Extraction, TDS and the Brewing Control Chart

Brewing coffee is a solvent problem. Roasted coffee is roughly 28–30 % soluble by mass¹; the rest is cellulose and other structural material that will never dissolve in water. Every recipe is a strategy for pulling the right portion of that soluble fraction into the cup.

Two numbers describe the result of that strategy.

Extraction yield (EY)

Extraction yield is the percentage of the coffee dose that ended up dissolved in the beverage²:

EY (%) = (TDS × beverage mass) / dose × 100

If you brew 20 g of coffee into 300 g of beverage at 1.35 % TDS, you dissolved 4.05 g of coffee, which is a 20.25 % extraction.

The Specialty Coffee Association's long-standing target is 18–22 %², a range that descends from Lockhart's 1957 consumer-preference work at the Coffee Brewing Institute³. Below 18 % the cup tends to taste sour, salty and thin, because acids and light aromatics dissolve first and sweetness hasn't caught up yet; above 22 % the heavier, slower-dissolving compounds — bitter phenolics, chlorogenic acid lactones, drying tannin-like compounds — dominate¹. A 2020 UC Davis sensory study confirmed the pattern with trained panels: sourness fell and bitterness rose as extraction increased, largely independently of strength⁴. Modern light roasts on excellent grinders are often brewed happily at 22–24 %, a practitioner observation rather than a measured standard⁵, so treat the range as a compass rather than a fence.

Total dissolved solids (TDS)

TDS is the concentration of dissolved coffee in the liquid you drink. It is measured in percent by mass and is what your tongue reads as strength¹. The SCA Golden Cup standard puts filter coffee at 1.15–1.35 %²; Cup Engineering's calculator uses a slightly wider 1.15–1.45 % window to reflect current specialty practice, and that widening is our editorial choice, not a standard. Espresso runs around 8–12 % because it uses far less water per gram of coffee, a range reported from practitioner measurements rather than a formal standard⁶.

TDS is measured with a refractometer: dissolved solids change how light bends through the liquid, and coffee-specific refractometers convert that refractive index to a TDS reading⁵. A drop of filtered, cooled coffee is enough.

Strength versus extraction

These two axes are independent, and confusing them is the most common dialling-in mistake.

• Ratio (water per gram of coffee) mostly moves strength.
• Grind, temperature, time and agitation mostly move extraction.

The brewing control chart plots TDS on the vertical axis against EY on the horizontal, with diagonal lines for each brew ratio³. The ideal box sits at 18–22 % EY and 1.15–1.35 % TDS². Because a given ratio constrains you to a diagonal, changing ratio alone slides you along a line; to move perpendicular to it you have to change extraction. The 2020 UC Davis panels found the same independence in the cup: strength and extraction each shifted the sensory profile in different directions⁴, and a companion study found that once strength and extraction are held fixed, brew temperature itself makes little sensory difference⁷.

Now drive the map yourself — every lever, one at a time:

Mass balance and retained water

Not all the water you pour makes it into the cup. Spent grounds retain roughly 2 g of water per gram of coffee in a filter brew, a figure Gagné's measurements put at about 2–2.5 g/g depending on grind and brewer⁵; Cup Engineering's model uses 2.1 g/g for pour over, more for immersion, and about 1.1 g/g for espresso, where pressure squeezes the puck. So:

beverage ≈ water poured − dose × retention

This is why the calculator in this app reports both water and beverage mass, and why measuring beverage weight matters when you compute EY from a TDS reading.

What actually dissolves, and when

Extraction is not uniform over time. Small, highly soluble molecules leave the grounds first and large, slow-diffusing ones last⁸. In rough order:

1. Acids and salts — fast, small molecules. Dominant early; taste sour/salty in isolation⁹.
2. Sugars and Maillard products — the sweet, caramel and brown middle. Balance arrives here¹⁰.
3. Large phenolic compounds and lactones — slow; drying and bitter. Take over when everything else is spent⁹.

Under-extracted coffee has too much of 1 and not enough of 2. Over-extracted coffee has too much of 3. Good extraction is capturing 1 and 2 while stopping before 3 takes over. Everything else in coffee brewing — grind, water, temperature, agitation — is a lever for controlling that curve.

How the numbers are computed

Three of the formulas on this page are definitions and mass balance; they are exact. The fourth is Cup Engineering's prediction model, and its coefficients are working assumptions.

water     = dose × ratio                      g
beverage  = water − dose × retention          g      retention: g water held per g coffee
dissolved = EY / 100 × dose                   g
TDS       = dissolved / beverage × 100        %
EY        = TDS × beverage / dose             %      (rearranged; what a refractometer gives you)

Worked example: 20 g dose at 1:16.7 → 334 g water. With 2.1 g/g retention, 42 g stays in the grounds and 292 g reaches the cup. At 20 % EY, 4.0 g of coffee is dissolved, so TDS = 4.0 / 292 × 100 = 1.37 %. If you instead measure 1.37 % TDS in 292 g of beverage from a 20 g dose, EY = 1.37 × 292 / 20 = 20.0 %.

The calculator predicts EY before you brew as:

EY = base(method)
   + 0.9  × (baseGrind − grind)
   + 0.15 × (T − T_default)
   + 1.4  × log2(time / time_default)
   + 0.8  × log2(ratio / ratio_default)
   + schedule term (bloom, pulses, pre-infusion, stirring) + freshness term

The directions of those terms are established: extraction rises with surface area and contact time and follows diffusion kinetics, which multiscale modelling reproduces⁸. The coefficients (0.9 % per grind step, 0.15 % per °C, 1.4 % per doubling of time, 0.8 % per doubling of ratio) are Cup Engineering working assumptions chosen to reproduce the rough size of effects reported in that literature and in practitioner measurements⁵. They are not fitted to a dataset. Trust the direction and the order of magnitude; do not trust the second decimal. The schedule and freshness terms are small and are documented in Blooms, pulse pours and pre-infusion. The full model and its per-method parameters are in lib/coffee.ts and explained on the calculator page.