Hojicha is low in caffeine because it is roasted: the line turns up on packets, on mail-order pages, in the words spoken in front of the shelf. The reason attached is usually temperature — the heat is high, so it flies off, it sublimes and escapes. Said out loud, it sounds plausible enough. The three sections so far were about published values; what is left is the roasting itself.
Does roasting reduce caffeine?
Within the range of the published trials, not by much. One trial recorded a fall of 6 to 13% at the dark roast. What produced it, though, cannot be read out of the report.
What the Kochi trial recorded
The report “Roasting Level and the Quality of Hojicha” (research period FY2013–2014), from the Tea Research Station of the Kochi Prefectural Agricultural Technology Center, varied the roast level on an automatic continuous hojicha roaster and indexed caffeine against aracha, unrefined tea, at 100. Tables 3 and 4, which carry caffeine, hold the 2014 data: first-flush tea came to 101 at the light roast and 94 at the dark, karibancha to 99 and 87. The summary: the light roast barely falls, the dark fell by 6 to 13%.
The roasting ran in two stages: the first at an exhaust temperature of 90°C with the drum at 22 revolutions per minute, the second at 100°C. The 2014 tables that carry caffeine (Tables 3 and 4) run to two roast levels, and what changes between them is the rotation speed at the second stage — 17 revolutions for the light roast, 10 for the dark (the 2013 tables in the same report, the ones that carry catechins, run to three levels, light, medium and dark, with first-flush tea at 17, 12 and 8). The slower it turns, the longer the leaf stays in the machine. Only exhaust temperature and rotation speed are recorded.
A record from higher temperatures, and the different places measured
There is a record from higher temperatures. A 1981 report put green tea aracha into a forced-hot-air electric constant-temperature oven at oven temperatures of 130°C and 160°C, 30 minutes each. Amino acids, vitamin C and free reducing sugars fell markedly, while caffeine, the text says, hardly differed.
The two measure different places. The 100°C is the air leaving the roaster; the 130°C and 160°C are temperatures inside an oven. They do not line up on one axis.
The band of real roasting temperatures is still open
Outside that, the space is open. The 2020 experiment that roasted bōcha in a drum roaster set the drum's inner surface to 180°C, 200°C and 220°C, on manufacturers' advice. No public data measuring caffeine in that band appeared in this search.
The sublimation line is worth following too. The temperature most often quoted for it is not in the official chemistry database (side note below).
Stems before roasting, and the same question split for coffee
Before roasting, the same station's measurement put kukicha at 77% of first-flush tea. Table 3, though, adds one line for kukicha to a run of tea flushes; it is not a leaf-against-stem comparison (chapter 7).
The same question is split for coffee. A 2024 study of filter coffee reported that with the brewing held level the darker roast comes out lower, and with the extraction yield held level, higher. Only the shape is shared: change what is held steady and the answer changes. (Raw material and extraction differ, so it cannot be carried across to hojicha.)
Where did the 6 to 13% go?
So where did the dark roast's 6 to 13% go? Broken down by heat, or crumbled fine and lost as dust — the report does not say. The mechanism is, for now, not understood. What stayed largely still through the roasting was caffeine. What is moving, then? Part 2 goes there, one condition at a time.