diff --git a/components/pid/TODO.md b/components/pid/TODO.md
index a4db2cf..a3e43e8 100644
--- a/components/pid/TODO.md
+++ b/components/pid/TODO.md
@@ -73,11 +73,18 @@ git history for `temp_controller.py`/`temp_controller_smith.py`).
   clamp was considered and rejected: sustaining a genuine 1.5 K/min ramp
   needs `y≈0.7-0.75` from `yi` alone at steady state, the same range the
   disturbance itself peaked at, so no single clamp value can suppress the
-  windup without also capping legitimate ramps. Plan is instead to gate
-  `pid_heat`'s activity by FSM state (frozen/offline while `HOLD`, engaged
-  by command or a new `HoldHeatEngage` threshold) — see
-  `docs/overshoot_hold_windup.md` for the full writeup, open design
-  question (threshold relationship with `HoldHeat`), and test plan.
+  windup without also capping legitimate ramps. An FSM-gating alternative
+  (freeze the loop's output in `HOLD` unless engaged) was also superseded.
+  Current plan: rename `pid_hold`/`pid_heat`/`pid_cool` to `pid_outer`/
+  `pid_inner`/`pid_inner_cool` (matching what actually runs when), and
+  split the inner loop's config into `Inner.Heat`/`Inner.Hold`/
+  `Inner.Cool` so the *same* PID instance gets a tight `yi_max` only while
+  `HOLD` is driving it and stays unclamped for real `HEAT` ramps — no
+  freeze/thaw, bumpless transfer preserved for free. See
+  `docs/overshoot_hold_windup.md` for the full writeup and test plan. This
+  is also a breaking config change (`Hold`/`Heat`/`Cool` → `Outer`/
+  `Inner.*`) — every deployed `config.json` needs migrating, not just the
+  repo templates.
 
 - [ ] **`kalman.py` is now dead code in production.** Neither
   `temp_controller.py` nor `temp_controller_smith.py` uses `Kalman`
diff --git a/docs/overshoot_hold_windup.md b/docs/overshoot_hold_windup.md
index 4c8d257..ed04bbb 100644
--- a/docs/overshoot_hold_windup.md
+++ b/docs/overshoot_hold_windup.md
@@ -13,12 +13,12 @@ are pulled from `logs/log_latest.json`.
 - `diff` never reached `Thresholds.HoldHeat` (1.0°C, `temp_controller_fsm.py`),
   so **the FSM never left `HOLD`** — the entire episode happened inside what
   the system still considered "holding".
-- `pid_hold` (outer loop, `Hold: {kp:0.6, ki:0, kd:0, kt:0}` — pure
+- The outer loop (`pid_hold`, `Hold: {kp:0.6, ki:0, kd:0, kt:0}` — pure
   proportional) tracked `0.6 * max(0, diff)` exactly, tick for tick, peaking
-  at `pid_hold_y ≈ 0.31` — nowhere near its `y_max=1.0` ceiling. Confirmed
-  directly against logged samples: `rate_soll` == `0.6*diff` to the last
-  digit throughout.
-- `pid_heat` (inner loop, `Heat: {kp:0.08, ki:0.02, kd:0, kt:1.5}`) then
+  at `y ≈ 0.31` — nowhere near its `y_max=1.0` ceiling. Confirmed directly
+  against logged samples: `rate_soll` == `0.6*diff` to the last digit
+  throughout.
+- The inner loop (`pid_heat`, `Heat: {kp:0.08, ki:0.02, kd:0, kt:1.5}`) then
   chased that `heatrate_soll` target, driving commanded power from a
   baseline ~500W burst-cycle up to **~2600W** (`y ≈ 0.74` of the 3500W max,
   see `tasks/heater.py:58-59`: `power = get_power_max() * y`).
@@ -34,8 +34,8 @@ are pulled from `logs/log_latest.json`.
   resulting ~0.4°C plateau above soll does not visibly decay within any
   reasonable observation window.
 
-**Root cause:** `pid_heat`'s integral term (`yi`) accumulated during the
-~130s the outer loop demanded a real (if modest) heat rate, and nothing
+**Root cause:** the inner loop's integral term (`yi`) accumulated during
+the ~130s the outer loop demanded a real (if modest) heat rate, and nothing
 bounded that accumulation — the existing anti-windup in `Pid.process()`
 (`components/pid/pid.py:45-48`) is back-calculation that only corrects `yi`
 when the combined output `y` is actually clamped at `y_min`/`y_max`. Since
@@ -44,12 +44,11 @@ time and the anti-windup mechanism never engaged. The integrator just had to
 unwind naturally against real (delayed) negative error, which took ~35+
 seconds after the outer loop had already zeroed its target.
 
-## Rejected approach: flat `yi_max` clamp on `pid_heat`
+## Rejected approach: a flat `yi_max` clamp on the inner loop
 
-The first fix considered was bounding `yi` directly
-(`self.yi = max(-yi_max, min(yi_max, self.yi))` in `Pid.process()`,
-independent of the existing `y`-clamp anti-windup). Numeric check against
-the real plant model kills this as a standalone fix:
+The first fix considered was bounding `yi` directly, with one constant for
+the inner loop regardless of state. Numeric check against the real plant
+model kills this as a standalone fix:
 
 | scenario | P needed | y needed |
 |---|---|---|
@@ -57,91 +56,169 @@ the real plant model kills this as a standalone fix:
 | 1.5 K/min ramp @ 50°C | 2547 W | 0.728 |
 | 1.5 K/min ramp @ 66°C | 2636 W | 0.753 |
 | the actual HOLD disturbance (peak) | 2600 W | 0.74 |
+| steady HOLD loss compensation @ 66°C | ~257 W | ~0.073 |
 
 At steady state (`err=0`), `kp*err` contributes nothing — the entire
 `y≈0.7-0.75` needed to sustain a genuine 1.5 K/min ramp has to come from
 `yi` alone, for as long as the ramp lasts. That's the *same* range the
-disturbance transient itself peaked at. So:
+disturbance transient itself peaked at, but far above what real steady-HOLD
+loss compensation ever needs (~0.07-0.15 at realistic brew temperatures). A
+single constant can't serve both: tight enough to matter for the disturbance
+(well under ~0.74) permanently starves a real ramp of the ~0.7 it needs;
+loose enough not to interfere with ramps (~0.75+) never engages during the
+disturbance at all. **Rejected as a single global constant.**
 
-- `yi_max` set low enough to meaningfully shorten the observed hangover
-  (well under ~0.74, e.g. ~0.3) would permanently cap a real 1.5 K/min ramp
-  far below its commanded rate — not a transient dip, a persistent,
-  uncorrectable shortfall for the entire ramp.
-- `yi_max` set high enough not to interfere with legitimate ramps (~0.75+)
-  sits at or above what the incident already peaked at, so it never
-  engages and does nothing for the hangover.
+An FSM-gating alternative (freeze the inner loop's output to 0 while in
+`HOLD` unless explicitly engaged by command or a new threshold) was also
+sketched, but left an open question about how a new engage threshold should
+relate to the existing `HoldHeat` FSM threshold, and needed extra
+engage/disengage hysteresis to avoid chatter. **Superseded** by the plan
+below, which reaches the same effect with less new machinery.
 
-A flat magnitude clamp cannot distinguish "leftover integral from an
-already-resolved disturbance" from "integral correctly holding up a real
-ongoing ramp" — in this plant they occupy the same output range. **Rejected.**
+## Refined plan: split the clamp by *which state is driving the same loop*,
+## not by freezing the loop
 
-## Refined plan: gate `pid_heat` by FSM state instead
+The insight above — real HOLD-time demand (~0.07-0.15) and real ramp demand
+(~0.7-0.75) occupy clearly different ranges — means a **per-state parameter
+set** on the *same* PID instance solves this more cleanly than gating the
+loop on/off: same accumulated `yi`/`d` state carried across `HOLD↔HEAT`
+transitions (bumpless transfer for free, no explicit reset, no engage/
+disengage hysteresis to tune), just a different `yi_max` ceiling depending
+on which state is currently active.
 
-Bound (or freeze) the *rate* at which `pid_heat` reacts based on FSM state,
-not a constant on `yi`, since real sustained-rate demand only ever happens
-in `HEAT`; in `HOLD` any large `yi` is by definition windup, since the
-target rate there is always ~0 outside of a real disturbance.
+### Rename for honesty
 
-1. **New flag `heat_loop_active`**, scoped to `HOLD` only, in
-   `TempControllerFsm`. `is_holding()` stays `state == HOLD` regardless —
-   Sud step "have we reached target" logic is unaffected.
-2. **While `heat_loop_active` is False in `HOLD`:** force `y=0`, skip
-   `pid_heat.process()` entirely (freeze it, don't reset — same
-   bumpless-resume convention already used for `HOLD→HEAT`, see
-   `temp_controller_fsm.py:99-101`). `pid_hold` keeps running every tick
-   regardless (it's pure-P, no windup risk, and its output is needed to
-   evaluate the engage condition below).
-3. **Engagement (`False→True`)**, checked each tick while in `HOLD`:
-   - Explicit command — a new `engage_heat_loop()`/`force_heat()` call,
-     wired to a manual UI action and/or auto-fired from `set_theta_soll()`
-     when the new target is a genuine change (not the same value re-sent
-     every tick).
-   - `diff >= Thresholds.HoldHeatEngage` (new config key).
-4. **Disengagement:** once `diff` settles back inside the existing
-   `HoldHeat`/`HoldCool` band, held for a short dwell time (a few seconds),
-   to avoid chatter right at the boundary.
-5. **Config/back-compat:** new threshold merges into `DEFAULT_THRESHOLDS`
-   the same way existing ones do (`temp_controller_base.py:30`).
+The current names don't match what actually runs when:
+- `pid_hold` already runs unconditionally *every* tick regardless of state
+  (`process_pid()`'s first line, `temp_controller_base.py:134`) — it's the
+  outer loop, not "the HOLD-state PID". Rename to **`pid_outer`**.
+- `pid_heat` already runs in *both* `HOLD` and `HEAT` today (only `IDLE`/
+  `COOL` skip it, `temp_controller_base.py:151-158`) — it's the inner loop
+  for the heating direction. Rename to **`pid_inner`**.
+- `pid_cool` only ever runs in `COOL` — no `HOLD`-time ambiguity, since a
+  disturbance that pushes temp *above* soll during `HOLD` is still handled
+  by `pid_inner` (the outer loop's `max(0.0, pid_hold_y)` floor sends
+  `heatrate_soll` to 0, and `pid_inner` reacts to the resulting negative
+  `heatrate_err` — the state only escalates to real `COOL` past
+  `Thresholds.HoldCool`). Rename to **`pid_inner_cool`** for symmetry, kept
+  as its own instance — today's explicit `reset()` calls when crossing
+  between heat-direction and cool-direction states
+  (`temp_controller_fsm.py:104,108,115,123`) stay exactly as they are; there
+  is no reason to share integrator state across a heater/chiller boundary.
 
-### Open design question (unresolved)
+### Config: `Hold`/`Heat`/`Cool` → `Outer` / `Inner.{Heat,Hold,Cool}`
 
-`HoldHeatEngage` needs to sit **below** the existing `HoldHeat`/`HeatHold`
-FSM threshold (currently `1.0°C`) to ever fire from inside `HOLD` — if
-`HoldHeat` stays at `1.0`, the FSM fully escalates to `HEAT` (where
-`pid_heat` always runs anyway) before any larger engage threshold is ever
-reached. Two ways to resolve, not yet decided:
-- Raise `HoldHeat`/`HeatHold` too, widening the HOLD↔HEAT band so `HOLD`
-  covers the full deadband and the new flag is the only gate within it.
-- Keep `HoldHeat` at `1.0`, set the engage threshold below it (e.g.
-  ~0.7-0.8°C) — loop only goes offline for small disturbances, engaging
-  before the FSM would escalate to full `HEAT` anyway.
+```json
+"TempCtrl": {
+  "pid_type": "Smith",
+  "beta": 0.9,
+  "Outer": { "kp": 0.6, "ki": 0.0, "kd": 0.0, "kt": 0.0 },
+  "Inner": {
+    "Heat": { "kp": 0.08, "ki": 0.02, "kd": 0.0, "kt": 1.5 },
+    "Hold": { "kp": 0.08, "ki": 0.02, "kd": 0.0, "kt": 1.5, "yi_max": 0.3 },
+    "Cool": { "kp": 0.08, "ki": 0.02, "kd": 0.0, "kt": 1.5 }
+  },
+  "Thresholds": { "...": "unchanged" }
+}
+```
+- `Inner.Heat` keeps today's `Heat` gains, no `yi_max` (or a very loose one)
+  — a real ramp must be able to reach `y≈0.75`.
+- `Inner.Hold` starts as a copy of the same gains, with `yi_max≈0.2-0.3`
+  added — comfortably above realistic steady-loss compensation (~0.07-0.15)
+  but well below what turned a 0.5°C dip into a 2600W burst.
+- `Inner.Cool` is `Cool`'s existing gains, moved under `Inner` purely for
+  structural consistency — introduced now, not because we've observed a
+  cooling-side incident. `pid_inner_cool` never runs during `HOLD`, so it
+  doesn't need its own `Hold` variant the way `Heat` does; one params block
+  is enough.
+- **This is a breaking config change** — no backward-compat shim for the
+  old flat `Hold`/`Heat`/`Cool` keys (per the "no compat hacks" convention).
+  Every deployed `config.json` needs migrating, not just the repo's
+  `config-real.json.tpl`/`config-sim.json.tpl` templates.
+
+### Code changes (planned, not yet implemented)
+
+1. **`components/pid/pid.py`** — add the symmetric `yi_max` clamp inside
+   `process()` (already planned): `self.yi = max(-yi_max, min(yi_max,
+   self.yi))` when `self.params.get('yi_max')` is set, applied right after
+   accumulating `yi` and before it's summed into `y`.
+2. **`components/pid/temp_controller_fsm.py`** — rename `self.pid_hold` →
+   `self.pid_outer`, `self.pid_heat` → `self.pid_inner`, `self.pid_cool` →
+   `self.pid_inner_cool` (constructor at lines 33-34/40, all `reset()` call
+   sites and comments at lines 11-12, 83, 87-89, 99, 104, 108, 111, 115,
+   118-119, 123).
+3. **`components/pid/temp_controller_base.py`**:
+   - `set_params()` (line 28-34): `self.pid_outer.set_params(params['Outer'])`;
+     store `self._inner_heat_params = params['Inner']['Heat']` and
+     `self._inner_hold_params = params['Inner']['Hold']` for the per-tick
+     lookup below; `self.pid_inner_cool.set_params(params['Inner']['Cool'])`.
+   - `process_pid()` (lines 133-160): rename `pid_hold_y` → `pid_outer_y`
+     and the `self.pid_hold.process(...)` call. In the combined `HOLD`/`HEAT`
+     branch (today's `else`, lines 156-158), select the active param set
+     before processing:
+     ```python
+     else:
+         inner_params = self._inner_heat_params if self.state == States.HEAT else self._inner_hold_params
+         self.pid_inner.set_params(inner_params)
+         self.pid_inner.process(heatrate_err, -self.heatrate_ist)
+         self.y = self.pid_inner.get_y()
+     ```
+     `set_params()` is a cheap dict-reference assignment (`pid.py:22-23`),
+     so calling it every tick has no meaningful cost. Because `kp`/`ki`/`kd`/
+     `kt` are identical between `Inner.Heat` and `Inner.Hold` in the starting
+     config, switching the active set at a `HOLD↔HEAT` transition changes
+     no term of `y` at that instant — only the `yi_max` ceiling going
+     forward. No bump at the transition, unlike the freeze/thaw approach.
+4. **Config files** — `config.json`, `config-real.json.tpl`,
+   `config-sim.json.tpl`: restructure into `Outer`/`Inner.{Heat,Hold,Cool}`
+   as above. Inline `"Cool": {...}` dicts in `scripts/demos/pid/
+   demo_temp_controller_smith.py:21`, `demo_temp_controller.py:21`, and
+   `scripts/demos/sud/demo_sud.py:32` need the same restructure.
+5. **`utils/replay_sim.py`** — `_apply_gain_overrides()` and the CLI flag
+   loop (lines 50, 214-222) iterate `('Hold','hold'), ('Heat','heat'),
+   ('Cool','cool')`; becomes `('Outer','outer'), ('Inner.Heat','inner-heat'),
+   ('Inner.Hold','inner-hold'), ('Inner.Cool','inner-cool')` (nested dict
+   access needed since these aren't flat top-level keys any more). The
+   `for section in ('Hold','Heat','Cool')` print loop at line 284 and the
+   `_infer_heatrate_soll_set()` docstring's "hold PID" reference (line 66)
+   need the same rename.
+6. **`components/pid/TODO.md`** — update the cross-link entry added for the
+   previous (superseded) plan to point at this rewritten section instead.
 
 ## Test plan (not yet implemented)
 
 Stdlib `unittest`, no pytest — `tests/components/pid/test_pid.py`,
 discoverable via `python -m unittest discover -t . -s tests/components/pid`.
+Simpler than the FSM-gating plan's test plan: no engage/disengage hysteresis
+or threshold-relationship behavior to cover, since the loop is never turned
+off — only its `yi_max` ceiling changes with state.
 
 **A. Unit-level, isolated `Pid`** — no plant involved. Feed a synthetic
 `err` sequence shaped like the incident (positive `heatrate_err ≈ 0.3` held
 for ~130 ticks, then decaying/negative tail, matching the real
-`rate_soll - rate_ist` pulled from the log) into `pid_heat` with the loop
-gated (frozen while "HOLD"/inactive) vs ungated (today's behavior). Assert:
-recovery time (ticks after error goes negative until `y` drops back under a
-small threshold) is measurably shorter when gated; `yi` never grows during
-the frozen window.
+`rate_soll - rate_ist` pulled from the log) into two `Pid` instances with
+identical gains, one with `yi_max` set (the `Inner.Hold` case) and one
+without (`Inner.Heat`). Assert: recovery time (ticks after error goes
+negative until `y` drops back under a small threshold) is measurably
+shorter when clamped; `yi` never exceeds the configured bound.
 
 **B. Closed-loop, self-contained synthetic scenario** — no dependency on
 the multi-MB log file. Real numbers from `PlantParams`/`config.json`:
 `Pot(dt)` with `M=27.96, C=3403.43, L=0.2, Td=17`, ambient from
 `config.json`; `TempController` via `PidFactory.create('Smith', dt)` with
-the real `Hold`/`Heat`/`Cool` gains. Run closed-loop (controller's own `y`
+the real `Outer`/`Inner.*` gains. Run closed-loop (controller's own `y`
 drives the plant, unlike `utils/replay_sim.py`'s open-loop observe-only
-mode): hold at 30°C until settled, knock `plant.temp` down ~0.5°C to
-emulate the cold-water event, keep ticking for several minutes. Run twice —
-gating on vs off — and assert peak overshoot above 30.0°C is measurably
-smaller with gating, while a separate run driving a genuine 1.5 K/min ramp
-confirms gating does *not* reduce the sustained ramp rate (guards against
-reintroducing the flat-clamp regression above).
+mode):
+- **Disturbance case**: hold at 30°C until settled, knock `plant.temp` down
+  ~0.5°C to emulate the cold-water event, keep ticking for several minutes.
+  Assert peak overshoot above 30.0°C is measurably smaller with `Inner.Hold`'s
+  `yi_max` set than without.
+- **Ramp case**: command a genuine 1.5 K/min ramp (state reaches `HEAT`) and
+  assert the sustained heat rate actually reaches ~1.5 K/min — guards against
+  reintroducing the flat-clamp regression from the rejected approach above.
+- **Transition case**: drive a `HOLD→HEAT→HOLD` sequence and assert `y` has
+  no discontinuity at either transition beyond what the changing `heatrate_err`
+  itself would explain — confirms the per-state param swap is truly bumpless.
 
 ## Status