GroupSoup Wine Scale

GSWS-100 β€” peer-weighted quality score

Every drink in the GroupSoup catalog has a score from 0 to 100. The 50–100 band uses the same format as Wine Spectator, Decanter, Wine Enthusiast, and Robert Parker β€” so the number means what you expect it to mean. The 0–40 band is a confidence ramp that lets you tell β€œnot enough data yet” apart from β€œmediocre wine.”

What's different from the critic format is how we calculate it: instead of one palate, the GSWS aggregates the tastings of every GroupSoup member who has rated the bottle, weighted by three transparent factors below.

Confidence ramp (0–40)

A new bottle starts at 0β€” no one has tasted it. Each tasting adds 10 points until the score reaches 50, at which point the peer-weighted algorithm takes over. The ramp exists so a newly-added bottle with one enthusiastic 5-star rating doesn't pretend to be a 100; the score has to earn its quality reading.

  • 0 β€” never tasted by a member.
  • 10 / 20 / 30 / 40β€” 1, 2, 3, or 4 tastings recorded. The score is purely β€œhow settled is this” and not yet a quality reading.
  • 50–100 β€” 5 or more tastings recorded. The peer-weighted algorithm below is in effect.

What the bands mean (50–100)

95–100
Classic
A great wine. Defines its category, vintage, or region.
90–94
Outstanding
Superior character and style. A wine of distinction.
85–89
Very good
Solid, well-made wine with notable character.
80–84
Good
A sound, drinkable wine.
70–79
Average
Drinkable but unremarkable.
50–69
Below average
Not recommended.

How the score is computed

For every tasting recorded against a drink, the rater's 1–5 overall score (s) is converted to the 100-point scale and then averaged using a weight made of three components:

Experience

1 + log₁₀(1 + n)

Tasters who have rated more distinct drinks contribute slightly more. The log curve keeps a single prolific rater from dominating.

Recency

exp(βˆ’days / 730)

Tastings have a ~2-year half-life. Old ratings still count, but recent ones move the score more β€” important for wines that evolve.

Consensus

1 if |s βˆ’ median| ≀ 1 else Β½

A tasting more than one star off the bottle's median is dampened to half-weight. Reduces outlier and troll impact without removing dissenting voices.

if tastings < 5:
GSWS = tastings Γ— 10
else:
weight = experience Γ— recency Γ— consensus
weighted_avg = Ξ£(weight Γ— s) / Ξ£(weight)
GSWS = clamp(round(20 Γ— weighted_avg), 50, 100)

Confidence labels

Every score also carries a confidence label so readers know how settled it is:

  • Low β€” fewer than 3 distinct tasters. The score is provisional.
  • Medium β€” 3 to 9 distinct tasters. Reasonable signal.
  • High β€” 10 or more distinct tasters. The score is well-supported.

Why a 100-point format?

Collectors and casual drinkers alike already know the 100-pt format. A β€œ91” means something instinctive in a way that 4.3 stars never will. We adopted it for the same reason restaurants list ABV in % and not in grams per decilitre β€” meet readers where they already are.

← Back to Collections