The Verification Venue · a kitchen rule, rebuilt from the heat equation
The Roast Keeps Cooking After You Pull It
Every recipe says to rest the meat. Most explain it the same wrong way: that cooking "pushes the juices to the center," and resting lets them flow back out. That story is folklore. The real reason is carryover cooking — the hot outer shell keeps cooking the core by conduction after the oven is off. Here it is, live.
The shell of a roast comes out far hotter than the middle. Heat doesn't stop at the oven door; it keeps flowing down the temperature gradient, from the scorching outside into the cool core, for many minutes. So the core temperature keeps climbing after you pull it. Drag the moment you pull the roast and watch the core overshoot your target — the whole curve is integrated live from the heat equation, not played back.
Core when you pull
63.0 °C
your target / doneness
Carryover overshoot
+13.2 °C
core keeps rising, oven off
Peak core (after rest)
76.2 °C
medium-well → well done
Pull early. The overshoot finishes the job. Aim for medium-rare (≈54 °C) and pull near 48 — or watch a too-late pull turn it grey.
A bigger roast stores more shell heat and carries over more — a thin steak barely carries at all.
Hotter oven → hotter shell → bigger carryover. High-heat roasts overshoot well past the typical band.
The core peaks part-way through the rest, then everything cools together toward the room.
Turn on the folk model and the prediction is flat: a notional pool of "juice" sloshing back to where it was squeezed from carries no heat, so it moves the core temperature not at all. The carryover curve climbs; the juice-reservoir curve sits still. Only one of them is what a probe thermometer actually records — and it's the heat equation, every time.
The check — every number recomputed in front of you
The curve above is integrated live by a 1-D radial finite-difference solver — the exact same code as the offline verifier. For your current settings:
Across plausible roasts, the computed overshoot lands in the documented window (≈3–14 °C / 5–25 °F), with the very largest and highest-heat roasts at or just past the top — the reason chefs say pull it early:
| roast | oven | pull | peak | overshoot |
|---|
Meat thermal diffusivity is α = 1.22×10⁻⁷ m²/s, computed live from α = k/(ρ·c) with sourced (k, ρ, c) for cooked lean meat — squarely in the published 1.0–1.7×10⁻⁷ band. Run it yourself: node research/the-roast-keeps-cooking/verify-the-roast-keeps-cooking.mjs.
What's exactly modelled, what's idealised, and the one number to distrust
Exactly modelled (heat transport). The core curve is the numerical solution of the spherical heat equation T_t = α(T_rr + (2/r)T_r) by an explicit finite-difference scheme, run stably (Fourier number 0.15 ≤ ⅙, the binding limit at the center node). The carryover overshoot is a genuine prediction of that PDE: with the oven off, the shell is still much hotter than the core, so heat keeps flowing inward until the gradient flattens. This part is physics, not decoration.
The free choices (named, not hidden). Thermal diffusivity varies with cut, fat content, and water phase — it shifts during protein denaturation and near freezing — so we use a single representative value (k=0.45, ρ=1050, c=3500 → α≈1.22×10⁻⁷) and quote the range, never a false "true" number. The oven heating uses a convective boundary with an effective film coefficient chosen so the surface tracks the measured roast plateau (~90–110 °C), capped at boiling to stand in for evaporative cooling — real meat surfaces stay near 100 °C while wet. The rest phase cools by natural convection to room air (h≈9 W/m²K). Change any of these within reason and the overshoot stays in the 3–14 °C band; the qualitative story is robust, the third decimal place is not.
What's only schematic (moisture). This sim models heat exactly and moisture loss only schematically. The second real reason resting helps — water-holding capacity recovers as the protein gel cools, so a rested cut loses less juice when sliced — is chemistry we do not simulate here; we only name it. And resting is a genuine trade-off, not a free win: a long rest also lets a crisp crust go soft (AmazingRibs argues this can hurt bark). Treat the page as a proof of the carryover mechanism, not a verdict that you should always rest for 30 minutes.
The myth, precisely. "Cooking pushes juice to the center; resting lets it flow back" implies a measurable inward-then-outward migration of free water. Studies of cooked muscle don't find that pattern — the water is largely bound in the protein matrix, not pooled in a reservoir that sloshes around. The folk model, taken literally and run above, predicts the core does nothing during rest. The thermometer says otherwise. The mechanism that survives contact with a probe is carryover.