HYPER
HYPER
로딩 중...
로딩 중...
Hyper-Δ — state-based roasting control read through two deltas (ET & BT RoR). A roasting theory formalized by Hyper Log, cross-validated against 2020–2025 peer-reviewed literature, and implemented in the app exactly as written.
Grab the guide that fits your level — print it or read offline.
This theory defines roasting not as 'matching a target curve' but as an energy-state management problem with time lag. The goal is not the single best cup — it is minimizing the probability of structural failure: systematically removing the ways a roast goes wrong, like stalls, scorching and underdevelopment.
We don't promise the best cup. We systematically remove the ways to ruin one.
The strength of the warm air entering the room — how much energy you're pushing into the coffee right now. It responds fast.
How warm the coffee has actually become. Just as a room takes time to warm up after you raise the boiler, it follows slowly.
Change the supply (ET) and the coffee (BT) responds tens of seconds later. That lag is exactly the information to read. RoR is nothing to overthink — ET RoR asks 'is the supplied energy rising or falling?', BT RoR asks 'is the coffee heating faster or slower?'. Just read the arrows: ▲▼.
A crucial truth — BT rising does not mean the coffee is 'receiving more heat'. From around first crack, coffee generates its own heat (exothermic reactions). BT can keep climbing even after you cut the gas — and if you misread that as 'still absorbing heat' and hold the flame, it burns.
m·c_p·(dT_b/dt) = Q_in − Q_evap ± Q_rxn
Q_in = h·A·(T_env − T_b)Energy conservation + Newtonian convection (Ranz–Marshall) + Fourier conduction + Stefan–Boltzmann radiation, plus moisture loss (Q_evap) and reaction enthalpy (Q_rxn). All established laws — the shared skeleton of modern roasting models. The key: BT RoR is not a 'rate of endothermic change' but the net of inflow − evaporation ± reaction.
Never raise the flame — with exactly one exception: +2% on a stall. The rule engages only after the RoR peak. Trust a signal only when it persists for 8–10 seconds.
Energy and heating accelerating together. A departure from the smooth decline → pull it back.
Energy piling up while the coffee isn't taking it. Left alone, it spikes later → cut preemptively.
Supply and heating falling together — the ideal progression. The only hold · no further cuts (conservative).
Supply is down, yet the coffee heats faster. Residual R (rise left over after subtracting what the lagged supply predicts) >0 = exotherm (strong) / R≈0 = inertia (weak).
The only heat-increase exception.
Deadband — don't touch anything; wait.
The Hyper-Δ indicator on Hyper Log's live roasting screen runs this theory's state assessment in real time. A pipeline of median despiking → low-latency EMA → regression RoR → priority state assessment (stall first) runs every second — displayed as a single chip.
The only thing that differs between coffees (green beans) is the initial firepower — and its basis is HR (Heat Resistance): 'the time from charge to first crack under identical conditions'. Arrives late (high resistance)? Start hotter. Arrives early? Start lower.
Not a temperature 'value' but a fracture event you hear — so the reference holds even across different machines and probes
Measuring at a fixed BT would measure a different degree of cooking for every coffee; this removes that confound
Measuring up to just before the exothermic threshold activates is what is most meaningful in practice
By real-world sensory results, the classic 'monotonically declining ET RoR' method produced the best cups — data that must not be dismissed, and that physics explains (a smooth monotonic RoR → uniform development, preemptive offsetting of exothermic acceleration).
The classic method directly optimizes the outcome (the cup); Hyper-Δ diagnoses the process (energy state) to reduce failure probability and green-bean variance. They are not rivals — they play different roles.
The most sensible setup: keep smooth ET/BT RoR driving as your baseline and lay Hyper-Δ on top as a safety-and-diagnostics overlay — drive the way you always have, intervening only at the ET▼/BT▲ moments (coffee self-heating) the classic method could not see, to prevent accidents. Neither one raises the firepower — save for the single +2% stall exception.
Every quantitative claim was adversarially cross-checked against 2020–2025 peer-reviewed literature through five-way parallel searches across four domains (exotherm/reaction heat · first crack · heat transfer/airflow · HR/initial firepower). Confidence is labeled honestly on a three-level scale.
Settled by peer review
Supported, but condition- or single-study-dependent
No literature basis (experience heuristic)
The cut-% step, frequency, θ thresholds, gain multipliers — numbers the literature cannot provide. They are empirical calibrations that depend on machine, probe and batch, settled only by your roast logs. 'Logically closed, numerically uncalibrated' — that is the honest final state this theory can reach.
The Hyper-Δ indicator is free on both the web and the desktop app (Mac & Windows). No roaster? Five minutes in the simulation is all it takes to see it.
Theory v3.0 went through nine revisions — from first-principles physics through literature, 2025-paper, and logic passes to the HR + 2026 literature re-verification. Sensory validation is a single-user result; empirical generalization requires multi-batch data.