[plastikman]

Problem

Since v1.1.16 the chamber controller can't hold setpoint — it stalls ~2 °C low ("gets to 64, can't reach 65"). v1.1.15 held.

Root cause: the approach cap in pb_heater_pid.h limited SSR duty by chamber distance-to-target (40% within 2 °C, 70% within 5 °C). That clamped the steady-state hold, creating a stable attractor at target − 2 °C for any enclosure whose hold duty exceeds the cap, and integral_max = output_max − proportional stopped the integrator from compensating. It also throttled the element below its safe capability, adding SSR duty-modulation (wear) the prior bang-bang firmware never had — the concern Justin raised about frequent SSR trips.

The element (PTC) foldback already in pb_heater.c is the correct near-target limiter (a real thermal/safety limit). The chamber-error cap ceilinged the wrong variable.

Fix

• Rate-gated soft anti-overshoot (pb_heater_pid_approach_soft_cap): full authority unless within 3 °C of target and genuinely climbing (>0.06 °C/s), so a settled/oscillating hold is never clamped, while a fast approach on a low-loss enclosure still gets overshoot damping. integral_max frees up at hold.
• EMA-filtered gate rate (α 0.20): the NTC reports in ~0.1 °C steps, so a single tick spikes the one-sample rate to ~0.2 °C/s and would flicker the cap through the whole hold. Filtering rejects isolated ticks while a sustained climb still registers (bench: 27% → 8% of hold samples capped).
• SSR wear: widen the time-proportioning window 10 s → 30 s + a 2 s minimum on/off dwell (steady-state switching 12/min → ~4/min).
• Applied soft cap reported via state->last_approach_cap telemetry (DragonSniff//diag still see when damping is active).

Fixed safety cutoffs (105 °C PTC / 85 °C chamber / 70 °C target) untouched. All DragonBreath-side — no dc_pid change.

Bench validation (enclosed U1, 65 °C target, bed 100 °C)

	chamber hold	vs 65	SSR/min	chamber cap
v1.1.18 (bug)	62.9 °C	−2.1	12	pinned 0.40–0.70
v1.1.15 (good)	64.3 °C	−0.7	2.3	n/a
this fix	64.3 °C	−0.7	4.4	released (1.0)

The fix recovers v1.1.15's reach exactly and cuts SSR switching ~63% vs the bug. Raising the already-exposed fb_cut advanced setting 102 → 103 °C then reached and held a full 65.0 °C on the same rig, with the element still ~90 °C at steady state (huge margin to the 105 °C cutoff) — so fb_cut is the per-enclosure element-headroom knob for lossy setups.

Tests

Host suite (tests/pb_heater_pid_host_test.c) rewritten to cover the soft cap, EMA quantization-spike rejection, the 30 s window + dwell, a settled hold reaching full authority, and fast-approach damping. All four heater/policy host suites pass; firmware builds (esp32c3).

Follow-ups (not in this PR)

• Optional: expose SSR window / soft-cap params as advanced settings (foldback-chain mirror) — deferred since fb_cut already covers the highest-impact knob.
• The residual ~8% plateau flicker could go to ~0 by latching the anti-overshoot off after target is first reached this session — a small state-machine change worth discussing separately.

🤖 Generated with Claude Code

https://claude.ai/code/session_012B9YKQ5yUenkyDUqdbn6HP