.

Shelf-life regression

This section appears with the title Shelf-life regression (OLS + 95% CI). For each monitored attribute, AreaCacao fits an ordinary least squares (OLS) linear model over the panel's means and computes:

• n: number of points used.
• Slope per day: how much the attribute changes each day.
• Intercept (day 0): estimated value at the start.
• r²: goodness of fit.
• 95% CI slope: 95% confidence interval.
• Direction: Worsens going down or Worsens going up.
• Min threshold or Max threshold: the one you set for that attribute, depending on its direction.
• Days to threshold and estimated cross date: when the curve reaches the threshold.
• 95% CI conservative and 95% CI optimistic: range of cross dates according to the confidence interval.

Per-attribute regression statuses

Each attribute falls into one of these statuses:

• Cross detected: the curve is moving toward the threshold (falling or rising, depending on the direction), and the app estimates the date it will cross it.
• No decay (or No rise, if it worsens going up): the attribute shows no trend toward the threshold.
• Already below threshold (or Already above threshold, if it worsens going up): the attribute is already past the threshold from the start.
• Insufficient data: there aren't enough evaluated points to fit the model.

While AreaCacao computes the model, the section shows the message "Computing OLS regression…" before the results appear. If the study has no target attributes configured, you'll instead see "No target attributes configured in this study."

Study status

The header shows the study's current status (Draft, Active, Paused, Completed, or Canceled) as a badge and, next to it, the buttons that apply to that status:

• Activate: on a Draft study (every study starts as a draft).
• Pause: on an Active study; Resume sets it back to Active.
• Complete: on an Active or Paused study.
• Cancel: on a Draft, Active, or Paused study. It asks for confirmation: the study is closed, can't be reopened from here, and the recorded measurements are kept.
• Delete: in any status. It asks for confirmation (Delete this study?): the study disappears from your list along with its points and measurements; this can be reversed from the admin panel.

The example study shows none of these buttons.

Next steps