Home › Experiments › Light vs dark roast

Light vs dark roast on a V60, measured

One variable changed · simulated on the CoffeeScope V3 engine · updated 18 July 2026

Brew a light roast and a dark roast on the exact same recipe and they do not land in the same place in the cup. The simulator lets us isolate why: we changed only the roast level on one V60 recipe and measured how far the extraction and the predicted taste moved.

The question

Roasting develops the bean. A darker roast is more soluble and less dense, and it has been degassing CO₂ for longer, so — brewed identically — it tends to extract more and push toward bitterness. A lighter roast holds onto its acids and its delicate aromatics but resists extraction, so the same recipe can leave it tasting sour and underdeveloped. The practical question: on a fixed recipe, how much does roast alone move the cup, and which way?

The setup

Both lanes are the baked Hoffmann Ultimate V60 recipe (15 g, 1:16.7, 96 °C, 600 µm, center pour). The only difference is roast level on the model's 0–1 scale: 0.15 (light) vs 0.75 (dark).

Exact recipe (reproducible in the simulator)

{
  "brewer": "v60", "doseG": 15, "ratio": 16.7, "grindUm": 600,
  "waterTempC": 96, "roastAgeDays": 7, "grindSpread": 0.24,
  "bloomWaterX": 3, "bloomTimeS": 45, "pourRate": 6,
  "pourHeight": 16, "pourPattern": "center", "bedLevel": 0.5,
  "pourHeadroom": { "stopMm": 10, "resumeMm": 13 },
  "resolution": "medium", "numericBackend": "js",

  "roastLevel": 0.15   // Lane A — light
  "roastLevel": 0.75   // Lane B — dark
}

Every number on this page is read from these two runs, taken to brew completion on the deterministic V3 engine.

Measured results

Measure	Light · 0.15	Dark · 0.75	Takeaway
Extraction yield (EY)	19.1 %	20.4 %	The darker, more-soluble roast gives up more of itself on an identical recipe.
Strength (TDS)	1.28 %	1.37 %	More dissolved solids in the same water — a heavier, stronger cup from the dark roast.
Drawdown time	162 s	162 s	Identical: roast changes chemistry, not the bed's flow resistance here.
Cup volume	223.5 ml	223.5 ml	Same beverage volume — roast shifts what is dissolved, not how much water passes.
Caffeine	159 mg	159 mg	Unchanged — caffeine is set by dose and extraction, not by roast colour.

Predicted taste axes

The taste axes are directional calibrations, not sensory-panel scores. A difference is flagged tasteable when it clears the model's own threshold of 0.045 (0–1 scale).

Axis	Light	Dark	Δ	Tasteable?
Clarity	0.79	0.63	0.16	yes — the light roast reads cleaner
Astringency	0.02	0.12	0.11	yes
Bitterness	0.29	0.38	0.09	yes — dark climbs, as expected
Acidity	0.57	0.49	0.08	yes — brightness fades in the dark cup
Sweetness	0.75	0.71	0.04	no — below threshold
Body	0.59	0.62	0.03	no — below threshold
Aroma	0.96	0.96	0.00	no — unchanged in the model

Roast moves the axes you would predict — clarity down, bitterness and astringency up, acidity down — but only four of the seven clear the tasteable line, and the shifts are modest next to what grind does. On a fixed recipe, roast tilts the cup rather than transforming it.

How it builds over time

Cumulative extraction yield, sampled during the brew:

Time	Light — EY	Dark — EY
60 s	10.4 %	11.4 %
120 s	16.8 %	18.1 %
Final	19.1 %	20.4 %

The dark roast runs about one extraction point ahead throughout — the gap opens early and holds, rather than appearing only in the tail.

Why it happens

Roasting drives off mass and restructures the bean. In the model, a darker roast carries a larger pool of slow, soluble compounds (the bitter, woody polyphenols) and a smaller fast (bright, acidic) pool, is less dense, and degasses more CO₂. On an identical recipe, that higher solubility raises extraction yield and TDS, and the taste readback follows the chemistry: bitterness and astringency rise off the larger slow pool, while acidity and clarity fall as the bright fast fraction shrinks. Because roast does not change grind or geometry, the bed's permeability — and therefore drawdown — is untouched. See the methodology for the fast/slow/refractory solubles model and the CO₂ degassing term.

Caveats

This is a simulator, not a measurement of your specific coffee. Read the numbers as a well-reasoned direction, not a lab result.

Roast level here is a single 0–1 knob that shifts solubility, density and CO₂ together — it is not a real roast curve, and it does not model the specific Maillard and caramelisation products that give a given roast its character.
The aroma axis does not separate the two roasts in this run; real light and dark roasts smell very different, and the model's aroma term (directional) does not capture that development.
The taste axes are calibrated to move the right way by a plausible amount, not validated against a trained sensory panel.
Bean origin, process and freshness beyond a simple roast-age input are not represented.

Run it yourself

Load this exact A/B in the live simulator and compare the extraction fields side by side:

Run light vs dark live →

Related experiments: Fine vs coarse grind 1:14 vs 1:17 ratio 85 °C vs 96 °C

Methodology, validation & limitations →