Document EDFA max-power characterisation; restore safe pump clamp

Swept pump-1 to find real output: 1 W (30.1 dBm) at 4000 mA, 1.22 W at
4500 mA still unsaturated. External Thorlabs meter read 1.05 W vs 1.22 W
internal PD -> telemetry is ~0.5-0.6 dB optimistic, path-dependent; treat
amp/output*_power_dbm as indicative, not calibrated.

- README: full pump-vs-output table, PD-vs-meter calibration note, >1 W
  output-connector warning, ramp/overshoot guidance
- config.h: PUMP_CURRENT_MAX_MA back to 4000 mA (1 W rating) after the
  supervised over-rating sweep; added get_thresholds probe to agiltron driver

README.md

@@ -46,23 +46,47 @@ The rack is a **MOPA** (master-oscillator power-amplifier) at 1550 nm:
 Run the booster in **ACC (constant-current)** mode — APC stepping quantises the
 output. Below transparency the booster erbium fibre *absorbs* the pre-amp
 output, so output stays at the PD floor (−60 dBm) until the pump is high
-enough. Measured 2026-06-17 with ~6 mW (7.8 dBm) seed into the EDFA:
+enough.
 
-| Pump-1 | Output-1 | note |
-|-------:|---------:|------|
+### Pump-1 vs output characterisation (2026-06-17)
+
+Two sweeps. The transparency sweep used ~6 mW (7.8 dBm) seed into the EDFA; the
+high-power sweep added a 3 dB pad (input 3.2 dBm, within the ≤5 dBm benchtop
+spec) and a direct output-to-power-meter connection.
+
+| Pump-1 | Output-1 (internal PD) | note |
+|-------:|------------------------|------|
 | 500–900 mA | −60 dBm | below transparency, signal absorbed |
 | 1200 mA | 21.0 dBm | booster turns on |
-| 1500 mA | 23.3 dBm | |
+| 1500 mA | 23.1 dBm | |
 | 2000 mA | 25.9 dBm | |
-| 2500 mA | 27.1 dBm | documented operating point, ~515 mW |
+| 2500 mA | 27.1 dBm / 514 mW | documented 0.5 W operating point |
+| 3000 mA | 28.3 dBm / 679 mW | |
+| 3500 mA | 29.3 dBm / 857 mW | |
+| 4000 mA | 30.1 dBm / 1026 mW | **1 W rated ceiling** |
+| 4500 mA | 30.85 dBm / 1216 mW | over-rating; curve still ~linear (not saturated) |
 
-Pump current overshoots ~2× briefly on each step up and settles in 2–4 s, so
-**ramp gradually** rather than jumping straight to the setpoint.
+The unit (factory APC setpoint 30 dBm) comfortably exceeds 1 W: 1.22 W at
+4500 mA with the output still climbing ~0.74 dB/500 mA and module temp only
+~30 °C — so its true saturated maximum is higher, but driving a 1 W-rated pump
+diode past rating trades lifetime for diminishing return. Pump current
+overshoots ~2× briefly on each step up and settles in 2–4 s, so **ramp
+gradually** rather than jumping to the setpoint.
 
-> **Safety:** this is a Class 4, ~1 W, invisible 1550 nm source. The firmware
-> clamps pump-1 to `PUMP_CURRENT_MAX_MA` (3000 mA) and seed current to
-> `SEED_CURRENT_MAX_MA`, but those are not interlocks — treat the output as
-> hazardous unless it terminates in a power meter / beam dump / closed system.
+> **Internal PD vs external power meter:** at 4500 mA the internal PD reported
+> 1.22 W while a Thorlabs meter measured **1.05 W** at the output (~0.6 dB
+> high / residual loss). Treat the `amp/output*_power_dbm` telemetry as
+> indicative, ~0.5–0.6 dB optimistic, not a calibrated power reading. (The
+> earlier lab notes saw the opposite sign through a different connection, so
+> the offset is path-dependent.)
+
+> **Safety:** Class 4, >1 W, invisible 1550 nm. The firmware clamps pump-1 to
+> `PUMP_CURRENT_MAX_MA` (4000 mA ≈ the 1 W rating) and seed current to
+> `SEED_CURRENT_MAX_MA` — guardrails against accidental over-drive, **not
+> interlocks**. Above 1 W the Agiltron manual requires matched high-power
+> connectors or a splice on the output, else the output fibre/amplifier can be
+> damaged. Always drop the pump to 0 before unmating any fibre. Treat the
+> output as hazardous unless it terminates in a power meter / beam dump.
 
 ## Hardware