Prove the M101 dose-model switch; record the judged dose curves

The stream harness gains rules 18 to 21: the floor is derived from the
selected model's config key at the arm and a typed $35 is overwritten;
M101 switches the rendering exactly at the boundary in both directions
with no continuous FIRE at full duty across it; a refused switch (the
spindle on) leaves the stream unchanged, and the harness resyncs with
an empty line because the core skips G-code after an error until the
sender resyncs; M2 reverts a program-scoped switch and Q1 holds. The
analog sessions pin laser_floor_analog at the density floor so the
existing duty expectations stand, and the density ladder's unfloored
run moves from a chained $35 write to the laser_floor_density key.

The catalog's laser.power-floor becomes model-aware: it reads the
configured model and the floor keys from forgectrl, switches to the
configured model with M101 so the derivation runs without a fire, and
expects $35 to be that model's floor. The new laser.power-model-switch
switches to each model with the spindle off, checks the reported
message and $35 after each switch, and checks the M2 revert. The new
mswitch bench drill runs the switch on the machine in one armed run.

Docs follow: BRINGUP's Laser control section describes the switch, the
derived floors and the measured dose response of both models; the
MOTION settings table gains the five keys; LIGHTBURN gains a Power
models section and drops the stale 30 percent floor advice; SAFETY
names the switch's refusal rule; the CAMPAIGN-LOG records the judged
depth-witness runs of 2026-08-30 and the switch's host and bench proof.

scripts/bench/README.md

@@ -32,7 +32,7 @@ page's takeover does that; from a host, stop them first.
 | `gate_a_kernel_drills.py` | Kernel laser-safety drills (run on the board with forgectrl stopped so the pulse device is free): `K1` controlled-stop deceleration floor, `K2` resume waypoint honors the locked latch, `K3` a mid-ramp latch unlock never re-arms the FIRE drive. Software witnesses (`cnc/state`, `laser_enable`, `laser_on`, `laser_on_sampled`, interlock bit 3) plus the PSU-connector LASER_ON scope point; K3 refuses to run if HV reports good. |
 | `laser_stream_test.py` | Host-side laser pulse-stream emission harness: runs the native null-sink controller with `GFSINK_DUMP`, drives small laser jobs over TCP, and checks the dumped bytes against the kernel feeder contract (leading power byte, no back-to-back power bytes, FIRE only inside cutting moves, every stream ends FIRE-clear, no FIRE on a stepless gap, no FIRE leak across cycle churn, the rapids after an M5 executed at idle ship dark, and the next job in the same process fires at the level the previous one ended at). Runs in the grblHAL repo's CI. |
 | `laser_lifecycle_test.py` | Host-side operator-armed-window lifecycle harness (null-sink controller): arm once per job with M5/M3 persistence, the M2 close, sender-change re-consent, the disarm grace counting down in Hold, and arm refusal under a blocking cooling verdict. Runs in the grblHAL repo's CI. |
-| `live_fire_drills.py` | **LIVE LASER** drills, on the board (the bench page) or from a LAN host (`GF_HOST`): `live_fire_drills.py <drill> [S] [F]` - `witness` (emission witness, lid-IR peaks vs the ambient baseline, HV current, job-based disarm on M2), `hold` (disarm grace in Hold), `faultpos` (armed job refuses a stale origin after an underrun), `ircut` (lid-IR characterization cut at S/F), `pthresh` (laser power-threshold ladder: 13 constant-power rungs from 2 % to 30 % of full on scrap; the lowest rung that marks is the tube's striking threshold and reads directly as the `$35` value - requires `$35` = 0 for the run), `dladder` (density ladder at a chosen base period), `pcurve` (laser performance-curve ladder: one 100 mm line per level at 10 mm/s under M3, the laser off between rungs and a mid-ladder rung repeated at the end; reads `pic/hv_current` and the head thermopile `head/beam_detect_analog` (a scatter detector in the beam path upstream of the final mirror, so it sees the beam, not the material) from sysfs at ~25 Hz on the board, brackets each rung on the controller's Run/Idle states, and reports per rung the current with a clipped-at-1023 flag, the thermopile delta over its laser-off baseline and in-line drift, then the normalized curve, monotonicity, a line fit with its threshold intercept and the repeat-rung drift; JSON record with the raw trace in the bench data directory; rungs follow `laser_power_model`, a comma list overrides; a curve measurement wants `$35` = 0), `dpatch [F] [pitch] [length]` (depth witness for the dose curve of the configured `laser_power_model`: two rows of small serpentine-filled patches, row A CW at feeds giving relative doses 1.0 to 0.25 of the reference feed, row B at the reference feed at 100/80/60/45/30 % density or duty; the operator matches each row-B patch to the row-A patch of equal depth, which reads the model's light fraction off the material beside the thermopile's prediction; JSON record), `m5dark` (the rapids after an M5 ship dark: one 20 mm line at M3 S400, M5, dwell, rapid back, dwell, rapid forward; PASS when the 25 Hz current trace shows one discharge segment and reads dark after the M5 and `laser_on_sampled` never re-lights; the catalog's `laser.m5-rapid-dark` is its port), `flowload` (cooling under laser load, one armed run per invocation, the conf keys it writes put back at the end, the pump never commanded off: `t1` reproduces the flow-check trip with the check on at its defaults and two 30 x 4 mm CW fills at F1500 starting on the press with no dark dwell, and reports the engine's rise/dT verdict beside the 25 Hz trace of both coolant sensors, the current, the digital witness and the heater output in 5 s bins across the window, with the shape at fire start; `t2 <secs> [pct]` runs with the check off and one fill of about `secs` lit seconds at CW or at `pct` density, and reports the lag to each sensor, the rise per raw-second of `hv_current` and what a full 50 s window would add against the 1.6 C margin; `fit` fits rise against dose over every t2 record; JSON records), `expstop` (armed kill on the expected-stop path; needs the panel token - `GF_TOKEN`, or the board's token file) and `ctrlstart` (the separate controller restart after it). Every drill waits for the operator's physical arm press; eye protection, fire watch, extinguisher, and exhaust are mandatory. |
+| `live_fire_drills.py` | **LIVE LASER** drills, on the board (the bench page) or from a LAN host (`GF_HOST`): `live_fire_drills.py <drill> [S] [F]` - `witness` (emission witness, lid-IR peaks vs the ambient baseline, HV current, job-based disarm on M2), `hold` (disarm grace in Hold), `faultpos` (armed job refuses a stale origin after an underrun), `ircut` (lid-IR characterization cut at S/F), `pthresh` (laser power-threshold ladder: 13 constant-power rungs from 2 % to 30 % of full on scrap; the lowest rung that marks is the tube's striking threshold and reads directly as the `$35` value - requires `$35` = 0 for the run), `dladder` (density ladder at a chosen base period), `pcurve` (laser performance-curve ladder: one 100 mm line per level at 10 mm/s under M3, the laser off between rungs and a mid-ladder rung repeated at the end; reads `pic/hv_current` and the head thermopile `head/beam_detect_analog` (a scatter detector in the beam path upstream of the final mirror, so it sees the beam, not the material) from sysfs at ~25 Hz on the board, brackets each rung on the controller's Run/Idle states, and reports per rung the current with a clipped-at-1023 flag, the thermopile delta over its laser-off baseline and in-line drift, then the normalized curve, monotonicity, a line fit with its threshold intercept and the repeat-rung drift; JSON record with the raw trace in the bench data directory; rungs follow `laser_power_model`, a comma list overrides; a curve measurement wants `$35` = 0), `dpatch [F] [pitch] [length]` (depth witness for the dose curve of the configured `laser_power_model`: two rows of small serpentine-filled patches, row A CW at feeds giving relative doses 1.0 to 0.25 of the reference feed, row B at the reference feed at 100/80/60/45/30 % density or duty; the operator matches each row-B patch to the row-A patch of equal depth, which reads the model's light fraction off the material beside the thermopile's prediction; JSON record), `mswitch` (the M101 dose-model switch on the machine, one armed run: a density line at S500, M5, M101 P0, the same line back under analog, M2; asserts the arm/switch/revert reports with each model's floor, $$ showing the floor in force, no re-prompt across the switch, exactly two discharge segments with a pulsed character under density and a steady one under analog, and dark after), `m5dark` (the rapids after an M5 ship dark: one 20 mm line at M3 S400, M5, dwell, rapid back, dwell, rapid forward; PASS when the 25 Hz current trace shows one discharge segment and reads dark after the M5 and `laser_on_sampled` never re-lights; the catalog's `laser.m5-rapid-dark` is its port), `flowload` (cooling under laser load, one armed run per invocation, the conf keys it writes put back at the end, the pump never commanded off: `t1` reproduces the flow-check trip with the check on at its defaults and two 30 x 4 mm CW fills at F1500 starting on the press with no dark dwell, and reports the engine's rise/dT verdict beside the 25 Hz trace of both coolant sensors, the current, the digital witness and the heater output in 5 s bins across the window, with the shape at fire start; `t2 <secs> [pct]` runs with the check off and one fill of about `secs` lit seconds at CW or at `pct` density, and reports the lag to each sensor, the rise per raw-second of `hv_current` and what a full 50 s window would add against the 1.6 C margin; `fit` fits rise against dose over every t2 record; JSON records), `expstop` (armed kill on the expected-stop path; needs the panel token - `GF_TOKEN`, or the board's token file) and `ctrlstart` (the separate controller restart after it). Every drill waits for the operator's physical arm press; eye protection, fire watch, extinguisher, and exhaust are mandatory. |
 | `pacing_test.py` | Protocol-loop pacing check (runs on the board, dry motion): idle and parked-in-Hold states are coarse-paced, active motion is tight-paced, and a feed-hold/resume mid-move preserves position with no feeder starve. |
 | `gfbench.py` | Not a tool: the helper the board/host tools share - `HOST`/`LOCAL` from `GF_HOST`, `board(cmd)` (local `sh -c` or ssh), the factory coolant conversion `degc()`, `data_path()` (`FORGETEST_BENCH_DATA` or next to the tool), forgectrl's HTTP API with the panel token, `setting(key)` (from forgectrl, or from `/data/forgefirm.conf` on the board while forgectrl is stopped). |
 | `fan_test.py` | Fan/coolant bench (board or host; controller running): snapshots fan PWMs/tachs/temps, drives M8 → cut fans, M9 → cooldown → idle, verifying via tach readbacks. |